Clients bring ChatGPT tax answers to their accountant. What should a Canadian firm publish?
Canadian accountants report clients challenging advice with AI answers. The guidance pages, fee and service facts, and triage step a firm should put in place.

The short answer
A Canadian accounting firm should publish dated guidance pages that answer the questions clients put to ChatGPT, plain facts about its services, provinces served and fees, and a named author on each page. Then add a triage step for AI-sourced questions, so an accountant reviews what the client read before replying, instead of arguing with it.
Key takeaways
- Canadian accountants surveyed for Dext in December 2025 reported clients leaning on public AI tools and using AI outputs to challenge their advice.
- In the Thomson Reuters Institute's survey of 1,816 professionals in 62 countries (March to April 2026), 78% of corporate clients called AI-enabled quality improvements from their providers very important or essential, and 6% said most or all providers deliver them.
- In Similarweb's US data for May 2026, under 4% of ChatGPT answers about professional services carried a citation, so being named and described correctly matters more than links.
- Guidance pages that state the tax year, who the rule applies to and when it does not give engines and clients something precise to repeat.
- Our view: treat an AI answer a client brings in as an intake document to review, not an argument to win.
In this article
What Canadian accountants are reporting
Clients now arrive with an answer already in hand. A survey of Canadian accountants and bookkeepers commissioned by Dext, which sells accounting software, was fielded in December 2025 and published in January 2026. Most respondents had seen clients rely more on public AI tools, and many had seen clients use AI outputs to question or challenge their advice.1 We describe it without figures because it is not in our statistics bank.
The pattern is familiar to anyone who has worked a client desk. A business owner asks ChatGPT whether a vehicle expense is deductible, gets a confident answer, and then asks the accountant why the return says otherwise. The AI answer may be right in general and wrong for that client's facts, province or tax year.
Our view: this is a publishing problem as much as a client-service problem. The answer the client read was assembled from pages someone published. If the firm's own pages are vague, undated or missing, the engine builds its answer from other people's.
Why does the AI answer carry so much weight?
Because clients now expect their providers to work with AI, and most think providers fall short. In the Thomson Reuters Institute's Future of Professionals 2026 survey of 1,816 professionals in 62 countries (March to April 2026), 78% of corporate clients said AI-enabled quality improvements from their providers were very important or essential.2 Only 6% said most or all of their providers deliver them.2
That gap shapes the conversation at the desk. A client who believes the firm is behind on AI gives the chatbot's answer more credit than it deserves. The firm then spends time defending advice instead of giving it.
There is a second reason. In Similarweb's US data for May 2026, under 4% of ChatGPT answers in professional services carried a citation, the lowest of the sectors it reported.3 The data is US, but the shape of the questions is the same in Canada: most clients read a general answer with no link to check. Our note on citation rates by sector covers what that means for measurement.
What should a firm publish?
Three kinds of page, each written so an engine can quote it accurately: guidance pages for the questions clients actually ask, plain facts about the firm, and author pages that show who stands behind the advice. None of it requires a new website, only pages that are precise where most firm sites are vague.
Guidance pages that state their limits
Write one page per recurring client question, in the client's words. "Can I claim my home office as an employee?" is better than "Home office expenses: an overview". Open with the short answer, then say which tax year it applies to, who it applies to and when it does not.
The limits are the part AI answers usually drop. A page that says "this applies to employees with a signed form from their employer, for the 2025 tax year; self-employed clients follow different rules" gives an engine a sentence it can repeat correctly. It also gives your accountant a link to send when a client arrives with a broader answer.
Date every page and review it each filing season. Name the Canada Revenue Agency forms and guides you rely on and link to them. An undated page from three seasons ago is how outdated rules end up in AI answers.
Facts about the firm, stated plainly
Engines describe firms from what they can find. If your site never says which provinces you serve, whether you work remotely, which client types you take or how you charge, the answer to "which accountant should I use for a small corporation" is built from directory listings and review sites.
State the facts on one page: services, client types, provinces, designations held, languages and how fees work. Publishing fee ranges is a judgement call for each firm. Our view: a range, or at least the basis of your fees (fixed per return, hourly, monthly package), answers a question clients already ask AI. It also filters enquiries before they reach you.
Author pages
Put a named accountant, with their designation, on every guidance page, and link to a short profile. It tells clients a qualified person stands behind the page. It also gives engines an entity to attach the advice to, which helps when someone asks who explains a topic well.
The guidance page pattern
- Question as the heading
In the words clients use when they ask ChatGPT.
- Short answer first
Two or three sentences that stand on their own.
- Scope and limits
Tax year, who it applies to, when it does not.
- Sources and author
CRA links, a named accountant, a review date.
- Next step
When to book a call instead of relying on the page.
How should a firm handle a client's AI answer?
Treat it as an intake document. Ask the client to paste the question and the answer they received, then have an accountant review it against the client's facts before replying. The reply should say which parts are right in general, which do not apply to the client and why.
The habit of checking AI answers with a person is already common among business buyers. In Gartner's survey of 645 B2B buyers (August to September 2025; geography not stated), 69% said they prefer to validate AI-generated insights with sales reps.4 Clients bringing a tax answer to their accountant are doing the same thing. Make it easy and the conversation starts from the facts instead of from a disagreement.
A triage agent for AI-sourced questions
A small agent can handle the intake. It reads what the client pasted, tags the topic and tax year, finds the firm's matching guidance page and drafts a reply for the accountant to review. It never sends advice on its own; a person approves every reply.
The value is in routing and preparation. In filing season, an accountant who opens a question with the client's AI answer, the relevant page and a draft already attached saves time on every message. Our AI automation page describes how we build agents that end in a draft for a person to approve.
Our view: keep this agent away from final advice, deadlines and anything that commits the firm. Its job is to make the accountant faster, not to answer the client.
What to fix first
Start with the answers engines already give about you. Ask ChatGPT, Gemini and Perplexity what they know about your firm, and the five questions clients most often bring in. Note where your firm is missing, misdescribed or out of date.
Then fix in this order: the firm facts page, the five guidance pages, and author profiles. If an engine states something wrong about the firm, such as a closed office or a service you no longer offer, our guide to correcting what ChatGPT says about your company sets out the steps.
Law firms face a similar pattern with different rules on advertising, and our note on how people choose lawyers with AI covers what those surveys measure. For accounting firms, the publishing work is simpler: precise, dated pages and plain facts, reviewed every season.
Sources
- Dext, Canadian businesses losing money to ChatGPT-style tax and financial advice (Jan 2026)
- Thomson Reuters Institute, Future of Professionals 2026: 1,816 professionals in 62 countries, Mar–Apr 2026 (Jun 2026)
- Similarweb, AI search stats 2026: ChatGPT answers with a citation, US, Jun 2025–May 2026 (Jul 2026)
- Gartner, survey of 645 B2B buyers, Aug–Sep 2025; geography not stated (May 2026)


