Gemini Guide

Gemini is Google's AI platform: a chat interface at gemini.google.com plus deep integration into Google Workspace (Docs, Sheets, Slides, Gmail, Calendar, Drive, Chat, Meet, Vids) plus Gemini Notebook (formerly NotebookLM) as its sister product. As of October 2026: the new flagship model is Gemini 4 Argon (since 13 October 2026 in the Gemini App for AI Pro and Ultra, including in DACH, and in the API). Below it sit Gemini 3.8 Flash (since 2 September 2026), Gemini 3.6 Flash, Gemini 3.1 Pro and Gemini 3 Flash; 3.7 and 3.5 Flash have been deprecated in the API since 8 October 2026. Gemini 3.5 Pro was never shipped. Nano Banana 2 is the current image model (Imagen has been discontinued, see chapter 11), Veo 3.1 (Lite + Standard) the video model, plus Gemini Omni 1.1 Flash for conversational video editing. New since 26 August 2026 is Gemini 3.5 Transcribe as the speech-to-text model behind dictation and transcription. On the Workspace side, September added Google Pics as a standalone image application and Gemini's cross-app orchestration, plus the Gemini desktop app for Windows.

1.1 The seven building blocks

Workspace depth: Unlike ChatGPT or Copilot, Gemini is seamlessly embedded in Google Workspace. A question in Gmail can be grounded in Drive files immediately, a Doc gets live suggestions without switching tabs.

1.2 Security, privacy, EU hosting

• Training opt-out: In Gemini for Workspace Business and Enterprise it is excluded contractually and by default (customer data is not used for training, no human review without permission, no use for ads). In the consumer Gemini app you have to opt out actively ("Keep Activity" toggle off = no training).
• EU hosting: EU multi-region and in-country DE/IN region (with allowlist) for Workspace Enterprise Plus / Education Plus with the Data Regions add-on.
• Compliance: SOC 1/2/3, ISO 9001, ISO 27001, 27017, 27018, 27701, ISO 42001 (AI Management Systems, new), BSI C5, FedRAMP High, HIPAA-ready, CJIS.
• Workspace Vault: Audit log for Gemini usage, retention policies, eDiscovery. Vault limitations: search by user/OU/date only, export as XML only (roadmap improvement 2026).
• EU AI Act: Google has signed the GPAI Code of Practice. The AI Act GPAI obligations have been fully applicable since 2 August 2026.
• DPIA Support Guide has been published ("Supporting Your DPIA for Workspace with Gemini"), a mandatory artefact for DACH.

1.3 Plans at a glance

Details follow in chapter 3. Quick mapping (as of October 2026, German list prices):

• Gemini Free: EUR 0. Web chat plus image generation, Deep Research (limited), Gemini Live, Canvas, Gems, 15 GB storage.
• Google AI Plus: EUR 4.99/month. 2x limits, video generation, Daily Brief, 200 Flow credits, Gemini Notebook, 400 GB, family sharing for up to 5 people.
• Google AI Pro: EUR 21.99/month (formerly AI Premium, renamed). 4x limits, Gemini 4 Argon (since 13 October 2026) and 3.1 Pro, Deep Research, Deep Search, 1,000 Flow credits, 5 TB, YouTube Premium Lite.
• Google AI Ultra (Standard): EUR 99.99/month. 5x limits over Pro, top Argon tier, Deep Think, Spark, 10,000 Flow credits, for devs and knowledge workers.
• Google AI Ultra (Max): EUR 219.99/month. 20x limits over Pro, 25,000 Flow credits, full YouTube Premium, 20 to 30 TB, Gemini in Chrome, Project Mariner.
• Workspace Business Standard and above: Gemini features included since 2025 (no longer an add-on). NotebookLM Plus a Workspace Core Service since February 2025.
• Gemini Enterprise (separate top SKU): Custom pricing. Agent governance, cross-platform integrations (M365, Salesforce, SAP, Jira), EU region.

1.4 Who uses Gemini

In the DACH mid-market at 22 percent adoption according to Bitkom 2025, often as a by-product of Workspace licences. Global Workspace reach: 9 million Workspace customers, with a growing share actively using Gemini. NotebookLM gained viral traction in 2025, with +56 percent month-over-month growth in H1 2025.

Typical use cases:

• Docs and Slides: Drafting, rephrasing, slide skeleton from a briefing.
• Gmail: Smart Reply, email summaries, inbox triage.
• NotebookLM: Onboarding knowledge bases, study material, Audio Overviews.
• Deep Research: Competitive briefings, market research.
• Nano Banana and Veo: Visuals and video snippets for marketing.

Productive in ten minutes. What to switch on first and where Gemini shows up in everyday Workspace use.

2.1 Where to access Gemini

2.2 Login

• Personal: Google account. Free, AI Plus, AI Pro or AI Ultra.
• Business: Workspace account, automatic recognition via the domain.
• Multiple accounts: Account switcher top right. Workspace data is strictly separated from the personal account.

2.3 What to do in the first ten minutes

1. Activate Personalization: Settings, Personalization. Add custom instructions.
2. Check Connected Apps: Which Google services may Gemini read? The default is conservative.
3. First side-panel task in Docs: Click the Gemini button in a Google Doc and feed "Help me write" with a concrete brief.
4. First notebook in NotebookLM: Upload two or three PDFs, generate an Audio Overview. The lowest-threshold wow moment.
5. Try your first Gem: Click a premade Gem such as "Brainstormer" or "Writing Editor" and give it a task.

Gemini's licensing landscape was reshuffled after Google I/O 2026 (19 May): "AI Premium" is now called "AI Pro", "AI Ultra" was split into a 100 USD and a 200 USD tier. Workspace received massive updates: Gemini is no longer an add-on but included in every Business/Enterprise subscription.

3.1 Plan comparison (as of October 2026)

Limits since I/O 2026: compute-based with a reset every 5 hours, no longer daily.

3.2 Which models Gemini has

Gemini 4 and Gemini 3 series:

• Gemini 4 Argon (announced on 30 September 2026, generally available since 13 October 2026, new): Google's new frontier model and the successor to 3.1 Pro, built for long, multi-step tasks in software development, knowledge work (such as legal and finance, financial analysis, presentations) and cyber defence. Context of 1M tokens, with an output limit of up to 1M tokens as well (previously 64K). Benchmarks according to Google include DeepSWE v1.1 77.9 percent, LVBench 91.7 percent, first place in the Vals Index and a 0.7 percent success rate for prompt injection attacks (Gray Swan). Where to use it: in the Gemini App and in AI Mode in Google Search for AI Pro and AI Ultra (worldwide, including DACH; not in Free or AI Plus), for businesses via Gemini Enterprise, for development in the Gemini API (gemini-4-argon, stable) and on Vertex AI; in Antigravity and AI Studio initially for Ultra. API pricing: introductory price USD 2 input / USD 10 output per 1M tokens until 30 November 2026, then USD 4 / 20 from 1 December 2026; no free tier. EU: In Gemini Enterprise and on Vertex AI, Argon runs in the EU multi-region with data storage and ML processing in the EU; individual country regions (such as Frankfurt) do not offer this residency commitment for Argon. For training sessions: ready to demo now, provided participants have AI Pro or Workspace access with Argon.
• Gemini 3.8 Flash (since 2 September 2026, new): Google's currently strongest Flash model, built for long-running software engineering tasks, autonomous agents and multi-step reasoning in specialist domains. On hard tasks it noticeably thinks longer, takes additional reasoning steps and calls tools iteratively. Benchmarks versus 3.7 Flash: DeepSWE v1.1 73.7 instead of 65.3 percent, Terminal-Bench 2.1 89.4 instead of 85.8 percent, HLE-Verified 54.9 percent. Price unchanged from its predecessor: 0.75 USD input / 3.75 USD output per 1 million tokens as an introductory price until 31 December 2026, then 1.50 / 7.50 USD from 1 January 2027. Important for DACH: Unlike 3.7 Flash, 3.8 Flash does not run through Gemini Spark but is available to AI Pro and AI Ultra subscribers directly in the Gemini App, in AI Mode in Google Search and in Gemini in Google Sheets. For development the model is available in AI Studio, Android Studio, Antigravity and Stitch, for enterprises via Gemini Enterprise. EU data residency (correction, as of 29 September 2026): According to the current location documentation, 3.8 Flash also runs in the EU multi-region, with data storage and ML processing in the EU. In Gemini Enterprise it has been enabled by default since 23 September 2026. Outside the global region Google charges a surcharge of around 10 percent (see chapter 19.2).
• Gemini 3.8 Flash Cyber (since 2 September 2026, new): Security variant of 3.8 Flash, successor to 3.5 Flash Cyber. Finds vulnerabilities across 20 programming languages with a hit rate above 70 percent and patches them automatically (47.2 percent on CWE-Bench); the Chrome security team reports 2.6 times more correct patches than competing commercial models. Access only via the new Fairwind Program for government agencies, operators of critical infrastructure and maintainers of open-source software. Not generally available, so for mid-market training it is context only.
• Gemini 3.8 Live and 3.8 Live Extended Thinking (GA since 15 September 2026, new): New real-time voice models behind Gemini Live, Search Live and the Live features in Workspace. They understand visual context, recognise 97 languages and switch between them automatically during a conversation; audio output carries a SynthID watermark. The Extended Thinking variant keeps reasoning in the background while the conversation continues. In Workspace for AI Pro and Ultra, in Gmail and Keep for all AI subscriptions, plus in the API and AI Studio. "Gemini 3.8 Live with Avatar" (since 24 September 2026) is only available in Gemini Enterprise. In addition, the speech output models Gemini 3.8 Flash TTS and 3.8 Flash-Lite TTS have been GA since 22 September 2026.
• Gemini 3.7 Flash (GA since 13 August 2026): In Google's own words the "most intelligent workhorse model yet for coding and agents", with clear leaps in software engineering, web development and agentic workflows. Context window 1,048,576 tokens input / 65,536 output. Introductory price until 31 December 2026: 0.75 USD input / 3.75 USD output per 1 million tokens, then 1.50 / 7.50 USD from 1 January 2027. Important for DACH: The launch went through the Gemini API, AI Studio, Android Studio, Antigravity and the Gemini Enterprise Agent Platform. 3.7 Flash reaches the Gemini App exclusively via Gemini Spark for AI Pro and Ultra subscriptions in more than 160 countries. Spark is excluded in the EEA, Switzerland, the United Kingdom and Nigeria. Since 2 September 2026 this is no longer a disadvantage: the successor 3.8 Flash arrives in the app without the Spark detour, including in Germany, Austria and Switzerland. Deprecated since 8 October 2026: 3.7 Flash is deprecated in the Gemini API and requests are automatically redirected to 3.8 Flash; the model can no longer be selected in AI Studio.
• Gemini 3.6 Flash (launched 21 July 2026, GA): the Flash workhorse in the Gemini App, an update of 3.5 Flash. Around 17 % fewer output tokens per task, clearly better at coding (DeepSWE from 37 to 49) and computer use (OSWorld from 78.4 to 83). Knowledge cutoff moved forward to March 2026. Price identical to 3.7 Flash: 0.75 / 3.75 USD as an introductory price until 31 December 2026, then 1.50 / 7.50 USD. Available in the Gemini App (model dropdown), Google Antigravity, AI Studio and Android Studio.
• Gemini 3.5 Flash (launched 19 May 2026 at I/O): Previous Flash workhorse, replaced by 3.6 Flash since 21 July 2026. Native computer-use tool since 24 June 2026 (see chapter 17). 1.50 USD input / 9.00 USD output. Deprecated in the Gemini API since 8 October 2026; requests are automatically routed to 3.6 Flash.
• Gemini 3.1 Pro (Preview): Flagship reasoning/multimodal, long context >200K tokens.
• Gemini 3 Flash (Preview): Frontier at reduced cost.
• Gemini 3.5 Flash-Lite (launched 21 July 2026, new): lightweight model for high throughput and low latency (agentic search, document processing). Beats 3 Flash on coding benchmarks (SWE-Bench Pro 54.2 instead of 49.6). 0.30 USD input / 2.50 USD output per 1 million tokens. Available in the Gemini App and in Google Search. Replaces 3.1 Flash-Lite as the cheapest 3-series variant.
• Gemini 3.5 Flash Cyber (limited-access pilot, since 21 July 2026): Security-specialised model for finding and patching code vulnerabilities at scale. Only for governments and selected partners, not generally available.
• Gemini 3.1 Flash Live (Preview): High-quality, low-latency Live API model for voice agents (real-time audio in/out).
• Gemini 3.1 Flash TTS (Preview): Dedicated speech generation model with low latency, separate from Flash Live.
• Gemini 3.5 Live Translate (Preview): Specialised Live API model for fluent real-time speech translation (see chapter 17).
• Gemini 3.5 Pro: announced for June 2026 (postponed from I/O), never shipped as of 15 October 2026 and not listed in Google's model overview. With Gemini 4 Argon, Google has moved the Pro segment straight to the next generation; Google has not confirmed that this means 3.5 Pro has been skipped. In the Flash segment, 3.8 Flash remains in the lead.
• Gemini Omni 1.1 Flash (since 27 August 2026, new): natively multimodal model for video workflows with conversational editing, now production-ready. New compared with the public preview from early July 2026 (then a maximum clip length of 10 seconds): scene extension in 10-second steps up to 40 seconds total length (the model takes up to 10 seconds of lead-in as context), keyframe control (define first and last frame, for clean transitions and loops), a 360p draft mode (around 60 % faster and a third of the cost of 720p) and upscaling to 1080p or 4K. Video references of up to 3 seconds keep characters and look consistent. Available in AI Studio and the Gemini Enterprise Agent Platform, since 27 August 2026 additionally in Google Flow and the Gemini App for AI Plus, Pro and Ultra. See chapter 11.3.

