Tools
Gemini for Lawyers: Workspace Integration, Gemini Notebook and Data Caveats
Gemini for lawyers as of September 2026. Why Workspace is the only safe door, what Gemini 3.1 Pro and Gemini Notebook do well, and where Google's terms bite.
On this page
Gemini is a sound choice for a law firm that already runs on Google Workspace, and a poor one for anybody using the consumer app, whose privacy page asks users not to enter "confidential information that you wouldn't want a reviewer to see". As of September 2026 the flagship model is Gemini 3.1 Pro, the case-file tool is Gemini Notebook, and the enterprise layer has EU data residency. The door you enter through decides everything.
Two doors, and only one of them is for client work
Google sells Gemini through two channels with opposite data terms, and the Gemini Apps Privacy Hub describes the consumer channel with a candour most vendors avoid. Activity is kept for 18 months by default. A subset of conversations is read by human reviewers. Reviewed chats are retained for up to three years even after you delete them. Conversations are used to train models unless "Keep Activity" is switched off at myactivity.google.com, and even with it off, chats are kept for 72 hours. Ars Technica called the resulting menu structure a privacy maze in April 2026.
The personal subscriptions make this worse for firms, not better. Google's AI plans page states that Google AI Plus, Pro and Ultra are "available only for personal accounts". A partner paying for Google AI Ultra is still on consumer terms.
The other door is Google Workspace with the Gemini add-ons, and Gemini Enterprise on Google Cloud. There, according to the same Privacy Hub, Workspace and Vertex AI are excluded from training by default via the admin console. The confidentiality checklist sets these terms next to the other vendors; for Gemini the comparison is not between Google and its rivals but between Google's two channels.
| Consumer Gemini app (incl. AI Pro/Ultra) | Workspace with Gemini / Gemini Enterprise | |
|---|---|---|
| Training on your content | Yes, unless Keep Activity is off | No, by default |
| Human review of chats | Yes, a subset | Governed by enterprise terms |
| Retention | 18 months default; reviewed chats up to 3 years; 72 hours with activity off | Admin-controlled |
| EU data residency | No commitment | NotebookLM Enterprise from 3 Jun 2026; UK, Japan, India, Singapore from 30 Jun 2026 |
| Suitable for client matters | No | Yes, after your own risk assessment |
The models, and the one that has not arrived
Gemini 3.1 Pro, released 19 February 2026, remains the flagship as of September 2026. Google announced Gemini 3.5 Pro at I/O on 19 May 2026 for "next month"; as of late August it had still not shipped, while the Flash line moved through 3.5, 3.6, 3.7 and, on 2 September 2026, 3.8 Flash, per the Gemini model history and Coders Era's launch tracker. For a firm this matters in one way: build your prompts and workflows on 3.1 Pro and treat 3.5 Pro as a rumour until it is in your admin console.
| Model | Role | Context | API price (per million tokens in / out) |
|---|---|---|---|
| Gemini 3.1 Pro | Flagship reasoning model | 1M input | $2 / $12 (higher above 200K tokens) |
| Gemini 3.8 Flash | Everyday model, fast agent loops | 1M input | Cheaper than Pro; check the current rate card |
| Gemini Deep Research | Agentic research model, multi-step, cited reports | n/a | Preview endpoint |
The 1-million-token context window is real and useful: it is a full data room of leases in one request, and Gemini Deep Research typically reads 30 to 150 sources in five to fifteen minutes, the fastest of the major modes according to Presenc's May 2026 comparison.
Where Gemini beats Claude and ChatGPT
Price. Gemini 3.1 Pro at $2 in and $12 out is a fifth of Claude Fable 5.1 ($10/$50) and GPT-6 Astra ($10/$50) on the API. For bulk classification of a document set, the difference is a budget line.
