Lawyers keep asking us whether the firm should use Claude, ChatGPT or Gemini. Honestly, the gap between them on everyday drafting is smaller than the gap between a properly set up business plan and a carelessly used free account. Below: what each vendor officially publishes, a short decision guide, and a six-prompt test kit for your own redacted work.
For a small Canadian firm, the right AI is usually the one that fits your existing stack: Gemini for Google Workspace firms, ChatGPT or Claude for Microsoft 365 firms. Use a business plan, never a consumer tier, for client work. Then run the six-prompt test below on a redacted file and let your own scores decide.
Rule one: business plans only
Each vendor sells consumer tiers for individuals and business tiers with contractual data commitments and admin controls. For a law firm, that difference matters more than which model writes nicer paragraphs.
- ChatGPT: OpenAI's pricing page lists "opt-out available" for model training on the Free, Go, Plus and Pro plans, which means training is on unless the user turns it off.
- Claude: Anthropic's pricing page lists model training as "Opt-out" for Free, Pro and Max.
- Gemini: Google's consumer Gemini Apps privacy notice says a subset of chats is reviewed by human reviewers, and tells users not to enter confidential data if the Keep Activity setting is on.
The Law Society of Ontario's practice note uses exactly this scenario as its warning: uploading an agreement for proofreading into a public AI tool can expose the content. Its advice is to read the terms of use and not to input confidential or privileged information into a system without proper confidentiality, security and retention safeguards.
Never paste privileged facts into consumer tiers. Not a personal ChatGPT Plus account, not a free Claude login, not a personal Gmail Gemini chat. If a lawyer or assistant is already doing this, fix the account before you fix the prompts.
Claude vs ChatGPT vs Gemini: side-by-side for law firms
Everything below comes from the vendors' own pages, checked on October 9, 2026. Where a vendor doesn't document something, we say so rather than guess. Plans and prices change often, so confirm before you buy.
| What matters | ChatGPT (OpenAI) | Claude (Anthropic) | Gemini (Google) |
|---|---|---|---|
| Business plans | ChatGPT Business (from 2 users) and ChatGPT Enterprise | Claude Team (2 to 150 people) and Claude Enterprise | Gemini in Workspace apps is included in Google Workspace Business Standard, Business Plus and Enterprise; Business Starter gets limited access |
| Inputs used for training by default? | No. OpenAI says it does not train on Business or Enterprise data, including inputs and outputs, by default | No. Anthropic lists "None by default" for Team and Enterprise | No. Google says Workspace does not use customer data, including prompts, to train models without the customer's permission |
| Canadian data residency (as documented) | Canada is listed for storage at rest, for eligible new ChatGPT Enterprise and Edu customers. In-region inference is listed only for the US, Europe and the UAE. Business is not listed as eligible | Anthropic lists Canada for regional data residency through AWS Bedrock and Google Cloud Vertex AI, which are developer and cloud routes. We found no Canadian residency option documented for the Claude Team app | Workspace data regions offer United States, Europe or No preference. Canada is not an option in the Workspace documentation we checked |
| File upload and long documents | Supports file uploads and chatting with PDFs. Limits vary by plan and model; no single Business figure is shown on the pricing page | Supports uploads; the pricing page lists a context window of up to 1M tokens for Team and Enterprise, varying by model | Works on files where they live: Gemini in Drive can find, summarize and analyze files, and Gemini in Docs drafts and edits in place |
| Microsoft 365 fit | Extensions for Word, Excel and PowerPoint, and ChatGPT in Microsoft Teams | Claude for Microsoft 365 and Claude for Microsoft Outlook are included on Team and Enterprise | Built for Google Workspace; no Microsoft 365 add-in documented on the pages we checked |
| Google Workspace fit | Google Sheets extension listed on the pricing page | Gmail, Google Calendar and Google Drive connectors; an owner must enable them on Team and Enterprise | Native in Gmail, Docs, Sheets, Drive, Meet and Chat on Business Standard and up |
| Admin controls (SSO, retention) | Business: SAML SSO, MFA, roles, GPT controls. Enterprise adds SCIM and role-based access. Retention controls for qualifying organizations | Team: SSO, domain verification, central admin, usage analytics. Enterprise adds SCIM, audit logs, role-based access and custom data retention | Managed in the Google Admin console. Admins set Gemini app chat retention (up to 36 months) and manage retention of inserted content with Vault |
| Published business price (checked Oct 9, 2026) | Not captured: the per-seat Business price did not appear in the pricing page text we could verify. Check chatgpt.com/pricing | Team Standard: $20 per seat per month billed annually, or $25 monthly. Enterprise: US$20 per seat per month plus usage at API rates, billed annually | Gemini comes bundled with Workspace plans rather than as a separate seat. Workspace per-user prices did not appear in the page text we could verify |
Two takeaways. No vendor documents Canadian data residency for its small-business tier, so if that's a hard requirement you're looking at enterprise contracts or cloud deployments. And every business tier makes the same no-training-by-default promise, so privacy won't separate them. Workflow fit will.