Gemini 2.5 series (predecessor, still stable): 2.5 Pro, 2.5 Flash, 2.5 Flash-Lite, 2.5 Flash Live. Since 18 September 2026 only available in the Gemini API for projects that have recently used these models; according to Google, new projects should use 3.5 Flash-Lite or 3.8 Flash.

Special models:

• Nano Banana / Nano Banana 2 / Nano Banana 2 Lite / Nano Banana Pro (the current image family, officially "Gemini Image"): Nano Banana 2 on 3.1 Flash Image since 26 February 2026 and the default in the Gemini App; Nano Banana 2 Lite since early July 2026 as the fastest and cheapest variant (around 4 seconds per image); Nano Banana Pro on Gemini 3 Pro Image as the higher-resolution professional tier. All stable in the Generative Media API. New since 6 October 2026: Nano Banana 2.1 (gemini-nano-banana-2.1) is generally available in the Gemini API at around USD 0.03 / 0.05 / 0.11 per image at 1K / 2K / 4K, with no free tier, and also in the Vertex EU multi-region. This deprecates the API ID of Nano Banana 2 (gemini-3.1-flash-image); no shutdown date has been set yet. Google has not separately announced whether 2.1 is already the default in the Gemini App.
• Imagen 4 (discontinued): Fast, Standard and Ultra have been deprecated since 2026, and the shutdown on 17 August 2026 has been completed. Shut down were imagen-4.0-generate-001, imagen-4.0-ultra-generate-001 and imagen-4.0-fast-generate-001, together with the Gemini 3 Image preview models. This affected only the API model IDs (Gemini Developer API and Vertex AI), not the Gemini App. The successor is Nano Banana.
• Veo 3.1 / Veo 3.1 Lite (Lite since 31 March 2026): 8 seconds per generation, 1080p/4K, native audio (dialogue, SFX, music), scene extension, camera moves.
• Lyria 3 (Preview): Google's music models in the Generative Media API. Lyria 3 Pro generates full-length pieces, Lyria 3 Clip short clips up to 30 seconds, Lyria RealTime music in real time. Since September 2026, Lyria 3.5 (lyria-3.5) has been available in the Gemini App, AI Studio, Flow Music and Vids. Relevant for sound beds under video content, not for royalty-free music in the legal sense (check the terms of use).
• Gemini 3.5 Transcribe (since 26 August 2026, new): Google's most accurate speech-to-text model, successor to Chirp 3. Detects more than 85 languages automatically, copes with code-switching mid-sentence, assigns up to 8 speakers (diarization), delivers word-level timestamps and can be biased towards up to 1,000 technical terms via custom vocabulary. "Smart Transcription" removes filler words ("um"), resolves self-corrections cleanly ("Tuesday, no, Wednesday") and formats automatically. Word error rate according to Artificial Analysis 4.0 % in streaming and 2.6 % without streaming, and time to final transcript drops by around 70 % compared with Chirp 3. Two variants: gemini-3.5-transcribe-live (Live API, sub-second latency) and gemini-3.5-transcribe (Interactions API, recorded audio up to 1 hour, with diarization and word timestamps up to 30 minutes). Where it is already in the product: dictation in the Gemini app on macOS (see chapter 17), Gboard on Android (Rambler) and Google Antigravity. Chrome has been announced. Relevant for training because dictation and meeting recordings in DACH-language quality get noticeably better, without you having to select the model yourself.
• Gemini Embedding 2 (GA), Gemini Robotics ER 2 (GA).

Important EU restriction (as of 15 October 2026): The following run in the EU multi-region of Gemini Enterprise and Vertex AI with data storage and ML processing in the EU: Gemini 4 Argon (since 13 October 2026), 3.6, 3.7 and 3.8 Flash, 3.5 Flash-Lite, 3.8 Live and 2.5 Pro. Argon is therefore the newest model permissible with EU residency. Gemini 3.1 Pro has neither storage nor processing residency in the EU. Correction: For strict single-region residency in Germany, Vertex AI offers europe-west3 (Frankfurt), but only with 3.5 Flash and 2.5 Flash; the Netherlands region (europe-west4) currently has no Gemini text model at all. The consumer Gemini App and Workspace EU have the full model family.

3.3 Limits

For the consumer plans Google no longer publishes absolute numbers, only multipliers relative to the free version. The limit refills every 5 hours until the weekly limit is reached.

• Free: Standard limits. 3.6 Flash plus limited access to 3.1 Pro.
• AI Plus: 2x the Free limits, Gemini Notebook included.
• AI Pro: 4x the Free limits.
• AI Ultra Standard / Max: 5x or 20x over AI Pro, depending on the subscription.

Context window per plan: Free 32k tokens, AI Plus 128k, AI Pro and AI Ultra 1 million tokens each.

Workspace editions (from the admin edition comparison, relevant for rollout planning):

• Business Starter ("Basic access", limits change frequently according to Google): up to 25 prompts per 4 hours on the thinking model, up to 300 prompts/day on the fast model, 32k context window, 20 images/day with Nano Banana 2, 3 images/day with Nano Banana Pro.
• Business Standard / Plus ("Pro access"): 1 million context window, 100 images/day Nano Banana 2, 100 images/day Nano Banana Pro, 20 Deep Research reports/day.
• Enterprise Standard / Plus: same Gemini feature parity as Business Standard/Plus (25 prompts per 4 hours on Thinking 3.1, 300 prompts/day Fast, 1 million context, Canvas, Scheduled Actions, music and video generation). Enterprise Plus differs only in security (CSE, Advanced Reports).

Gemini's web chat at gemini.google.com is the main interface for consumers and Workspace users. This is where most Gemini conversations happen, and where history, Gems and the Workspace integration come together.

4.1 Anatomy of the chat

• Model selection: Top left, "Gemini 3 Auto", "Pro", "Deep Think". Separate per conversation.
• Tool buttons in the input field: Plus for file upload, Image, Canvas, Deep Research, Audio Overview.
• Left sidebar: Gems, chat history, activity log. Workspace users additionally see Notebooks (cross-sync with NotebookLM since April 2026).
• Right sidebar (in Docs/Gmail): Side-panel Gemini with context on the current file.
• Response actions: Copy, Edit response, Regenerate, Share link, Export to Docs.

4.2 Workspace integration as the default

Anyone logged in as a Workspace user automatically has access to Drive, Gmail and Calendar as data sources. Via "@" mention in the chat: "@Gmail What has Anna written in the last two weeks?" or "@Drive Search for the Q2 forecast".

4.3 Audio Overview in the chat

New since 2025: Any conversation can be turned into an Audio Overview. Click the audio button and Gemini generates a podcast-style summary of the answer so far. Handy for listening on the commute. More in chapter 10.

4.4 Image and video generation in the chat

• Image: Via Nano Banana 2 directly in the chat. "Generate an image of ..." or via the image button ("Create images").
• Video: Via Veo 3.1 plus Gemini Omni for conversational video editing (Ultra/Pro). More in chapter 11.

4.5 Gemini Live integrated into the chat (since 19 May 2026)

With the I/O launch, Gemini Live is no longer a separate mode but seamlessly embedded in the chat. Switch between speaking and typing at any time without losing the conversation. Live can now:

• Connect apps during the conversation (Maps, Weather, Gmail, Drive, Calendar, YouTube, Tasks, Keep, Smart Home).
• Render real-time info visually: map snippet, weather card, photo preview directly in the voice session.
• Use cross-session memory: what was discussed in Live mode is available in the chat (and vice versa).
• Start Deep Research directly from Live (since 19 August 2026): A research run can be kicked off mid-conversation without switching to text mode. The report then lands in the chat thread. Rolled out to all Gemini App users (see chapter 9).

Productivity features in Gemini Live (since 26 August 2026): Live moves from conversation partner to executor. Four building blocks are added:

• Spark connection: By voice you can set up multi-step tasks across Docs, Sheets and Drive that then run on their own. Requires Google AI Pro. DACH note: Spark is still not available in the EEA, Switzerland and the UK (see chapter 3.2), so this building block does not apply here.
• Daily Brief by voice: On "What is my Daily Brief?" Gemini reads out the day's priorities from Gmail and Calendar without you looking at the screen. From Google AI Plus.
• Hands-free inbox management: Sort, search and tidy emails by voice, directly in Gmail.
• Personal Intelligence: Live remembers earlier conversations and links them with Gmail, Photos, Search and YouTube so you never start from scratch.

The most honest framing for training sessions: Daily Brief, inbox handling and Personal Intelligence are usable in DACH, Spark automation is not. Anyone who gets quoted the blog post should be able to separate the two.

Restriction: Gems, Notebooks, Omni and Lyria do not run in parallel during a Live session.

New Live model (since 15 September 2026, new): Gemini Live now runs on Gemini 3.8 Live Extended Thinking (see chapter 3.2). Google explicitly promotes it for everyday organisation: asking for your Daily Brief, working through your inbox by voice, handing off to-dos. The Spark components remain unavailable in the DACH region.

