Almost every week, a lawyer tells me about a client’s annoying use of AI. Some clients are pushing back after running their lawyer’s advice through ChatGPT. Others are sending massive dumps of AI-generated slop and forcing their lawyer to comb through it.
I wrote the policy below to help Canadian lawyers combat the frustrations of client AI use. I’ve seen many lawyers take a heavy-handed response — some going as far as an outright ban on clients using AI to help manage their legal matter. The policy below takes a different approach. It’s rooted in the premise that AI is now a central component of modern life (like it or not), and it’s more productive to accept that fact than deny it.
When we prohibit clients from using AI, we’re asking them to give up one of the most powerful tools they can access, at what is often one of the most challenging moments in their life. Clients aren’t likely to accept that, and a rule in your retainer that nobody follows is worse than no rule at all.
The policy below accepts that clients are overwhelmed, dealing with significant stakes, and want all the help they can get. It steers them toward productive uses of AI, highlights the cost of using AI poorly, and avoids the extreme positions your clients will likely just ignore.
Everything in the policy is meant to be a starting point. Your practice area and your clients may call for a different tone, and the parts about your own AI use should describe what you do.
The policy
Dear client,
I want to give you the best representation I can. To do that, I need your help.
I know you're probably using AI. I use it too. Used well, it can save us both time.
But AI can also be counterproductive. Used poorly, it will increase the amount of time I spend working on your file. And since you're paying for my time, it will also cost you money.
I've adopted this policy so we can keep the positives of AI and avoid the negatives.
The risk to keep in mind
You may be forced to share your AI conversations with the other side.
Solicitor-client privilege protects our communications, but it's yours to lose. At least one US court has said that litigants give up the protection of solicitor-client privilege when they discuss their legal matter with tools like Claude or ChatGPT. Canadian courts haven't decided this issue yet, but neither of us wants your file to be the test case.
Given the risk, I strongly recommend that you keep my advice out of your chats.
Things to avoid:
Uploading my correspondence.
Asking AI for a second opinion on my advice.
Sharing context that you wouldn't share with the opposing party.
Productive uses of AI
AI can help in plenty of places. Some of my favourites:
Preparing for our meetings. Ask it what questions you should ask your lawyer, or have it explain legal terms in plain language.
Organizing your facts. A timeline of events in your own words, or a list of your documents. This is some of the most useful material you can hand me.
Expanding your perspective. Consider asking AI things like: "What am I overlooking?"
What I ask of you
If AI helped you create something you're sending me, read it first and cut what doesn't matter. Use your own words to explain what you want me to take from it. If you send me unfiltered AI answers, I'll be forced to review and comb through them at my hourly rate. Neither of us wants that.
Set your tools up properly. If you use AI for anything sensitive, use a paid plan and turn off model training in the settings. It takes two minutes, and it stops the tool from training on what you type. Ask me if you want help checking.
Check with me before you act. AI sounds confident whether it's right or wrong. Before you act on anything an AI told you about your matter, run it by me. That conversation is quick and much cheaper than unwinding a mistake.
Handling disagreement
At some point, AI might give you the impression that I've gotten something wrong. I'm happy to discuss these situations with you. But here are a few things you should think about beforehand:
- AI tools are built to be helpful. They will often agree with you even when you are wrong.
- AI can only work with what you give it. We often share the facts that help us, and leave out the facts that hurt our case.
- AI tools don't have a strong understanding of Canadian law, especially in nuanced areas. Most Canadian legal decisions can't be accessed by general-purpose AI tools (like Claude and ChatGPT). Those decisions don't form part of the tool's knowledge base, which increases the risk that AI will give you a bad answer.
- You're paying me to understand your entire matter (including the things you wouldn't share with a chatbot), and to give you advice based on years of experience in the Canadian legal system.
If you decide I should consider an answer you received from an AI tool, be prepared to share the full conversation (not just the AI tool's answer). This will help me understand how your conversation progressed, and potentially help me understand the source of the disagreement.
I promise to keep an open mind. But no matter what I hear from an AI tool, I’ll always need to form a professional opinion about your case and the strategy we adopt. I can't advance a position that my professional judgment doesn’t support. So, if we reach a point where you prefer a chatbot’s answer over my reasoned advice, I may need to consider ending our engagement.
My commitment to you
I will extend you the same courtesy I've asked of you in this policy. The work I deliver to you reflects my own independent judgment. I'll never send you a raw answer from AI and expect you to parse through it. And I won't use AI in a way that would jeopardize our solicitor-client privilege.
If you have any questions about how I use AI to make my practice more efficient, don't be afraid to ask.
[Signature]
Notes on the policy
The privilege warning
The risk is that a judge treats a chat about the file as a waiver of privilege, and your client ends up producing their AI conversations for the other side. A US court went that way in February 2026. No Canadian court has decided the issue, and the uncertainty is reason enough to warn your clients.
Be careful how you explain it, though. On a paid plan with model training switched off, today's mainstream tools aren't training on what your client types. Tell a client their chatbot is broadcasting their secrets and they can disprove you in ninety seconds, and you'll have spent your credibility on a scare story.
The disagreement section
When a client tells you the chatbot disagrees with you, arguing with its answer puts you in a debate against a tool built to flatter whoever is typing. Asking for the whole conversation puts the two of you on the same side of the table, diagnosing the answer together instead of treating it as true. Sometimes the client left out context that changes things. Sometimes the model doesn't know the Canadian cases or the nuance that decides the point. If you read the whole exchange, you'll be in a better position to explain what the client's AI tool missed.
Setting up their tools
Helping a client turn off model training costs you nothing, and it hands them something useful instead of taking something away. If you want to walk a client through the difference between the plans, what Claude's data terms say is a reasonable place to start. The same questions apply to whatever tool your client uses.
Your own AI use
The policy covers your client's AI use. Your own use raises different questions, starting with your law society's guidance. And if your firm doesn't have its own written AI policy, write that one next. It's hard to ask a client to be careful with a tool you haven't set any rules around yourself.
Work through it with other lawyers
The lawyers in Inn Laws deal with the same clients this post describes — the document dumps, the ChatGPT second opinions — and compare notes on how to respond. What your retainer should say about AI is a decision members work through together. If you're interested in working through these issues with a brain trust of sharp peers, apply below.