Can Canadian Lawyers Use ChatGPT for Legal Research? (2026 Rules)
September 28, 2026 · 10 min read · Casescout Team
The short answer is yes. No Canadian law society has banned lawyers from using ChatGPT, Claude, Gemini, or any other generative AI tool. But every regulator that has spoken on the subject says the same thing in different words: the tool is yours to use, and every consequence of using it is yours to own. Several courts have gone further and now require you to tell them when AI helped prepare what you filed.
This guide collects what Canadian law societies and courts actually require, jurisdiction by jurisdiction, then walks through the Canadian cases where lawyers filed AI-invented authorities and what happened to them. It ends with a practical checklist you can apply to any AI tool before its output goes anywhere near a client or a court.
This article is general information, not legal or ethical advice. Guidance changes; confirm current requirements with your own law society and the court you are appearing before.
The one rule every regulator agrees on
The Law Society of Alberta's Gen AI Rules of Engagement for Canadian Lawyers, a synthesis of guidance from Canadian courts, law societies, insurers, and the Canadian Judicial Council, reduces it to one principle: lawyers are responsible for the truth and accuracy of their work. Everything else follows from that. AI does not create new duties so much as put pressure on old ones: competence, confidentiality, supervision, candour to the court, and fair billing.
That framing matters because it answers the question most lawyers are really asking. "Am I allowed to use ChatGPT?" is the wrong question. The right one is "can I still meet my existing obligations if I use it this way?" For drafting a client letter from your own notes, usually yes. For finding case law to cite in a factum, a raw chatbot makes that very hard, for reasons we cover below.
What each law society says
Ontario (Law Society of Ontario)
The LSO published a white paper on licensee use of generative AI in April 2024, with a quick-start checklist, best-practice tips, and a summary of professional obligations. The key points:
- Competence includes understanding a tool's capabilities and limitations, specifically including hallucinations.
- Supervision: the LSO treats using generative AI as akin to receiving help from a non-licensee employee. You would not file a summer student's memo unchecked; the same applies here.
- Verification: licensees should build a process of human verification into any AI-assisted work.
- Confidentiality: additional protocols are expected to keep client information from being disclosed through prompts.
- Billing: fees must remain fair, reasonable, and properly disclosed. Billing hours that AI saved you is not.
Ontario also changed the Rules of Civil Procedure. Since December 1, 2024, a factum must include a certification that the lawyer is satisfied as to the authenticity of every authority cited. That is a personal certification. "The software found it" is not a defence to a false one.
Alberta (Law Society of Alberta)
Alberta publishes a Generative AI Playbook and the cross-Canada Rules of Engagement summary mentioned above. Its emphasis is on independent human review: never rely on an AI tool to judge its own accuracy, and verify against authoritative sources such as CanLII and official court websites. It is blunt about confidentiality, warning that the risk of inadvertently disclosing client information through AI tools cannot be overstated.
British Columbia (Law Society of British Columbia)
BC's guidance is most detailed on client information. If you cannot redact confidential information before it goes into a tool, the alternative it describes is informed, voluntary client consent, given in writing or orally with a written record of the conversation.
Manitoba (Law Society of Manitoba)
Manitoba's Generative Artificial Intelligence Guidelines for Use in the Practice of Law (April 2024) state that professional judgment cannot be delegated to generative AI and remains the lawyer's responsibility at all times. Manitoba also addresses billing directly: you can bill for the time you spend crafting and refining prompts, but not hourly time that the tool's efficiency saved.
Saskatchewan and the other provinces
The Law Society of Saskatchewan has issued its own generative AI guidelines, and the Alberta synthesis counts nine law societies with published guidance. If your jurisdiction has not issued a dedicated document, the general Code of Professional Conduct duties still apply in full. Competence, confidentiality, and candour did not wait for AI-specific rules.
Court practice directions: when you must disclose AI use
Law society guidance tells you how to practice responsibly. Court directions tell you what to put on the page. They vary more than the law society guidance does, so check the specific court every time.
| Court | What it requires |
|---|---|
| Federal Court | Notice issued December 20, 2023 (updated May 2024). If AI was used to create or generate content in a filed document, a declaration must say so in the first paragraph. The Court has said disclosure alone does not lead to an adverse inference. |
| Manitoba Court of King's Bench | Practice direction (2023): materials prepared with AI must indicate how AI was used. Manitoba was among the first Canadian courts to act. |
| Alberta courts | 2023 notice to the profession: requires "meaningful human control" and verification against authoritative sources, but no mandatory disclosure of AI use. |
| Ontario (civil) | Rules of Civil Procedure certification of the authenticity of cited authorities in factums, effective December 1, 2024. |
The trend is clear even where the rules differ. Courts are not banning AI. They are making lawyers certify, disclose, or both, so that there is no ambiguity about who stands behind a citation.
What happened to lawyers who got it wrong
Canadian courts now have a small but growing body of decisions about AI-invented authorities. Three are worth knowing, because they show the escalating seriousness with which courts treat the problem.
Zhang v. Chen, 2024 BCSC 285
In a BC family matter, counsel filed a notice of application citing cases that turned out to have been generated by ChatGPT and did not exist. Opposing counsel could not find them. The court ordered the lawyer personally to pay costs for the extra effort it caused the other side. It was the first widely reported Canadian decision on the issue, and it set the tone: an honest mistake is still the lawyer's mistake.