4.6 Interactive Responses (since 19 May 2026)

Gemini App answers are no longer just text. For Pro/Ultra, Gemini automatically renders for suitable topics:

• Zoomable high-resolution images (maps, diagrams, anatomy, architecture).
• Narrated Video Overviews, 30 to 60 seconds, with spoken voice-over plus visuals.
• Timelines for event and process answers.
• Interactive visualisations (since 19 August 2026): For suitable questions Gemini renders three-dimensional, rotatable models instead of static images. Rolled out together with the Student Hub to all Gemini App users, no longer just Pro/Ultra.
• Interactive simulations and models (since 24 August 2026, new): From a question Gemini builds not just an image but an operable model: molecular structures to rotate, physics simulations with adjustable parameters, process models to click through. Runs directly in the chat, without additional software. Available to all Workspace customers, Workspace Individual and personal accounts, controlled via the existing generative AI settings. For training sessions the fastest way to make an abstract topic explainable.

4.7 File generation from the chat (since 1 May 2026)

Gemini turns conversations directly into finished files, without a detour via Docs/Sheets/Slides. A prompt produces a formatted Google Doc, Sheet, slide deck or PDF that is displayed directly in the chat and saved to Drive.

• Typical triggers: "Turn this into a slide deck", "Make a sheet with the numbers", "Put this in a Doc", "Export as PDF".
• Availability: Workspace Business Standard and above, Google AI Plus/Pro/Ultra. EN first, DE follows.

4.8 Sharing

• Public link: Share button top right. Anonymised link, anyone with the link sees the conversation.
• Workspace sharing: Within Business and Enterprise with named people.
• Export to Docs: One click exports the answer as a Google Doc.
• Sharing via Drive (since 28 May 2026): Workspace users can share chats, Canvases and generated media from the Gemini App directly via Drive. The existing Drive sharing controls and org policies apply (DLP, out-of-domain warnings, audit log).

4.9 Study Notebooks

• What it is (since 25 June 2026): A personal learning area in the Gemini App. You state a learning goal, Gemini breaks it down into more than 100 concrete learning objectives, groups them into topics and delivers bite-sized lessons that build on your strengths and knowledge gaps.
• Diagnostic quiz: At the start Gemini creates an individual diagnostic quiz that determines your starting level and specifically uncovers strengths and gaps.
• Practice quizzes: Every lesson contains practice quizzes based on your uploaded material. The lessons update automatically according to your results and new uploads.
• Progress dashboard: Topics are marked as "Strengths", "Focus areas" or "Not started yet" and update automatically as you complete quizzes.
• Student Hub (since 19 August 2026): A dedicated entry point at gemini.google.com/students bundles Study Notebooks, interactive 3D visualisations and Deep Research in Gemini Live in one place.
• Availability: Since 25 June 2026 globally on the web via personal Google accounts, in all languages supported by the Gemini App. Since 19 August 2026 also on mobile and rolled out to all Gemini App users, together with the Student Hub.
• Workspace accounts (Rapid Release since 17 September 2026, new): Study Notebooks and the general notebooks in Gemini are now also available for Workspace organisations and Workspace Individual, provided the Gemini App and Gemini Notebook are enabled in the Admin console. Not in the EEA: for organisations in the European Economic Area the rollout only follows "in the coming weeks", so German and Austrian Workspace tenants are still waiting.

Gemini's prompt behaviour has become much more robust with Gemini 3. A clear task with context and expectations is enough. Tricks such as "You are an expert in ..." are no longer necessary.

5.1 Four building blocks of a prompt

• Goal: What should come out at the end?
• Context: Who is asking, in which role?
• Sources: @ mention of Drive, Gmail, Docs or file upload.
• Expectations: Format, length, style, language.

5.2 Gemini-specific strengths

• @ mentions: Direct access to Drive, Gmail, Calendar. "@Drive Q2-Forecast.xlsx" pulls the file in.
• Long context: Gemini 3 Pro handles up to 2 million tokens. Very long documents in one go.
• Multimodal: Mix text, image, video and audio in one prompt. "Here is a screenshot, here is an audio brief, write a reply that takes both into account."

5.3 Anti-patterns

• Too much at once: Gemini 3 can handle it, but the answer gets weaker in quality. Better two prompts.
• Asking in English, expecting German output: Inconsistent. The language of the prompt determines the language of the answer.
• Expecting live data: Gemini only fetches web data when Search is activated. For questions about current events, explicitly say "search current sources".

Personalization makes sure Gemini knows your persona, your tools and your style without you repeating it in every prompt. Three building blocks: custom instructions, Saved Info, Personal Context.

6.1 Custom instructions

Since 13 October 2026 in the new "Customise" menu in the left sidebar (web, mobile, Windows and macOS, all plans), which brings together response preferences, Skills and Connected Apps; previously Settings, Personalization, "Customize responses with your instructions". Free text, max. approx. 1,500 characters. Applies to all conversations outside Gems.

# About me
- Role: Marketing lead at DECAID
- Industry: AI enablement DACH

# How should Gemini answer?
- Language: German (even if I ask in English)
- Style: Direct, max. 18 words per sentence
- No emojis
- Sources with link

Custom instructions now in more Workspace apps (since 2 September 2026, new): Until now the stored instructions only applied to Gemini in Google Docs within Workspace. Since 2 September they also apply in "Ask Gemini in Drive", "Ask Gemini in Chat" and in the Gemini side panel of Slides, Sheets and Gmail. They are still maintained in one place under Personalization, and there is no admin switch. Rollout for Rapid and Scheduled Release over up to 15 days. Practical effect: Store "Answer in German, maximum 18 words per sentence" once, and the side panel in Gmail sticks to it without you repeating it per prompt.

6.2 Saved Info

Atomic facts that Gemini explicitly stores about you. "My name is Kiki", "My team has 12 people", "My main customers are X, Y, Z". Editable in Settings, Personalization, Saved Info.

6.3 Personal Context (Memory)

Implicit learning from the chat history. Gemini automatically distils facts and preferences without explicit memory entries. Can be deactivated separately.

6.4 Connected Apps for Personalization

Since 2026 Gemini can use your Connected Apps (Gmail, Drive, Calendar) as a personalisation source. From your communication Gemini learns who you regularly deal with, which topics recur and which tools you use. Deactivated by default, can be activated in the Personalization hub.

Gems are Gemini's counterpart to ChatGPT's Custom GPTs. Preconfigured persona bots with a name, instructions, knowledge files and (this is Gemini's USP) a public share link with which every recipient pulls their own copy. Fully shareable since September 2025.

7.1 What a Gem is

• Name plus description
• Instructions: System prompt, max. 8,000 characters. This is where everything the Gem should and should not do goes.
• Knowledge: Up to 10 files (Drive, PDF, Docs, TXT). Drive files are referenced live, so always current.
• Sharing setting: Private, with named people, with "Anyone with the link" or Public.

7.2 Premade Gems

Google ships several premade Gems that you can use right away:

• Brainstormer: Generate ideas, test options.
• Career Guide: Career coaching.
• Coding Partner: Code help.
• Learning Coach: Learning support for any topic.
• Writing Editor: Revise texts.

7.3 Creating your own Gem

1. Left sidebar, "Gems", "New Gem".
2. Enter name plus description.
3. Write the instructions (persona, tone, prohibitions, workflow).
4. Optionally attach knowledge files (max. 10).
5. Click "Save". The Gem is usable immediately.

7.4 Sharing mechanics (Gemini's USP)

Three sharing modes, all from the web (gemini.google.com), not from mobile:

• Private: Only you.
• With named people: Email list, recipients must be logged in.
• "Anyone with the link": The recipient does not need a Google account. Click the link, then "Add to my Gems", and their own editable copy is there.
• Public/Discoverable: Appears in Google Search. Caution: knowledge files end up in a Drive folder of the sender that is accessible to recipients.

7.5 Availability

• Free: Use premade Gems, creating your own is not possible.
• AI Plus, Pro, Ultra: Full, create your own Gems.
• Workspace Business Standard and above: Full, including side-panel Gems in Docs, Drive, Gmail.
• Education Plus: Full.

Canvas in Gemini is the two-column editor interface for longer writing and coding tasks. Chat on the left, a document or web app on the right that you both work on.

8.1 When to use Canvas

• Longer texts: Blog posts, briefings, memos, essays.
• Code: Develop, refactor, debug functions.
• Web apps: Since 2025 Canvas can generate a small HTML/JS app and render it live (mini demo apps, visualisations).
• Spec documents: Outline plus iterative refinement.
• Gemini 3 backbone (2026): Canvas runs on Gemini 3, for AI Pro and AI Ultra with a 1 million token context window. Enough for interactive dashboards, price sliders and algorithm visualisations from a single prompt.

8.2 Workflow

1. Click the "Canvas" tool button in the input field.
2. Give a first brief, Gemini generates a first version on the right.
3. Select a section in the document, in the mini menu "Edit", "Suggest Edits", "Adjust Length".
4. Give general instructions in the chat on the left: "Make the whole text more formal".

8.3 Export

• Export to Docs: One click, the Canvas moves to your Drive as a Google Doc.
• Code export: Download the code snippet as a file.
• Web app export: Single-file HTML for hosting elsewhere.

Deep Research in Gemini is a multi-step research agent. You ask a question, Gemini creates a research plan, autonomously searches hundreds of web pages (optionally also Gmail, Drive, Chat) and summarises the results in a structured way. 5 to 20 minutes of runtime, depending on depth. Now runs on Gemini 3 with an asynchronous task manager (notification on completion, robust error recovery). Since 21 April 2026: Deep Research Max with extended test-time compute for due-diligence depth.

9.1 When to use Deep Research

• Competitive briefing: "Who are the five most important competitors of Personio in DACH?"
• Market research: "How big is the DACH market for AI enablement consulting in 2026?"
• Literature review: "Which studies on AI adoption in the mid-market since 2024?"
• Due diligence: "What is publicly available about company X?"

9.2 Workflow

1. Click the "Deep Research" tool button.
2. Phrase the research task, as precisely as possible plus a description of the goal.
3. Optionally upload your own files and images as sources (since May 2026); they flow into the report in addition to the web research.
4. Gemini shows a plan, you can adjust it.
5. The research runs in the background, you can keep chatting in the meantime.
6. The result is a structured report in Canvas with citations.

9.3 Audio Overview twist

Any Deep Research report can be turned into an Audio Overview with one click. The research report as a podcast dialogue. Handy for commuters or as a briefing format for team calls.

9.4 Deep Research Max (since April 2026)

Second variant with extended test-time compute. For complex due-diligence cases or M&A research, choose Max mode. Runs longer, produces much deeper reports. Available in AI Ultra and Workspace Enterprise.

Note for automations: The old API agent model deep-research-pro-preview-12-2025 will be shut down on 23 October 2026. Its successors are deep-research-preview-04-2026 and deep-research-max-preview-04-2026.

9.5 Canvas and audio transformation

Every Deep Research report can be converted with one click into Audio Overviews, Canvas web pages, infographics or quizzes. Drastically increases the reusability of the research.

9.6 Availability

• Free: Heavily limited (10 runs/month).
• AI Pro: Standard Deep Research, regular access.
• AI Ultra / Workspace Enterprise: Deep Research Max (highest quality, ideal for due diligence).
• Available in 150 countries, 45+ languages (including German).
• Start from Gemini Live (since 19 August 2026): Deep Research can be started by voice in the middle of a Live session, rolled out to all Gemini App users.

Audio Overviews were the viral moment for NotebookLM in 2024 and have also been available in the Gemini App since 2025. From any content (chat, Deep Research report, notebook) Gemini generates a 5 to 30 minute podcast dialogue between two AI speakers.

10.1 What works

• From chat answers: Audio button below a longer answer.
• From Deep Research reports: Directly from Canvas.
• From notebooks (NotebookLM): With customisation (language, focus, length, style).
• From files (in the Gemini App): Upload a PDF, generate an Audio Overview.