Workspace. If the firm's documents already live in Drive, Gemini in Docs, Sheets and Gmail sees them under the existing permission model without a connector project. Freshfields, which signed with Anthropic in April 2026, has kept a separate Google Cloud and Gemini relationship since April 2025 and expanded it in 2026, which tells you the two are not mutually exclusive at scale.
Breadth of reading. In the deep-research comparisons Gemini consistently reads the most sources per query; for a regulatory sweep across a dozen national websites, breadth is the point. Deep research tools for lawyers has the head-to-head.
Academic legal knowledge. On the academic LegalBench suite (16 September 2026) Gemini 3.1 Pro scored 87.40 percent, within about a point of Claude Fable 5.1 at 88.51 percent, per Vals.ai's benchmark page.
Where it trails
The same benchmark pages show the other side. On Harvey's LAB, the long-horizon legal agent benchmark where every rubric criterion must pass, the 18 September 2026 leaderboard on LLM Stats puts Gemini 3.1 Pro at 0.0 percent against 13.3 percent for Claude Fable 5 and 11.7 percent for Opus 5; Gemini 3.5 Flash managed 0.8 percent in Harvey's initial run. On GC AI's vendor-run In-House Legal Bench (May 2026) Gemini 3.1 Pro scored 57.5 percent, last of the four models tested. Gemini is not on the podium of the Vals Index or the Vals Legal Research Bench.
The pattern is consistent: Gemini is a strong reader and summariser and a weak multi-step legal agent. Lacey v. State Farm (C.D. Cal., 6 May 2025), the $31,100 sanction against K&L Gates and Ellis George, involved Google Gemini alongside CoCounsel and Westlaw Precision among the tools used; the benchmarks guide explains why a high LegalBench score and a fabricated citation are not contradictory.
Gemini Notebook for a case file
NotebookLM became Gemini Notebook on 16 July 2026, with 30 million-plus users and a "secure cloud computer" per notebook that writes and runs code, per 9to5Google. It answers only from what you upload and cites the passage, which is the right shape for a bundle, a set of depositions or a statute pack. Source caps are 50 per notebook free and 300 paid, and it cannot search the open web. The full workflow, limits and data terms are in the Gemini Notebook review.
Gemini Enterprise: workflows, skills and residency
Gemini Enterprise is Google's answer to Claude for Legal and ChatGPT Work, and its release notes show the pace: Skills GA on 17 June 2026, workflow agents GA on 18 June, an Agent Registry with egress allow and deny lists on 25 June, pay-as-you-go on 1 August, A2A agent registration GA on 17 August and Workflow Builder GA on 3 September 2026. Google demonstrated a Gemini Enterprise for Legal configuration on 17 September 2026, the day OpenAI launched Astra for Law; treat it as new. For firms in Austria, Germany and the rest of the EU, the 3 June 2026 EU data-residency note for NotebookLM Enterprise is the line to quote in the vendor questionnaire, alongside the caveat in GDPR and AI tools for lawyers: a residency toggle changes where data sits, not which government can compel its production.
A Workspace setup for a firm
Before any lawyer types a client name into Gemini, the admin runs this list.
- Confirm every user is on a Workspace account with the Gemini add-on, and that no personal Google AI Pro or Ultra subscriptions are being expensed.
- In the admin console, verify Gemini for Workspace is excluded from training and set retention to match the firm's policy.
- Turn off Keep Activity on any consumer Gemini account that will exist on firm devices, and tell people why: 18 months, human review, three years. The screenshots are in AI privacy settings step by step.
- Create one Gemini Notebook per matter and one shared Drive per matter; keep drafts in Docs, not in the chat.
- Set Deep Research as the default for regulatory sweeps and press checks, never for case-law research; verify authorities in Westlaw, Lexis or the court record.
- For Gemini Enterprise, register agents in the Agent Registry with egress deny lists before granting any tool access.
- Decide, in writing, which tasks stay with Claude or ChatGPT: long agentic drafting and multi-step diligence are where Gemini's benchmark results are weakest.