Ko v. Li, 2025 ONSC 2766
In an Ontario estates matter, Justice Myers found a factum with citations that could not be located on any case law service, links that went to unrelated decisions, and one case cited for the opposite of what it decided. The lawyer was ordered to show cause why she should not be cited for contempt. She acknowledged using AI, apologized, withdrew the factum, and undertook remedial steps. The decision is a reminder that the stakes now extend past costs to the court's contempt power.
R. v. Chand, 2025 ONCJ 282
In a criminal aggravated assault trial, Justice Kenkel found "numerous and substantial" errors in defence submissions on self-defence: at least one case that appeared not to exist, several citations that led to unrelated civil decisions, and others that did not support the point they were cited for. The court ordered counsel to personally prepare new submissions, hyperlink every citation to CanLII or another source so the court could check it, and not use generative AI, including commercial legal software built on it, for the legal research.
Notice the pattern in all three: the problem was never that a lawyer used AI. It was that nobody checked. For a longer look at why these fabrications happen, see AI hallucinated cases and how to avoid them.
Where ChatGPT is fine, and where it is risky
With the rules in view, here is how general-purpose chatbots map onto everyday legal work. The dividing line is simple: is the model working from material you gave it, or from its own memory?
| Task | Risk level | Why |
|---|---|---|
| Polishing a letter you wrote | Low | The substance is yours; watch for confidential details in the prompt. |
| Summarizing a document you supply | Low to medium | Grounded in your text, but check that it did not add facts or drop qualifiers. |
| Explaining an unfamiliar area at a high level | Medium | Useful orientation, but may blend American and Canadian law. |
| Finding cases to cite | High | Answers from statistical memory with no database behind it; this is exactly how the cases above happened. |
| Stating what a specific case held | High | Can misstate holdings of real cases as confidently as it invents fake ones. |
If a chatbot names a case, treat it as a lead to check, not a citation to file. And be aware that pasting client facts into a consumer chatbot may send them to a third party on terms you have not reviewed, which is precisely the confidentiality risk the LSO, LSA, and LSBC guidance warns about.
Retrieval-grounded tools: what changes and what does not
The architectural fix for fabricated citations is retrieval: the tool searches a real database of Canadian decisions first and answers only from what it found, linking each proposition to its source. CanLII Search+, Westlaw, Lexis+ with Protégé, and Casescout all take this approach in different ways. A tool that can only cite what it retrieved cannot invent a case that does not exist.
What retrieval does not change is your obligation. A real case can still be summarized wrongly, cited for a proposition it does not support, or overturned. The Chand order is instructive here: the court barred commercial legal software built on generative AI too, not just ChatGPT, because the failure was one of verification. The right tool makes verification fast. It never makes it optional.
That is the design goal behind Casescout's verified research mode: every answer links to the Canadian decision or Criminal Code section it relies on, and verbatim quotes are mechanically checked against the full judgment text before you see them, so checking a citation is one click rather than a fresh search.
A compliance checklist for any AI research tool
- Know where your prompts go. Read the data terms. Is your input stored, for how long, where, and is it used for training? If you cannot answer, do not enter client information.
- Redact or get consent. Strip identifying details, or follow the LSBC approach of informed client consent with a written record.
- Open every authority. Read the cited paragraph in the actual judgment on CanLII or a commercial service. Confirm it says what the tool claims.
- Note it up. Confirm each case is still good law before you rely on it.
- Check the court's direction. Federal Court filings may need a first-paragraph AI declaration; Ontario factums need the authenticity certification regardless of how you did the research.
- Bill honestly. Bill your time, including prompt work, not the hours the tool saved.
- Keep a record. Note which tool you used and how you verified the output. If a question is ever raised, you will be glad you did.
The bottom line
Canadian lawyers can use ChatGPT and other AI tools, and regulators expect competent lawyers to understand them. What no regulator accepts is unverified output. Use general chatbots for drafting and summarizing material you control, use a retrieval-grounded tool for finding authority, and read every case before your name goes on the filing. If you want a research tool built around that verification step, try Casescout free on your next research question.
Frequently asked questions
Is it against the rules for a Canadian lawyer to use ChatGPT?
No. No Canadian law society prohibits it. The guidance requires you to use it competently, protect client confidentiality, supervise and verify its output, and bill fairly. The professional consequences of a mistake fall on you, not the tool.
Do I have to tell the court I used AI?
It depends on the court. The Federal Court requires a declaration in the first paragraph of a filed document when AI generated content in it. Manitoba's Court of King's Bench requires materials to indicate how AI was used. Alberta courts require human verification but not disclosure. Always check the current practice directions of the court you are in.
Do I have to tell my client I used AI?
There is no universal rule, but the guidance points toward transparency, especially where client information goes into the tool. BC's guidance describes informed client consent when confidential information cannot be redacted, and the LSO lists client expectations among the factors to weigh.
Can I bill for work done with AI?
You can bill for your own time, including time spent writing and refining prompts and reviewing output. Ontario and Manitoba guidance both say you should not bill hourly time that AI efficiency saved.
What happens if I file a fake AI-generated case?
Canadian courts have ordered lawyers to personally pay costs (Zhang v. Chen, 2024 BCSC 285), required them to show cause why they should not be held in contempt (Ko v. Li, 2025 ONSC 2766), and barred them from using AI for research on the redo (R. v. Chand, 2025 ONCJ 282). Law society discipline is also possible.
Are legal-specific AI tools safer than ChatGPT?
For finding authority, yes, if they retrieve from a real database and link every claim to its source, because they cannot cite a case that is not in the database. They can still misread a real case, so the verification duty is unchanged. Our comparison of the best AI legal research tools in Canada covers how each one handles this.