10.2 Customisation

In NotebookLM you can steer the Audio Overview:

• Language: 50+ languages (German sounds very good).
• Focus: "Explain for beginners" or "In-depth discussion".
• Length: Default, Shorter, Longer.
• Interactive mode: You can intervene in between and ask questions.

10.3 Use cases

• Commute listening: Listen to briefings and reports on the way to work.
• Team briefing: 5-minute audio for stand-ups.
• Learner support: A complex topic presented as a dialogue is often more accessible than reading material.
• Marketing: Share a whitepaper as a podcast snippet.

Nano Banana 2 is Google's current image model and what you actually use in the Gemini App, Veo 3.1 (Lite since 31 March 2026) the video model. Both usable directly in the Gemini App plus in Workspace via Slides/Docs and in Google Vids (10 free videos/month for personal accounts since April 2026).

11.1 Nano Banana (current image family)

Invoked via the "Create images" tool in the input field or by voice ("Generate an image of ..."). Three tiers:

• Nano Banana 2 Lite (Gemini 3.1 Flash-Lite Image, since early July 2026): fastest and cheapest variant, around 4 seconds per image. For high volume and prototyping.
• Nano Banana 2 (Gemini 3.1 Flash Image, since 26 February 2026): the default in the Gemini App, balanced between speed and quality. Variants Fast / Thinking / Pro.
• Nano Banana (Gemini 2.5 Flash Image, the first generation): still usable. Correction (as of 15 October 2026): The previously announced shutdown on 2 October 2026 did not happen; Google is keeping the API ID, but since 18 September 2026 only for projects that already use it. Anyone still using it in an automation should switch to Nano Banana 2.1.
• Nano Banana 2.1 (gemini-nano-banana-2.1, GA in the Gemini API since 6 October 2026, new): successor to the API variant of Nano Banana 2, also in the Vertex EU multi-region. The new default choice for automations and custom tools.
• Nano Banana Pro (Gemini 3 Pro Image): higher-resolution professional tier for final art, print and hero visuals.

Strengths: text rendering (packaging, comics, logos), photorealism, art styles (impressionism, abstract, illustration), multilingual prompts (including German). Existing images can be edited further conversationally ("Make the whole thing warmer ..."). Aspect ratio selectable as square, 16:9, 9:16 and 4:3. Available in the Gemini App, AI Studio, Whisk (Labs sandbox), Vertex AI, Workspace, Google Photos and NotebookLM.

11.1a Predecessors (Imagen 4 and Imagen 3)

Imagen 4 was Google's image model until early 2026 (Fast 0.02 USD, Standard 0.04 USD, Ultra 0.06 USD per image, native 2K resolution). All Imagen models are now deprecated, shutdown on 17 August 2026. Anyone who built their own applications against an Imagen model ID has to migrate to Nano Banana by then. The old marketing page at deepmind.google still advertises Imagen 4, but Google's deprecations registry is what counts.

11.2 Veo 3.1 (instead of Veo 2)

• Veo 3.1: 8 seconds per generation, scene extension for longer cuts, 1080p or 4K.
• Veo 3.1 Lite (since 31 March 2026): <50 % of the cost of Veo 3.1 at comparable speed, ideal for high volume.
• Native audio: Dialogue, ambient, SFX, music. Benchmark-leading audio-video alignment.
• Creative controls: Reference images for character consistency, camera moves (zoom/pan/dolly), first/last frame, object insert/remove, motion paths.
• Availability: Gemini App (Pro/Ultra), Google Flow, Google Vids (also for personal accounts, 10 free videos/month since April 2026), AI Studio, Gemini API, Vertex AI.
• Invocation: In the chat "Generate video of ...".

11.3 Gemini Omni for video creation (since 19 May 2026)

With Google I/O 2026, Gemini got a conversational video creation layer on top of Veo. Instead of fixing a prompt and waiting for the finished video, you build in the chat:

• Mix of text, photos and clips: Upload sources, describe what should come out, iterate conversationally.
• Custom AI avatars: Your own presenter avatars that guide through the video.
• Iterative editing: "Make scene 2 shorter", "swap the voice-over", "add an outro slide", without a new full render.

Rolled out globally for all Google AI subscriptions (18+), via the Gemini App and Google Vids. Cannot be combined with Live mode.

Gemini Omni 1.1 Flash (since 27 August 2026): The model behind the video layer is production-ready and gains real directing control:

• Scene extension: Extend scenes in 10-second steps up to 40 seconds. The model takes up to 10 seconds of lead-in as context, so look and narrative flow stay stable.
• Keyframe control: Define the first and last frame, Gemini computes the transition. This enables targeted camera moves and clean loops.
• 360p draft mode: Drafts around 60 % faster and at a third of the cost of 720p. Iterate first, then render.
• Upscaling to 1080p or 4K for the final version.
• Video references up to 3 seconds keep characters and visual context consistent across scenes.

Available in AI Studio and the Gemini Enterprise Agent Platform, since 27 August 2026 additionally in Google Flow and the Gemini App for AI Plus, Pro and Ultra.

Expansion in Google Vids (since 16 July 2026):

• Better clip quality and video editing by text: Noticeably cleaner text rendering and physics. Existing videos can be edited by instruction (colour grade, styles, remove objects), including third-party footage.
• Personal avatar: After a verification process you can cast yourself as an avatar in AI clips. English only, 18+.
• Control emotion and pace: Bracket tags such as [excitedly] control emphasis, pace and sound effects of AI avatars and voice-overs. Auto-tagging is available. Available for all Business, Enterprise and Education Plus plans as well as Google AI Pro.

11.3a Google Pics (since 1 September 2026, new)

Google Pics is a standalone image application in Google Workspace that runs on Nano Banana. It closes the gap between "generate an image in the chat" and "revise an image precisely".

• What it can do: Import images from your computer, from Google Drive or from Google Photos, generate or refine them by text prompt, select individual objects in the image and change them specifically, and edit text within an image. Intended for posters, social posts and illustrations.
• Where it appears: as a standalone application and embedded in Workspace apps, at launch in Slides, Docs and Drive.
• Availability: Business Standard/Plus, Enterprise Standard/Plus, Google AI Pro and Ultra, Google AI Pro for Education and AI Expanded Access. No surcharge, included in the subscription.
• Rollout: Rapid Release from 1 September 2026, Scheduled Release from 15 September 2026, each gradually over up to 15 days.
• Admin: on by default, can be switched off per domain, OU or group (see chapter 18.1).

11.4 Privacy

Generated images and videos carry SynthID watermarking (for AI detection). In Workspace plans, outputs are not used for training; in personal plans optionally.

File upload in Gemini is generously sized. PDFs, Word documents, spreadsheets, images, audio, video: upload once, Gemini extracts and answers questions.

12.1 Accepted formats

12.2 Limits

• Per prompt: 10 files
• Per conversation: Max 2 GB total
• Long context: Gemini 3 Pro handles up to 2 million tokens, very long documents in full

12.3 What happens to the content

• In the Gemini App: Files are read in the current conversation, not persistent.
• In Gems with knowledge: Files are stored as Gem knowledge, persistently available.
• In NotebookLM: Files are stored as sources, permanently.

Gemini's deepest strength: the side panel in Docs, Sheets, Slides, Gmail and Drive. Context-aware, with live access to the current document and all your Workspace data.

13.1 In Google Docs

• Help me write: First drafts, rephrasing, changing tone.
• Summarize: Boil long documents down to a bullet list.
• Custom instructions in Docs (since May 2026): Store your own instructions, tone and style guidelines per document.
• Across tabs: Gemini knows all tabs of a document and can synthesise between them.
• Gemini in Docs in 19 languages (since 16 July 2026): "Help me create", "Help me write", "Match writing style" and "Match doc format" now support 11 additional languages (including Mandarin, Dutch, Polish, Turkish, Czech, Swedish, Danish, Norwegian). German was already included. Runs under Workspace Intelligence, available in Business Standard/Plus, Enterprise Standard/Plus, Education Plus and Google AI Pro.
• Visuals with Gemini in Docs (since 28 July 2026): Gemini now generates and edits diagrams and infographics directly in the document by prompt, not just text. In addition there are Gemini-supported comment workflows: summarise comment threads, suggest replies and extract topics from many comments.
• Gemini Notebook as a source in Docs (since 23 September 2026, new): Using "@" in the Gemini side panel or the bottom bar, you can specify a notebook as context; Gemini then answers with inline citations to the notebook sources. Available in Business Standard/Plus, Enterprise Standard/Plus, Education Plus and Google AI Pro/Ultra; Workspace smart features must be enabled.
• Markdown files natively in Docs and Drive (since 5 October 2026, new): .md files can be edited and commented on together in Google Docs without conversion, and Drive shows a rendered preview. Google explicitly cites collaboration between people and agents as the reason. Handy for prompt libraries and SKILL.md files (chapter 7). For all Workspace and personal accounts.

13.2 In Google Sheets

• Help me organize: Understand data, suggest table structure.
• Formula generation: Derive a formula from natural language.
• Fix formula errors (since 22 June 2026): Gemini in Sheets explains formula errors (e.g. #REF!, #VALUE!) and suggests the correction directly. Increased usage limit as a promo until 15 July 2026, after that the regular Pro user limits apply.
• Charts and pivots: From an instruction, without a click cascade.
• Data clean-up: Detect duplicates, inconsistencies, missing values.
• Anomaly detection in Connected Sheets (since 27 May 2026): Zero-shot anomaly detection on BigQuery data via TimesFM, without SQL or model training. Handy for revenue, traffic or sensor series.
• Build and edit tables in 28 languages (since 18 June 2026): Creating and editing tables with Gemini in Sheets now supports 28 additional languages, including German, French, Italian, Spanish, Portuguese, Dutch, Polish and Japanese. Previously this function was limited to English.
• Fill with Gemini in 19 languages (since 7 July 2026): Fill with Gemini and the AI function (generate text, summarise, categorise, analyse sentiment, directly in the cells) are now available in 19 languages. German was already included, 11 more languages are new (including Mandarin, Dutch, Polish, Turkish, Swedish). From 15 July 2026 higher usage limits apply to licences with AI Expanded Access.
• New chart types (since July 2026): Sheets can now create and import combo charts (since 20 July 2026) and display scatter charts with multiple x series (since 29 July 2026), including import from Excel.
• Gemini in Sheets on Android (rollout from 9 September 2026, new): The analysis functions of Gemini in Sheets are now also available in the Android app: ask questions about your own data, generate analyses and build charts, with suggested prompts such as "Summarise this table". Limit: Complex editing, formatting and formula generation remain reserved for the web. Available in Business Standard/Plus, Enterprise Standard/Plus, Google AI Pro for Education and Google AI Pro and Ultra.
• Sheets Canvas: interactive mini apps (since 13 August 2026): Gemini turns a spreadsheet into an interactive mini app by prompt, for example a Kanban board, a tracker or a dashboard. The app reads from and writes back to the spreadsheet, with no code at all. Included in Business Standard/Plus, Enterprise Standard/Plus, Google AI Pro/Ultra and AI Pro for Education. Rapid Release since 10 August 2026, Scheduled Release from 31 August 2026. For training sessions the fastest way to turn an existing list into something operable without going through AppSheet.