Which to pick if...
Your firm runs on Google Workspace
Start with Gemini. It already sits inside Gmail, Docs and Drive, it's covered by your existing Workspace agreement, and Google states the same no-training commitment there. One catch: Google's privacy hub notes that Workspace data region settings do not apply to data in Gemini Notebook, so set a firm policy on what goes there.
Your firm runs on Microsoft 365
Shortlist Claude and ChatGPT. Both now document Microsoft integrations: Claude for Microsoft 365 and Outlook, and ChatGPT's Word, Excel and PowerPoint extensions plus Teams. Pick by running the test kit below on your own documents, not by brand.
You do heavy document review
Claude's pricing page lists a context window of up to 1M tokens on Team and Enterprise, which helps with long agreements. That's capacity, not accuracy. For discovery-scale review, a legal-specific platform may fit better. Prompts 1 and 2 below show quickly whether a tool invents clauses.
You mostly draft client emails
Pick the tool that lives where you write. Gemini works inside Gmail; Claude for Outlook covers Microsoft mail. ChatGPT drafts well in its own window, so check its current connectors for your mail system. Prompt 3 tests tone and whether the tool sneaks in advice or promises.
You work in French and English
None of the three vendor pages we checked makes a specific claim about Canadian legal French. Don't assume. Run prompt 4 with a real bilingual paragraph and have a francophone lawyer do the scoring. If French legal research is central to your practice, note that LexisNexis launched Lexis+ AI in French for Canadian users in 2024.
Same task, three tools: the test kit
Benchmarks show how a model did on someone else's work. This kit shows how it does on yours, and gives you a score you can defend to your partners.
How to run it
- Use business accounts only, set up as trials if needed.
- Pick one real but fully redacted file: replace names with labels like Party A, remove addresses, account numbers and anything that identifies the client.
- Paste the exact same prompt into a fresh chat in each tool. No follow-ups on the first pass.
- Time each run from paste to a version you'd actually send or file.
- Have a second lawyer score the outputs without knowing which tool wrote which.
Prompt 1: Summarize a redacted agreement
Prompt 2: Extract dates into a table
Prompt 3: Plain-language client update
Prompt 4: French translation check
Prompt 5: Checklist from a practice direction
Prompt 6: Cite-check reminder
Prompt 6 is a deliberate trap. A good tool builds your verification list. A weaker one starts confirming cases it hasn't read.
Scoring rubric
Score each prompt 1 to 5 on each criterion. Six prompts times four criteria gives each tool a maximum of 120.
| Criterion | A 5 looks like | A 1 looks like | Claude | ChatGPT | Gemini |
|---|---|---|---|---|---|
| Accuracy | Every fact traces back to the source document | An invented date, clause, quote or case | __ / 5 | __ / 5 | __ / 5 |
| Follows instructions | Word limit, headings and columns respected; "Not stated" used when needed | Ignores the format or fills gaps with guesses | __ / 5 | __ / 5 | __ / 5 |
| Tone | Sounds like your firm; Canadian spelling; no promises | Robotic, American spelling or overconfident | __ / 5 | __ / 5 | __ / 5 |
| Time saved | Usable draft in a fraction of your normal time, fixes included | Fixing it took longer than writing it yourself | __ / 5 | __ / 5 | __ / 5 |
Accuracy is a gate, not a score. If a tool invents a date, a clause or a case on any prompt, mark that prompt zero across the board. A fast wrong answer costs more than a slow right one.