The standard Gemini-in-Docs instruction for a first draft, which stops the model inventing authority, reads:
Draft a first version of the client update described below in British
English, plain language, under 400 words. Use only the facts in the
attached Doc and the attached engagement letter. Do not cite cases,
statutes or regulations; where a legal source is needed write
[SOURCE NEEDED]. Put every assumption in a bullet list at the end.
Module 3 of the AI Lab for Lawyers course covers multimodal and file-handling work, where Gemini's long context and Workspace integration are demonstrated against Claude and ChatGPT on the same documents. If your firm is choosing between the two ecosystems rather than adding one, Claude vs Gemini for legal work is the direct comparison.
Frequently asked questions
Is Google Gemini safe for confidential legal work?
Only through a Google Workspace or Gemini Enterprise account. The consumer Gemini app keeps activity for 18 months by default, lets a subset of chats be read by human reviewers, retains reviewed chats for up to three years and uses conversations for training unless Keep Activity is off. Workspace and Vertex are excluded from training by default via the admin console.
What is the difference between Gemini and Gemini Notebook?
Gemini is the general assistant, inside the consumer app, Workspace apps and Gemini Enterprise. Gemini Notebook is the July 2026 name for NotebookLM, a source-grounded tool that answers only from documents you upload, with passage citations, and now runs a secure cloud computer for each notebook. For a case file, use the Notebook; for drafting and email, use Gemini in Docs and Gmail.
How does Gemini compare with Claude and ChatGPT for lawyers?
Gemini 3.1 Pro is cheaper at $2 and $12 per million tokens, reads the most sources in Deep Research and sits inside Workspace. Claude Fable 5.1 and Opus 5 lead the Vals Index and Harvey's LAB, where Gemini 3.1 Pro scored 0.0 percent on the all-pass metric. For long agentic legal tasks pick Claude; for Workspace-native summarising and drafting Gemini is competitive.
Does Gemini Enterprise offer EU data residency?
Google's release notes record EU data residency and ML-processing compliance for NotebookLM Enterprise from 3 June 2026, with UK, Japan, India and Singapore residency from 30 June 2026. Confirm in writing which Gemini Enterprise features are covered for your tenant, and remember a residency toggle changes where data sits, not which government can compel its production.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Tools
Gemini Notebook (NotebookLM) for Lawyers: A Grounded Case-File Assistant
Gemini Notebook (ex-NotebookLM) for lawyers as of September 2026. How source-grounded answers work on a case file, three workflows, source caps and data terms.
6 min read
Comparisons
Claude vs Gemini for Legal Work: Context, Workspace and Data Terms
Claude vs Gemini for legal work: close on benchmarks, far apart on data terms. Workspace vs Microsoft fit, Gemini Notebook vs Claude Projects, EU residency.
9 min read
Guides
Deep Research Tools for Lawyers: Claude, ChatGPT, Gemini and Perplexity Compared
Deep research AI for lawyers compared: what Claude, ChatGPT, Gemini and Perplexity research modes do, where they fail on law, and a verification table.
10 min read
Guides
Which AI Tools Train on Your Data? A Confidentiality Checklist for Lawyers
Which AI tools train on your data, plan by plan as of September 2026: ChatGPT, Claude, Gemini, Copilot, legal vendors, plus bar rules and a 10-point checklist.
11 min read
Regulation & Ethics
GDPR and AI Tools for Lawyers: Lawful Basis, DPIAs and Vendor Terms
How GDPR applies when law firms use Claude, ChatGPT or Copilot: controller and processor roles, lawful basis, DPIA triggers, transfers and vendor terms.
11 min read
Playbooks
AI Privacy Settings for Lawyers: Step-by-Step in ChatGPT, Claude, Gemini, Copilot and Perplexity
Exact steps to turn off training and check retention in ChatGPT, Claude, Gemini, Copilot and Perplexity as of September 2026, plus admin settings and a log.
10 min read