13.3 In Google Slides and Vids

• Generate slides from a brief: First slide sequence on instruction.
• Native, editable presentations with Gemini in Slides (since 30 June 2026): Gemini now creates complete, multi-page and fully editable slide decks that are grounded in your Drive files and adopt the existing style. Previously there were only individual slide suggestions.
• Insert image: Via Nano Banana directly from the slide.
• Speaker notes: Auto-generated from slide content.
• Convert Slides to Vids with AI avatars (since 15 May 2026): When converting Google Slides to Google Vids, AI avatars are now available as presenters, in addition to the existing auto-script and voice-over.
• Slides to video in 7 more languages (since 8 July 2026): Converting Google Slides into a video with AI-generated script, voice-over, music and animations now supports German, French, Italian, Japanese, Korean, Portuguese and Spanish in addition to English.
• Expanded AI avatars in Google Vids (since 17 June 2026): The avatar selection grows from 23 to 53 presets and now covers 24 languages (16 new). Custom avatars can additionally be steered by text prompt (gestures, tone, emphasis).
• Record presentations directly in Slides (Rapid Release from 20 August 2026, Scheduled Release from 7 September 2026): A new "Record" entry in the Slides menu records the presentation including camera and audio directly from Slides, with transcript editing and generated voice-over via Google Vids. For all Workspace customers, Workspace Individual and personal accounts, without an admin switch.
• Documents to video summaries in Google Vids (Rapid Release since 2 September 2026, Scheduled Release from 5 September 2026, new): Google Docs, PDFs and Word files can be turned directly into an explanatory video. Vids writes the script and narration and adds matching visuals. Included in Business Starter/Standard/Plus, Enterprise Starter/Standard/Plus, Education Plus, Nonprofits, Google AI Plus/Pro/Ultra and Google AI Pro for Education. No admin switch. For training: the shortest path from an existing document to an onboarding clip, but proofread beforehand, because the script shortens and weighs.
• Co-presenter in Google Meet with one click (rollout from 31 August 2026, new): In the "Ask Gemini in Meet" panel Gemini suggests a second person as co-presenter for the running Slides presentation, and one click hands over control. Included in Business Standard/Plus, Enterprise Standard/Plus, Frontline Plus and Google AI Pro for Education. On by default when "Ask Gemini in Meet" is active.
• Longer Veo videos and multi-generation in Vids (since 17 June 2026): Veo clips in Google Vids can now be extended with character consistency, and several clips can be generated at the same time. Saves time on longer video sequences.
• Gemini Omni 1.1 Flash in Vids (since 23 September 2026, new): Video extensions with more consistent characters and audio, generation and upscaling in 1080p, and control over clip duration. Included in Business and Enterprise Starter/Standard/Plus, Education Plus and Google AI Pro/Ultra.
• More natural AI voiceovers in Vids (since 29 September 2026, new): Vids now uses Gemini 3.8 Flash-Lite TTS for narration: more natural voices, more realistic pacing, context-aware emphasis and faster generation. Same editions as for Omni 1.1 Flash.
• Caption styling in Vids (since 30 September 2026, new): Font, position and animation of captions can now be customised. Useful for training videos, which are often watched without sound.

13.4 In Gmail and Chat

• Help me write: Email drafting with style options.
• Help me write with context (since 8 May 2026): Drafts now automatically pull in relevant content from Drive and Gmail history and match the tone of your earlier emails to the same recipient. Fewer generic drafts, less rework.
• Summarize thread: Summarise long discussion threads.
• Smart Reply: Three reply suggestions under every thread.
• Inbox search by natural language: "Which emails from Anna are unanswered?"
• Message refine in Google Chat now multilingual (since 14 May 2026): Message improvement (tone, clarity, grammar) now supports German, French, Italian, Japanese, Korean, Portuguese and Spanish in addition to English.
• Free-form refining in "Help me write" in Gmail (since 14 July 2026): Instead of the fixed preset options ("formalise", "shorten") you can now refine the draft with your own prompt, including undo and redo. Available in all Business and Enterprise plans as well as Google AI Plus/Pro/Ultra.
• Group chats with external people (since 15 July 2026): In Google Chat, group direct messages with several external people can be started by email invitation. Controlled via the existing admin setting for external spaces.
• Reply-all BCC warning (since 30 July 2026): Gmail now warns when you reply to an email on which you were only BCC'd, before your reply unintentionally goes to the whole distribution list.
• "Ask Gemini in Chat" (rollout from 26 August 2026): Google Chat gets a unified command line: invoked via the shortcuts menu or with Ctrl+G (Windows/ChromeOS) or Cmd+G (macOS). This lets you search across Gmail, Drive and Calendar without leaving the chat, generate an image or an update, schedule a meeting and catch up on the backlog across all conversations. Available in Business Standard/Plus, Enterprise Standard/Plus and Google AI Pro for Education, for Rapid and Scheduled Release over up to 15 days. Prerequisite: Gemini in Chat and Workspace Intelligence are active, then the function is there automatically. Initially only with English account language, further languages to follow. For DACH teams it therefore only becomes a training topic with the language rollout.
• AI Overviews in Gmail search (Rapid from 3 September, Scheduled from 21 September 2026, new): Type questions in natural language into Gmail search and Gemini answers with a summary drawn from the matching emails. For Business and Enterprise editions and Google AI Plus/Pro/Ultra. English only for now; personal accounts in the EEA, Switzerland and the UK are excluded.
• Voice instead of typing in Gmail, Docs and Keep (since 9 September 2026, new): Emails, texts and notes can be dictated and revised by voice. In Gmail and Keep from Google AI Plus, in Docs from Google AI Pro.

13.5 In Drive

• Search by natural language: Semantic search instead of filename search.
• File summarisation: Click-and-summary for PDFs, Docs, Sheets.
• Multi-file Q&A: One question about several files at the same time.
• Organize My Files (GA since 1 June 2026): Gemini suggests how loose files should be sorted into your existing folder structure and files them on confirmation. Helpful when tidying up Drives that have grown over time.
• Gmail as a source in "Ask Gemini in Drive" (GA since 3 June 2026): In "Ask Gemini in Drive", Gmail threads can now be included as an additional source, so that one question delivers answers across emails, files and folders.
• "Ask Gemini in Drive" now on mobile (since 30 June 2026): Multi-step conversations about Drive content now also work in the Android and iOS apps, no longer only on the web.
• AI Overviews in Drive on mobile (since 30 June 2026): Gemini search summaries across several Drive files are now also available on mobile (Android and iOS).

13.6 In Google Forms

Create a quiz by prompt (since 30 July 2026): Gemini in Google Forms builds a finished quiz from a description, including questions, answer options and grading. Files from Drive can also be specified as the basis, for example a training document or a slide deck. Included in Business Standard/Plus, Enterprise Standard/Plus, Education Plus, Google AI Pro/Ultra, AI Pro for Education and Teaching and Learning.

For training this is the fastest path from the material to a knowledge-check quiz. Caution: Gemini likes to generate questions where the correct answer is the longest option, and often puts it in the same position. Review once and shuffle the options before rolling out.

13.7 Live data vs. snapshot

Gemini accesses live data, not snapshots. If you change a Doc and ask Gemini, it sees the current version. Unlike Custom GPT knowledge, which is static.

13.8 Cross-app orchestration (since 2 September 2026, new)

The biggest change to the Workspace integration so far: Gemini is no longer tied to the app you are currently sitting in. From Gmail you get a formatted Doc or a finished slide deck, from Docs a calendar appointment, from the side panel a task. You no longer have to switch apps to file the result where it belongs.

• What is possible: Generate documents and spreadsheets, build slide decks, draft emails, schedule meetings and create tasks, each from the Workspace app currently open.
• Availability: Business Standard/Plus, Enterprise Standard/Plus, Frontline Plus (scheduling only), Google AI Pro (without scheduling), Google AI Ultra, Google AI Pro for Education and AI Expanded Access.
• Rollout: from 2 September 2026 for Rapid and Scheduled Release, gradually over up to 15 days. No separate admin switch, it works via the existing Gemini enablement.

Gemini Notebook is Gemini's sister product for source-grounded knowledge work. You upload up to 300 sources (PDFs, Docs, web links, YouTube, audio files) and Gemini answers exclusively on that basis. With citations, Audio Overviews, mind maps, Video Overviews and, since 2026, cross-sync with the Gemini App.

14.1 What Gemini Notebook can do

• Source-grounded Q&A: Answers are based exclusively on the uploaded sources. With a source pin per answer, you can call up the original passage.
• Audio Overview: Generates a podcast-style dialogue between two AI speakers about your content. 5 to 30 minutes, depending on the volume of sources.
• Video Overview: Slide video with voice-over, similar to the Audio Overview but visual.
• Mind Map: Topic map from the sources.
• Briefing Doc, FAQ, Study Guide, Flashcards, Quiz, Slide Deck, Infographic: Various structured outputs from the sources.
• Notes: Your own notes, which serve as an additional source.
• Copy notebook (since 17 August 2026): A complete notebook can be duplicated including sources and Studio content, instead of setting everything up again. Ideal for template notebooks that are cloned per project or per customer. For all Workspace customers and personal accounts with Notebook access. This is controlled not by the admin but by the notebook owner via "Allow copies" in the sharing settings (default: on).
• Secure Cloud Computer (since 16 July 2026): Every notebook now has a secure cloud environment in which Gemini can natively write and execute code, grounded in your own sources. Ideal for complex data analyses directly from the uploaded files. Available for Google AI Ultra as well as Workspace Business with AI Ultra or Expanded Access, AI Pro users on the web to follow.

14.2 Which source types

14.3 Sharing

• Private (default): Only you.
• With named people: Email list, viewer or editor.
• Public link: Since June 2025. Anonymous viewers can ask Q&A plus generate audio/video/mind map, but cannot change sources.
• Featured Notebooks: Google curates public example notebooks from partners (Economist, Atlantic etc.).
• Admin controls for external sharing (rollout from 10 September 2026, new): Until now, external sharing of Gemini Notebook could not be controlled granularly for Workspace organisations. Now there are four levels: off (default), trusted domains only, on and on including public notebooks. Configurable per domain, OU or group, for all Workspace customers. Important for DACH: The default is "off", so anyone who wants to collaborate with customers via a notebook today needs a deliberate admin decision.

14.4 Expert Intelligence (since 27 August 2026)

Google is opening Gemini Notebook to licensed third-party content. Expert Intelligence is a Google-wide initiative that brings curated sources from publishers, authors and experts into the AI products. Gemini Notebook is the first product, the Gemini App and AI Mode in Search are to follow.

• What is possible: E-books you have bought at Google Play Books can be added to a notebook as a source. Answers are then grounded directly in the book, and you can generate infographics, Audio Overviews or quizzes from it.
• Scope at launch: more than 100,000 titles, including from Penguin Random House, Macmillan and O'Reilly Media. Plus Featured Notebooks from more than 15 authors, including Steven Pinker and Michael Pollan.
• Licensing logic: Anyone who wants to work with a book in a notebook must own it themselves at Play Books. When a notebook is shared, co-readers only see the book content if they buy their own copy, otherwise the source remains locked for them.
• Planned: Third-party subscriptions, business research reports and textbooks as further source types.
• Workspace availability (since 17 September 2026, new): Expert Intelligence is now open to all Workspace editions, provided Gemini Notebook, Google Play and Google Books are enabled in the Admin console. Google does not explicitly say whether the book catalogue already works in German tenants; test in your own tenant before a training session.

14.5 Cross-sync with the Gemini App (since April 2026)

Since the rollout in April 2026, notebooks are available in both apps:

• NotebookLM side: Audio Overview, Video Overview, Mind Map, quizzes, Study Guide.
• Gemini App side: Deep Research, Canvas, Veo video from notebook sources, Scheduled Actions, binding a Gem to a notebook.
• Data flow: The source lives physically in the notebook namespace, the Gemini App only references it. Deleting in one app = global soft delete (30 days recovery).