Your professional duties don't change
In April 2024 the Law Society of Ontario published a white paper, Licensee use of generative artificial intelligence, which encourages licensees to experiment with these tools while understanding the risks. Its companion practice note, Generative AI: Your professional obligations (dated April 10, 2024), sets out six duties that apply when you use AI:
1. Competence. 2. Confidentiality. 3. Honesty and candour. 4. Supervision. 5. Reasonable fees and disbursements. 6. Not misleading the tribunal.
The practical tips are clear: read the tool's terms of use, redact identifying details, don't input confidential or privileged information where safeguards are lacking, get informed consent if anonymizing isn't enough, and independently verify anything you intend to rely on. The note also flags the Federal Court's December 2023 notice requiring parties to disclose AI-generated content in documents filed with the court.
The LSO doesn't name a preferred vendor. Your duty is the same whichever logo is on the screen.
What about legal-specific tools?
General assistants aren't your only option. Several legal platforms build AI into tools lawyers already use, for example Clio Duo inside Clio's practice management software, Lexis+ AI from LexisNexis (generally available to Canadian customers since July 2024), and CoCounsel from Thomson Reuters (launched in Canada in February 2024). We mention them only so your shortlist is complete. We haven't tested them and this isn't an endorsement. The same six prompts work on them too.
If you want to go deeper on day-to-day ChatGPT use, read our guide to ChatGPT for lawyers in Canada. For the front-office side, where AI answers calls and books consults, see AI for law firms and our walkthrough of law firm intake automation in Ontario.
Claude vs ChatGPT vs Gemini for law firms: FAQ
Is Claude safe for legal work?
On the Team and Enterprise plans, Anthropic says it does not train on your content by default, and Enterprise adds audit logs and custom data retention. Safe use still depends on you: redact client identifiers, follow your law society's guidance and verify every output.
Is ChatGPT or Claude better for lawyers?
Neither wins across the board. Both offer business plans with no training on your data by default and both document Microsoft 365 integrations. Run the same redacted test prompts in each and score accuracy, instruction-following, tone and time saved.
Can any of these tools keep firm data in Canada?
OpenAI lists Canada for data storage at rest for eligible new ChatGPT Enterprise customers. Anthropic lists Canada through AWS Bedrock and Google Cloud Vertex AI. Google Workspace data regions offer only the US, Europe or no preference.
Can I use free ChatGPT for client work?
Not with client information. OpenAI's pricing page shows consumer plans may use content for training unless you opt out. The LSO says not to input confidential or privileged information into tools without proper confidentiality, security and retention safeguards.
Which AI is best for a law firm on Google Workspace?
Gemini is the natural first trial because it is built into Gmail, Docs and Drive on Business Standard and up, and Google says Workspace data isn't used to train models without permission. Test it against Claude or ChatGPT before committing.
Sources
All pages checked October 9, 2026. Quotes are paraphrased; read the originals before relying on them.
- OpenAI: Business data privacy, security and compliance
- OpenAI Help Center: Data residency and inference residency for ChatGPT
- ChatGPT pricing
- ChatGPT Business
- ChatGPT in Slack and Microsoft Teams
- Claude pricing
- Claude regional compliance
- Claude Help Center: Use Google Workspace connectors
- Google Workspace: Generative AI in Google Workspace Privacy Hub
- Google Workspace Admin Help: Choose a geographic location for your data
- Google Workspace pricing
- Gemini Apps Privacy Hub
- Law Society of Ontario: Licensee use of generative artificial intelligence (white paper, April 2024)
- Law Society of Ontario: Generative AI: Your professional obligations (practice note, April 10, 2024)
- Clio: Clio Duo announcement
- LexisNexis: Lexis+ AI launch in Canada and French-language expansion
- Thomson Reuters: CoCounsel Core launch in Canada and Australia
General information only, not legal or procurement advice. Vendor plans, prices and terms change; confirm current terms and your law society's guidance before you adopt any tool.
Want help running the test on your own files?
We'll set up the trial accounts, redact a sample with you and score the three tools side by side. Free, founder-led, 30 minutes.
Get the free audit →