Blanket disclosure on every deliverable trains clients to ignore it. The question worth asking is narrower: does the disclosure change what a reasonable client thinks they are paying for?
Key takeaways
There is no general Canadian rule requiring a professional firm to tell a client that a tool helped produce their work, and stamping every letter with a disclaimer is worse than saying nothing — it is unread text that implies the firm is uncertain about its own product. The useful test is whether the disclosure would change what a reasonable client believes they are buying. If it would, say it. If it would not, you are adding noise.
That test is not invented. The Law Society of Ontario’s Futures Committee white paper on licensees’ use of generative AI, recorded in the Law Society of Alberta’s survey of Canadian guidance, declined to require disclosure every time and instead listed four considerations: whether the use will necessarily be disclosed publicly, whether the client reasonably expects the material to be prepared by a licensee, whether there is reputational or other risk to the client, and whether the use requires inputting the client’s personal or proprietary information. It works as a general test for any profession.
Canada’s Voluntary Code of Conduct on the Responsible Development and Management of Advanced Generative AI Systems is frequently cited in firm policies as though it created a disclosure duty. It does not. By its own terms it identifies measures to be applied by organisations developing or managing the operations of generative AI systems with general-purpose capabilities, with additional measures for systems made widely available. Its transparency outcome is that sufficient information is published to allow consumers to make informed decisions and for experts to evaluate whether risks have been adequately addressed — a commitment made by developers and managers who sign it, and its list of signatories is published on the page.
The code is also explicit that it does not change existing legal obligations, giving federal privacy law as its example. So it is a useful description of what good practice looks like and a bad citation for "we are required to tell you". Firms should describe it accurately or not at all.
The exposure is not on the deliverable. It is on the website, the capability deck and the proposal — the places where you already told the client what they were buying.
Paragraph 74.01(1)(a) of the Competition Act makes it reviewable conduct to make, for the purpose of promoting the supply or use of a product or any business interest, a representation to the public that is false or misleading in a material respect. The Competition Bureau’s guidance on false or misleading representations explains that material information is information that could influence consumer behaviour, and that it is unnecessary to show that anyone was actually deceived or misled.
Read that against a firm page promising that "a senior specialist personally prepares every report". If a model prepares the first draft and a junior edits it, the problem is the sentence on the website, not the model. The cheapest fix is usually to rewrite the promise rather than to bolt a disclosure onto the output.
The other half of section 74.01 is the half firms walk into when they start marketing their own AI capability. Paragraph 74.01(1)(b) makes it reviewable conduct to make a representation to the public in the form of a statement, warranty or guarantee of the performance, efficacy or length of life of a product that is not based on an adequate and proper test — and the section places the proof of that test on the person making the representation.
"Our AI review is ninety-nine per cent accurate" and "we cut turnaround by half" are both performance claims. If the test behind them is a partner’s impression of last quarter, do not publish them. Describe the mechanism instead: what the tool reads, what it produces, and who checks it.
The advertising standards side reaches the same conclusion by a different route. Clause 1 of the Canadian Code of Advertising Standards assesses a claim by the general impression it conveys rather than the precise legality of the presentation, prohibits omitting relevant information where the omission makes the advertisement misleading, requires all pertinent details to be clearly and understandably stated, and requires claims to be supported by competent and reliable evidence which the advertiser will disclose to Ad Standards on request. Disclaimers, it adds, must not contradict more prominent aspects of the message.
Separately from marketing, an Ontario lawyer owes a duty of candour to the client on matters relevant to the retainer — rule 3.2-2 requires a lawyer advising a client to be honest and candid, and its commentary ties that to the lawyer’s fiduciary obligations. Answering "did you use AI on this?" evasively is the failure mode, not using the tool.
Accountants, engineers and brokers are under differently worded duties that land in the same place. In every case the answer that works is the plain one: what the tool did, who reviewed it, and who is responsible for the result.
If a disclaimer is doing the work of reducing your liability rather than informing the client, read Treadstone Law’s note on whether a disclaimer in an engagement letter avoids liability for negligence first. It does less than most firms assume.
An accounting practice adds AI-assisted drafting to its year-end reporting workflow and asks whether to add a line to every report. The partner group reviews its own marketing instead and finds three claims that no longer hold: a promise of "hand-prepared" analysis, a turnaround guarantee that predates the change, and a testimonial page implying a named partner performs every review personally.
The firm rewrites the three claims, adds two sentences to the engagement letter, and adds nothing to the reports. The disclosure question turned out to be a marketing accuracy question, which is the usual outcome.
Do clients actually want to know?
Some do, and the ones who ask are the ones whose expectations you needed to check anyway. Treat the question as useful information about that client rather than as a problem to be managed.
Is disclosure ever mandatory?
For court filings it can be, and the requirement differs by court — several Canadian courts require a declaration and several expressly do not. Check the rules of the specific court or tribunal before filing; the Alberta survey records the current position across jurisdictions.
What if the client asks what tool we use?
Answer. Naming the product, saying whether client information goes into it, and saying who reviews the output is a short and complete reply. Refusing to name it invites the inference that there is a reason.
Does telling clients increase our risk?
The opposite, in the common case. The risk is a client discovering something they were led to believe was not happening. The Office of the Privacy Commissioner’s openness principle is built on the same logic — information about a system should be understandable by its intended audience and readily available before, during and after use.
A 30-minute call is enough to tell you whether AI pays for itself here.