14.6 Plans (as of May 2026)

New limit model (since 2 September 2026, new): The fixed daily values in the table are being replaced by compute-based quotas. The quota renews every five hours up to a weekly cap and depends on complexity, model and feature: AI Plus gets twice the standard limit, AI Pro four times, AI Ultra 5 or 20 times the Pro limit. New is "Generate later" (web only): generations start automatically as soon as quota is available again. The table remains a guide for the caps on notebooks and sources.

14.7 New since early 2026

• Video Overviews as the visual counterpart to the Audio Overview; Ultra has a cinematic variant.
• Studio panel redesign: 4 tiles (Audio / Video / Mind Map / Reports), several outputs of the same kind per notebook, parallel use.
• Audio Overview formats Brief / Critique / Debate.
• Deep Research with Fast/Deep split directly in NotebookLM.
• New source types: Google Sheets, Drive URLs, Word DOCX, images, PDFs from Drive.
• Data Tables with Sheets export.
• Chat engine upgrade: 8x context, 6x memory, +50 % quality, custom personas / goals.
• Notebooks in Gemini (bidirectional, since April 2026).
• 40+ languages including German and Swiss German.
• Education Plus expansion (April 2026).
• "Ask NotebookLM" in Workspace Studio (since 15 May 2026): A new automation step in Workspace Studio uses an existing NotebookLM notebook as a knowledge source and delivers grounded responses within no-code flows. Handy for curated onboarding, policy lookups and knowledge bots without a separate RAG pipeline.
• Drive auto-sync (since 26 May 2026): Drive documents included as sources synchronise automatically. Content changes, deletions and permission changes take effect directly, no manual re-upload needed any more.
• NotebookLM LTI for PowerSchool Schoology + Moodle (since May 2026): Course materials from Schoology or Moodle can be included directly as sources in notebooks. Completes the Education Plus integration alongside Canvas.
• New learning tools (since 15 September 2026, new): Real-time voice conversations with your own sources in the mobile app (almost 100 languages, from age 18), an audio recorder for lectures and meetings, interactive Learning Overviews with new quiz formats (short answer, multiple choice, fill in the blank) and short Video Overviews of around 60 seconds in more than 80 languages. For all Workspace editions and personal accounts.

14.8 Agentic upgrade (since 8 June 2026)

The biggest NotebookLM update of the year: NotebookLM now runs on Gemini 3.5 plus Antigravity (around 65 percent win rate against the predecessor system) and becomes much more agentic.

• Agentic research with code execution: Every notebook gets a secure cloud computer that writes and executes code on its own. More than 100 curated "skills" for deeper analyses (data evaluation, charts, calculations).
• Source Discovery: You can now start from a loose idea or a chat, NotebookLM builds the knowledge base itself and suggests sources via Google Search (previously you had to bring all sources yourself). Control and attribution stay with you.
• Expanded output formats: In addition to audio, video and mind map now also charts, PDFs (with diagrams and tables), Excel (.xlsx), PowerPoint (.pptx), DOCX and Markdown, CSV/JSON as well as images via Nano Banana.
• Availability (gated): Initially for Google AI Ultra subscriptions and Workspace customers with "AI Ultra Access" or "AI Expanded Access". Since 16 July 2026 the rollout to Google AI Pro has been announced ("over the coming weeks"), but as of this guide it has not yet shipped there.

Scheduled Actions in Gemini let you run tasks at set times. "Send me a summary of this week's appointments every Monday at 8 am" or "Remind me of the quarterly planning on the 15th".

15.1 Setup

1. Phrase it in the chat: "Schedule an action to do X every Monday at 8 AM".
2. Gemini confirms with a plan overview.
3. Manageable in Settings, Scheduled Actions.

15.2 Use cases

• Daily briefing: "Show me my emails from the last 24 hours plus today's calendar every morning at 7 am".
• Weekly recap: "Every Friday a summary of my Drive plus highlights from Gmail".
• Reminders: "Remind me of the quarterly planning on the 1st of every month".
• Research trigger: "Check once a week whether a new pricing page has appeared on [competitor]'s site".

15.3 Workspace Studio: new steps (since 8 May 2026)

• Calendar "Block time" step: Automations can now set focus times or buffer blocks in the calendar, not just appointments.
• Meet starter expanded: A starter can address several meetings and pull both transcripts and "Take notes" notes as input.
• Granular admin controls (since 27 May 2026): Admins can enable or block individual Studio steps and starters per service or individually, for staged rollouts.
• "Repeat for each" step (since 2 June 2026): A new loop step lets flows iterate over a list or over Sheet rows and run the same sub-flow per entry. This makes batch automations (e.g. one email or one Doc per row) possible without a workaround.
• "Add source to Gemini Notebook" step (since 6 August 2026): A flow can automatically feed sources into an existing Gemini Notebook: free text, links to Drive files as well as YouTube and web URLs. This lets you fill notebooks continuously instead of adding sources by hand one by one (the counterpart to the "Ask" step from chapter 14.7). Available in Business Starter/Standard/Plus, Enterprise Standard/Plus, Education editions and AI Expanded Access; enabled as soon as the Gemini steps are active.
• Enterprise security controls (admin settings immediately, user features from 20 August 2026 Rapid / 1 September 2026 Scheduled): The biggest governance step for Studio flows so far. New are agent identities for automated flows, granular access management, auditing and DLP protection at flow level. Admins control step types, approval requirements, DLP rules and webhooks. Available in Business Starter/Standard/Plus, Enterprise Starter/Standard/Plus, the Education editions and AI Expanded Access; DLP only in Frontline Standard/Plus, Enterprise Standard/Plus, Education Standard/Plus and Enterprise Essentials Plus. Default: active for eligible users. "Identity Attribution" runs as a beta. Relevant for DACH: For the first time, Studio automations can be properly secured against works council and GDPR requirements, because it becomes traceable under which identity a flow accesses which data.
• New steps for Drive, Gmail and Chat (Drive and Chat from 8 September 2026, Gmail from 14 September 2026, new): Four new automation steps close the biggest gaps in the flow toolkit: move file and copy file in Drive, reply to an email in Gmail and send a Chat reply (with Markdown formatting). This lets you build filing routines and reply workflows without a script. Available in Business Starter/Standard/Plus, Enterprise Standard/Plus, the Education editions, Google AI Pro for Education, Teaching and Learning and AI Expanded Access. Admin controls are available one week before the user features in each case (Drive and Chat from 1 September, Gmail from 8 September); individual steps can be blocked and made subject to approval for external data transfer.
• Seven new UI languages (since 8 May 2026): Workspace Studio is now additionally available in German, French, Italian, Japanese, Korean, Portuguese and Spanish.
• Custom starters, custom steps, third-party integrations and webhooks (Rapid from 21 September, Scheduled from 30 September 2026, new): Flows now react in real time to events in other apps (custom starters), run their own logic, for example via Apps Script (custom steps), connect in beta to Asana, Confluence, HubSpot, Jira, Mailchimp, QuickBooks, Salesforce and Slack, and send HTTP requests to external endpoints (webhooks). Admin default: off. A webhook allowlist is only available in Business Plus and Enterprise and Education Standard/Plus. Important for DACH data protection: via webhooks, data leaves the Workspace tenant, so define the permitted target endpoints before switching it on.
• New logic and search steps (rolling out from 13 October 2026, new): Nested conditions, an option in the Gmail starter to skip replies in the same thread, a Meet step with more calendar metadata (full attendee list plus a template for the follow-up email) and the new Drive step "Search for a Doc". On by default in all editions with Workspace Studio.
• "Routines for this meeting" after the Meet call (rolling out from 5 October 2026, new): On the screen after a meeting, Gemini and Workspace Studio suggest ready-made routines. Two are available at launch: a follow-up email draft in Gmail and the action items as Google Tasks. "Take notes for me" must have been running; the routine can be enabled for individual meetings, recurring meetings or all meetings. On by default; admins can turn it off via the Studio settings or the take-notes setting per domain, OU or group. For Business and Enterprise Standard/Plus as well as Google AI Pro for Education.

15.4 Limits

• 10 active actions per user
• Available in: AI Pro, Ultra, Workspace Business Standard and above, Education Plus
• Connected Apps must be active, otherwise the trigger cannot be executed
• New location in the app (rollout from 28 September 2026): In the Android and iOS app, the "Scheduled" entry in the redesigned side panel replaces the previous "Scheduled actions" page.

Connectors link Gemini with external services. In the consumer area mainly Google Workspace plus YouTube, Maps, Spotify. In the enterprise additionally Salesforce, Jira, Confluence, ServiceNow, SharePoint via Workspace connectors.

16.1 Native Workspace connectors

• Drive: Read and search files, create new ones.
• Gmail: Read and search emails, write drafts.
• Calendar: Read, suggest, plan appointments.
• Tasks: Create to-do lists, tick them off.
• Keep: Read notes.
• Maps: Route planning, place lookups.
• YouTube: Process video transcripts.
• New Connected Apps (rollout from 23 September 2026, new): Airtable, Linear, monday.com, PandaDoc, Wispr AI, Zoho, Adobe, Picsart, Squarespace and Webflow, plus lifestyle services such as Peloton and SeatGeek. Enable them in settings or via @-mention in the chat; the apps can read data and carry out actions. Google does not name any countries; according to trade press most are initially only available for personal US accounts. Google Wallet as a Connected App (since 28 September 2026) is also US-only.

16.2 Enterprise connectors (Gemini Enterprise)

Available via the Agent Marketplace: Salesforce, ServiceNow, Jira, Confluence, SharePoint, Box, Dropbox, Workday, Zendesk and many more.

16.3 Custom MCP servers

Since 2026 Gemini Enterprise supports custom Model Context Protocol servers. This lets you connect your own data sources (internal APIs, proprietary systems), compatible with the Anthropic/OpenAI world.

Workspace MCP Server (public preview since 1 May 2026): Google offers an official Workspace MCP server for developers. Standardised access to Gmail, Drive, Calendar, Docs from third-party agents (Claude, ChatGPT, your own agents). Accompanied by a new tiering model for all agent tools and APIs. Since 30 September 2026, tools can read and create comments via the Docs, Sheets and Slides APIs and submit changes as suggestions via the Docs API instead of overwriting directly. The MCP server itself remains in developer preview.

MCP connectors in Gemini for Workspace (since 15 September 2026, new): Gemini in the side panel of Docs, Sheets and Slides and in Google Chat now talks via MCP to Asana, Atlassian Rovo, HubSpot, Intuit Mailchimp, Intuit QuickBooks, Monday and Salesforce. For Business Standard/Plus, Enterprise Standard/Plus, Enterprise Essentials Plus and Google AI Pro/Ultra, English only for now. On by default, controlled under Apps > Google Workspace Marketplace apps > Apps list per domain, OU or group (see chapter 18.2).

16.4 Skills in Chrome (since 8 May 2026)

A new surface for Gemini prompts directly in the Chrome browser. Recurring prompts can be saved as a "skill" and triggered with one click on any web page. Examples: "Summarise this page as a 5-point brief", "Translate into German", "Extract the price table as a sheet". Available for eligible Workspace users.

Language and region expansion (since 16 June 2026): Gemini in Chrome is now rolling out to further languages and regions (including Latin America, Africa, the Middle East). DACH availability is already active, but the feature set keeps growing.

16.5 Setup

• Personal: Settings, Connected Apps. Click "Connect", OAuth flow.
• Workspace: The admin enables in the AI Control Center what users are allowed to connect.

Project Astra is Google's research line for a "universal AI assistant": multimodal, with live video, audio, screen sharing, computer use. Not all Astra features are GA, but parts are available as Gemini Agent.

17.1 What works today (June 2026)

• Gemini Live: Real-time voice with live video and screen sharing. Since I/O 2026 seamlessly integrated into the chat (no separate full-screen mode any more), seamless switching between voice and text in the same thread, real-time info cards and deeper app connections. Available in the mobile app (Android and iOS).
• Live Translate (since I/O 2026): Based on Gemini 3.5 Live Translate, fluent, natural-sounding speech-to-speech translation in real time directly in the Live conversation. Automatically detects more than 70 languages, preserves emphasis and intonation and additionally rolls out via the Google Translate app.
• Gemini Agent: Multi-step browser tasks, similar to ChatGPT Agent. AI Ultra in the US, rollout to other regions.
• Computer use native in Gemini 3.5 Flash (since 24 June 2026): Computer use is now a built-in tool of Gemini 3.5 Flash (previously only a separate 2.5 computer-use model). The agent can see, reason and act across browser, mobile and desktop. With enterprise protection: explicit confirmation before sensitive actions and automatic stop on detected prompt injection. Available via the Gemini API and the Gemini Enterprise Agent Platform.
• Gemini Spark on macOS (since 30 June 2026): Desktop agent that automates actions on the Mac beyond the chat window. New app connections (Canva, Dropbox, Instacart, OpenTable, Zillow Rentals) plus Google Tasks and Google Keep, plus support for your own MCP servers. Since July 2026 with Chrome integration: Spark can execute actions directly in the Chrome browser (tabs, forms, research). As of August 2026 beta, Google AI Ultra only, 18 and over, US. Download at gemini.google/mac.
• Gemini desktop app for Windows (since 10 September 2026, new): After the Mac app, the native desktop app is now also available for Windows 10 and newer (x64 and ARM64). Invoked with Alt + Space from any running application without leaving the context: ask a question, draft a message, share the screen, generate an image or video, start Deep Research. Connected are Gmail, Drive, Docs and Calendar. Download at gemini.google/desktop. Available in all countries and languages where the Gemini App is available; some functions require a Google AI subscription and a minimum age of 18. For DACH: The app itself is usable, but the agent it contains, Gemini Spark, remains blocked in the EEA, Switzerland and the United Kingdom.
• Voice dictation on macOS (since 29 July 2026): In the Gemini macOS app you can dictate in any active window by holding the Fn key, and Gemini turns speech into polished text directly at the cursor. Includes reasoning for screen-context tasks (rewrite text, analyse a file, generate an image). Initially in English for all users of the Gemini macOS app.
• Google Antigravity in the Gemini macOS app (since 13 October 2026, new): A coding agent plans, builds and tests prototypes in the background directly from the Mac app. Initially English only, for personal and eligible Workspace accounts aged 18 and over in all regions where Antigravity is available.
• Project Mariner (discontinued): Browser automation, discontinued on 4 May 2026. Functionality moved into Gemini Agent.

17.2 What is coming

Astra is Google's bet on the AI assistant of the future: continuously aware of context (camera, microphone, screen), proactive, with long-term memory. Rollout step by step over the coming quarters.

Project Genie (Genie 3): DeepMind's world model that generates interactive, walkable 3D environments from a prompt. Listed in the Gemini App as part of Google AI Ultra, but still experimental for everyday business. For most mid-market teams currently not a productive tool, but a preview of generative simulations.

17.3 Availability in DACH

• Gemini Live: Available in Germany, AI Pro/Ultra.
• Gemini Agent / computer use: Consumer agent (AI Ultra) as of June 2026 still US only, EU rollout announced. The native computer-use tool of 3.5 Flash, on the other hand, is usable via the Gemini API and the Enterprise Agent Platform.
• Project Astra (full vision): Research mode, not an end-user product.

Gemini in Google Workspace is managed centrally via two new surfaces: the "Generative AI" area in the Admin Console (since April 2026) and the AI Control Center (GA since 8 May 2026 in Enterprise Standard/Plus). Required role: Super Admin or delegated Services Admin.

18.1 AI Control Center and Workspace Intelligence (GA since 8 May 2026)

Central control room for all Workspace AI settings:

• Gemini Enterprise on/off: Per OU/group, a uniform service switch for all Gemini editions.
• Workspace Intelligence (GA since April 2026): can be switched off per data source (Gmail, Drive, Calendar, Chat, Meet). Default: on.
• Gem sharing, conversation sharing, conversation history: Switchable separately, history retention can be shortened per domain or deactivated entirely.
• Gemini App apps access (connectors): Gmail/Drive/Calendar/Docs/Sheets/Slides/Chat/Meet individually.
• Audio Overview, video gen, image gen: Sub-toggles under Workspace Intelligence or NotebookLM. Audio/Video Overviews in NotebookLM not available in in-country regions (CA/IN).
• NotebookLM Enterprise: Separate on/off.
• Gemini in Meet: Granular in the AI Control Center.
• Beta features ("Alpha features" until 22 July 2026): Opt-in toggle for functions in the early launch stage (see 18.2).
• Admin Assist (since 17 August 2026): Gemini moves into the Admin Console itself, with two building blocks. The side panel via the One Google Bar guides step by step through complex admin tasks and troubleshooting. Search Overviews answer questions in the admin search bar by bundling Help Center articles into a summary instead of just listing links. Only for Business Starter/Standard/Plus, on by default for Super Admins (not for delegated admins), no effect on end users.
• Audit logs for Gemini Notebook (rollout from 3 September 2026, new): The Admin Console now keeps complete audit logs for Gemini Notebook, i.e. who created, shared or added sources to which notebook and when. Available for all Workspace customers with access to the security investigation tool or the audit and investigation tool. The logs are there by default, but the BigQuery export must be switched on actively. Closes the last big governance gap for notebooks.
• Control external sharing of Gemini Notebook (rollout from 10 September 2026, new): Four levels from "off" to "on including public notebooks", per domain, OU or group. Default is off. Details in chapter 14.3.
• Switch Google Pics on and off (since 1 September 2026, new): The new image application is on by default and can be deactivated per domain, OU or group. Anyone who has deliberately restricted image generation so far should check this switch after the rollout (see chapter 11.3a).
• Chat usage metrics in the Gemini reports dashboard (since 20 August 2026): Gemini reporting now also shows Gemini usage in Google Chat, at organisation and user level. Business Starter/Standard/Plus, Enterprise Starter/Standard/Plus and Google AI Pro for Education, automatically active for eligible organisations. Handy for measuring adoption after a rollout, but clarify with the works council before introduction: user-level evaluation is subject to co-determination.

18.2 New compliance and consent controls (May 2026)

• Explicit consent requirement for "Take notes" in Meet (since 8 May 2026): Admins can enforce that all participants must actively consent before automatic notes, recording or transcript start. Relevant for DACH compliance (employee co-determination, GDPR Art. 6).
• Out-of-domain file warnings (since 26 May 2026): Badges for external files now appear consistently in Android/iOS apps, email notifications and service account access, no longer only in the web UI.
• Audit logs expanded (since 1 May 2026): New data source fields and expanded functions in the Admin Console for security investigations.
• DLP for Google Calendar (GA since 3 June 2026): Data loss prevention rules now also scan appointment titles, descriptions and locations and can warn, block or log. Extends DLP coverage beyond Drive, Chat and Gmail.
• "Take notes for me" for in-person meetings (since 13 August 2026): Automatic notes now also work without Google Meet, i.e. in the meeting room on site. Gemini captures the conversation and files notes, action items and transcript in a Google Doc. Included in Business Standard/Plus, Enterprise Standard/Plus, Frontline Plus, Google AI Pro/Ultra and Google AI Pro for Education. DACH note: This moves the recording question from the video call into the physical room. Anyone who wants to keep works council and GDPR clean should clarify the consent requirement (see above) and the consent of everyone present before using the feature at scale.
• "Take notes for me" on the Meet hardware touch controller (rollout from 31 August 2026): Until now, automatic notes in the meeting room only ran via Companion Mode on a laptop. In future you control them directly on the touch controller of the Meet hardware: a dedicated badge on the display shows whether Gemini is currently taking notes and can be toggled with a tap. Notes can be paused for confidential passages and resumed afterwards. Rollout: Early preview devices from 31 August 2026, Rapid and Scheduled Release from 8 September 2026, each gradually over up to 15 days. Included in Business Standard/Plus, Enterprise Standard/Plus, Frontline Plus, Google AI Plus/Pro/Ultra and Google AI Pro for Education; Meet hardware still needs its own licence. Admins must have activated "Take notes for me" for the organisation. DACH note: The pause function is the genuinely useful detail for works council and GDPR conversations, because the recording can now be interrupted visibly and traceably in the room.
• Visual screenshots in Meet notes (rollout since 3 August 2026): "Take notes for me" now automatically captures shared slides, diagrams and charts as screenshots in the meeting notes, so the visual context of a meeting is no longer lost. Admins define whether screenshots are always allowed or only during an active recording; individual users can switch them off in the notes panel. Included in Business Standard/Plus, Enterprise Standard/Plus and Google AI Pro for Education. In addition, Meet automatically organises notes, transcripts and recordings in Drive folders.
• New "Take notes for me" settings in Meet (since 16 July 2026): Until now admins could only switch automatic notes on or off. New is the option to activate them only for meetings with 3 or more participants, which exempts one-on-one conversations. Caution, the defaults differ: in Business Standard/Plus the function is on by default, in Enterprise and Education plans off by default. The rollout runs until 3 August 2026, so the setting takes effect without you doing anything. End users can override it themselves from 21 September 2026. Anyone on Business Standard/Plus who wants to keep co-determination or GDPR clean should check this before 3 August (interplay with the consent requirement above).
• "Gemini Alpha" is now called "Gemini Beta" (since 22 July 2026): Google has renamed the early launch stage for Workspace features. A pure name change, neither admin controls nor privacy nor prices change. The previous Alpha opt-in toggle in the AI Control Center is accordingly now called Beta. Affects all Business, Enterprise and Education add-on plans.
• Context-Aware Access for Gemini Enterprise (rollout from 8 September 2026, new): The existing Context-Aware Access policies now also apply to Gemini Enterprise: access only from managed devices, only from defined countries or only when device security requirements are met. Configuration per OU or group. Available in Enterprise Standard/Plus, Education Standard/Plus, Frontline Standard/Plus, Enterprise Essentials Plus and Cloud Identity Premium; the policies only take effect if Gemini Enterprise is licensed. Rollout completes on 15 September 2026. For DACH: This is the lever to tie AI access to the company laptop without blocking Gemini entirely.
• Gemini-based data classification in Drive (open beta since 28 August 2026, new): Gemini assigns classification labels to Drive files based on instructions that admins store in their own words. Training data is no longer needed for this. Set up in the Admin Console under Security, Access and data control, Data classification. File owners can review and correct an automatically set label, and every label action lands in the audit log. Available in Enterprise Plus, Google AI Pro for Education and Frontline Plus, rollout until 30 September 2026. Why this counts: Only with clean labels do DLP rules and retention periods work reliably, and only then can autonomous flows be prevented from reaching sensitive files.
• Control for temporary chats and conversation deletion (since 16 June 2026): Admins now control per domain, OU or group whether users may have temporary chats in the Gemini App and delete conversations themselves. Default: on. Relevant for retention and eDiscovery requirements (interplay with Vault, see chapter 19.4).
• MCP connectors on by default (since 15 September 2026, new): The new Gemini connectors to Asana, HubSpot, Salesforce and other services (chapter 16.3) are active out of the box, which means Workspace content can flow to third parties. Recommendation for DACH: review them under Apps > Google Workspace Marketplace apps > Apps list and only enable them selectively per OU or group.
• "Quick notes" in "Take notes for me" (since 22 September 2026, new): Meet notes now open by default on the "Quick notes" tab, a one-page summary with outcomes and action items; the detailed version is under "Full notes". Since 29 September 2026 the "only for meetings with three or more people" option also applies to end users, with different defaults: on in Business Standard/Plus, off in Enterprise Standard/Plus, Frontline Plus and AI Pro for Education. In addition, the Drive folder "Meet Recordings" is being renamed "Legacy Meet Recordings" and moved into a new "Google Meet" folder structure.
• New permissions and limits in Gemini Enterprise (September 2026, new): Since 28 September granular IAM permissions per app and data store (the roles Gemini Enterprise Admin, Viewer and User can be assigned individually), since 21 September admins can transfer ownership of shared agents (the new owner has to re-enable deactivated schedules and triggers). Breaking change since 9 September: projects with VPC Service Controls can no longer add website URLs as a source. Agent Designer is now called "Workflow Builder" (GA), "Projects" is GA and must be enabled via a toggle. The Microsoft connectors (Outlook, OneDrive, SharePoint, Teams) are getting write actions in preview; since 29 September Outlook can create, reply to and forward emails. For governance this means agents now also have write access to M365.
• Gemini 4 Argon in Gemini Enterprise: switch flips to on on 27 October 2026 (new): Since 13 October 2026, Argon has been GA in the web app, mobile app and AI developer tools, and since 14 October also in Workflow Builder and workflow agents. The admin switch is initially off and will be enabled automatically on 27 October. If you want to roll out Argon deliberately or block it for now, decide under "Model availability" before then. In Antigravity, Argon is immediately part of the default selection; admins have to add it to custom model lists themselves. Also new: a separate switch for Gemini 2.5 Pro in the web app (since 1 October, on by default), admin controls for Antigravity Boost and Teamwork (since 6 October) and workflow connectors to Microsoft Teams, Notion and Salesforce (since 13 October), which means additional channels through which data can leave.
• "Routines for this meeting" after Meet calls on by default (rolling out from 5 October 2026, new): After meetings with "Take notes for me", Gemini suggests follow-up emails and action items in Google Tasks (details in chapter 15.3). Can be turned off per domain, OU or group via the Workspace Studio settings or the take-notes setting. Include it in works council agreements on Meet notes.
• Skills in the Gemini App without an admin switch: There are no admin controls for Skills in the Gemini App (rolling out since 13 October 2026), only for the postponed Skills in the Workspace apps (chapter 7).

18.3 Most important admin articles

• Turn Gemini app on/off
• Manage feature access
• Turn NotebookLM on/off
• Manage Gemini conversation history
• Adoption phases 1 to 3 (Microsoft-like rollout framework)

18.4 Rollout strategy (mid-market top 3)

1. AI Control Center first: Deactivate data sharing with Google for model training org-wide (default is already off, but verify).
2. Set conversation history: For DACH compliance requirements, 0 or 6 months as the default.
3. Pilot with power users: 10 to 50 users from marketing/sales, then roll out in stages.

Gemini for Workspace and Gemini Enterprise meet the important compliance standards. Data residency configurable, training opt-out contractual, audit logs available.

19.1 Certificates and frameworks

• SOC 1, SOC 2, SOC 3
• ISO 9001, ISO 27001, ISO 27017, ISO 27018, ISO 27701
• ISO 42001 (AI Management Systems), covers Google Cloud, Workspace and the Gemini App since 2026
• BSI C5 (relevant for DACH)
• FedRAMP High, CJIS
• HIPAA-ready
• GDPR-compliant Cloud Data Processing Addendum (CDPA)

19.2 Data residency

• Multi-region EU and US: Core and ML processing fully supported for Gemini Enterprise (access-controlled data, blended/Workspace/people search, assistant, answer generation).
• In-country regions in Gemini Enterprise: GA with allowlist only for Canada, India, Japan, Singapore and the UK. Correction: Germany is not on the list; for DACH the EU multi-region remains the option. The planned Sovereign Cloud Germany is operated by a German company owned by Thales (preview since 20 May 2026, GA planned for the end of 2026, with S3NS in France as the failover partner); Gemini has not yet been committed there.
• Data Regions support for the Gemini App (since 29 June 2026): The Gemini App now respects the organisation's data regionalisation requirements (storage and processing in the chosen region). Closes a gap for regulated DACH customers who could previously only regionalise Enterprise and Vertex paths.
• global (default): unrestricted, for free Workspaces.
• CMEK: only in US/EU multi-region.
• Important restrictions (as of 15 October 2026): In the EU multi-region (EU member states only, so excluding the UK and Switzerland), Gemini 4 Argon, 3.6 to 3.8 Flash, 3.5 Flash-Lite, 3.8 Live and 2.5 Pro run with data storage (DRZ) and ML processing (MLP) in the EU. Gemini 4 Argon has been GA in Gemini Enterprise since 13 October 2026 (regions global, us, eu); the admin switch flips to on automatically on 27 October 2026 (see chapter 18.2). 3.8 Flash is enabled by default in global, us and eu and, since 14 October 2026, also has in-country storage and processing in the UK region (europe-west2); 3.6 and 3.7 Flash remain on global there. Still not available in the EU: Gemini 3.1 Pro ("Limited Availability") and grounding with Google Search, which only runs globally. An EU-compatible alternative is Web Grounding for Enterprise with DRZ and MLP in the US and EU, where it is on by default. Images: 3.1 Flash Image (Nano Banana 2) has EU residency, 3 Pro Image (Nano Banana Pro) only runs globally. Single-region Germany (europe-west3) only with 3.5 and 2.5 Flash. Video generation not supported in EU DRZ. NotebookLM Enterprise: no MLP, DRZ only US/EU, Audio/Video Overviews not in CA/IN. Third-party models: Since 8 October 2026, Claude Opus 5.5 and Sonnet 5.5 can be enabled in the AI developer tools of Gemini Enterprise (off by default, only global, us and eu, billed by usage against the project budget).
• Assured Controls Plus: Add-on for public authorities/regulated industries (access approvals, CSE, local data storage). Since 12 October 2026, Access Management and Access Approvals can be controlled via classification labels on Drive files, and labels can also be applied automatically via DLP (Enterprise Plus or Frontline Plus with Assured Controls or Assured Controls Plus). This means the labels from data classification (chapter 18.2) now also determine who may access a file.
• Simplified CSE setup (since 1 October 2026, new): Client-side encryption can now be set up in minutes with Cloud HSM keys and Google Identity; the route via an external key management service remains available. Enterprise Plus with Assured Controls or Assured Controls Plus only. According to Google, CSE protects content from Google and Gemini as well, so Gemini cannot read encrypted files.
• Post-quantum cryptography in Meet (since 12 October 2026): Meet connections on the web and on Meet hardware are encrypted with quantum-safe cryptography, with mobile devices to follow. No admin action required.
• Data regions for Google Apps Script (GA since 16 September 2026, new): Storage and execution of Apps Script projects stay in the selected region. Available in Enterprise Plus, Education Standard/Plus and Frontline Plus. Relevant because custom steps in Workspace Studio (chapter 15.3) can run on Apps Script.

19.3 Training opt-out

In Gemini for Workspace Business and Enterprise, the use of inputs for model training is contractually excluded. In the personal version (Free, AI Plus, Pro, Ultra) it is on by default and can be switched off.

19.4 Workspace Vault

Audit, retention, eDiscovery for Gemini conversations. Availability differs per Workspace plan, Enterprise has the full range of functions. New since 11 June 2026: Vault now explicitly supports retention rules and litigation holds for the Gemini App (web and mobile), so that pure Gemini App conversations are also retained in a legally secure way and captured for eDiscovery.

Anyone using Gemini in DACH should know five points: EU AI Act, GDPR, EU hosting, Connected Apps availability and language.

20.1 EU AI Act Article 4

• Proof of training mandatory for all Gemini users.
• Workspace admin reporting provides hour and activity logs.
• Logs must be available in audits.

20.2 GDPR and Gemini

• Workspace Business and above: Data Processing Amendment as standard.
• Personal plans: No DPA, no training guarantee. Unsuitable for company data.
• Right to be forgotten: Via Workspace admin or privacy form.
• DPIA: Mandatory for critical use cases.

20.3 EU hosting

Correction: For Workspace you do not choose a cloud region but Data Regions (EU or US) in the Admin console. On Google Cloud (Vertex AI, Gemini Enterprise), the EU multi-region is the route for the new models; the Frankfurt region is called europe-west3 (not europe-west12, which is Turin) and only offers older Flash models (see chapter 19.2).

20.4 Availability in Germany

• Gemini App: Fully available in Germany.
• NotebookLM: Fully available.
• Connected Apps for Personalization: Personal accounts since Q1/2026, Workspace since Q4/2025.
• Gemini Agent/computer use: As of May 2026 US only, EU rollout announced.
• Veo Premium: AI Ultra, available in Germany.
• As of October 2026, what is still missing in Germany: Gemini Spark including the new Chrome and Photos connection from 9 September (US only), notebooks in Gemini and Study Notebooks for Workspace tenants in the EEA (rollout to follow), AI Overviews in Gmail search for personal accounts, the Student Hub for Workspace Individual customers in the EEA and most of the new Connected Apps (since 13 October 2026 more than 40 in the "Customise" menu, mostly US and English only). Conversely, according to trade press, Gemini has fully replaced Google Assistant on Android in Europe too since late September 2026; the option to switch back is gone.

20.5 EU AI Act: status after the Digital Omnibus

The GPAI obligations of the EU AI Act have applied since 2 August 2025, and Google has signed the GPAI Code of Practice. With the Digital Omnibus (in force since 27 July 2026), the high-risk obligations have been postponed: for systems under Annex III to 2 December 2027, for Annex I to 2 August 2028. Typical everyday office use of Gemini is hardly affected. The DPIA template for Workspace with Gemini has been published ("Supporting Your DPIA for Workspace with Gemini") and is a mandatory artefact for DACH data protection officers.

The postponement is no reason to wait: anyone who has not yet set up the Article 4 training records and the DPIA should catch up now rather than keep planning.

20.6 Language

• German is supported on a par with English.
• Voice (Gemini Live) in German sounds acoustically very good.
• Tip: Store "Always answer in German" in the custom instructions, otherwise Gemini may switch to English on multilingual topics.

20.7 What DACH compliance should check

1. Workspace Business or Enterprise (no personal account for business data).
2. Data Regions set to EU multi-region or in-country DE (allowlist).
3. AI Control Center: set the Workspace Intelligence toggles deliberately.
4. Verify the training opt-out (default is already OFF in Workspace).
5. Conversation history set to the desired retention.
6. EU AI Act training records (Art. 4) plus DPIA: Art. 4 has applied since 2 February 2025, catch up immediately if in doubt.
7. Deliberate decision on model choice in the EU: Gemini 4 Argon, 3.6 to 3.8 Flash and 2.5 Pro via EU multi-region, 3.1 Pro without EU residency, strict single-region Germany (europe-west3) only 3.5 and 2.5 Flash.
8. BSI C5 as an additional DACH compliance anchor.
9. NotebookLM Plus for power users.