For a long time the deal with a professional was simple: the client came to you because you knew things they did not, and you could act on that knowledge for them. That deal is changing. More and more clients arrive having already asked a chatbot, holding an answer that sounds complete and expecting you to carry it out. Sometimes that answer is right. Often it misses the context that decides the question. Tell the client they are wrong and you risk offending them; go along with it and you risk their money.
Accounting is not alone in this. Doctors see patients who walk in with an AI-generated diagnosis, and travel agents field itineraries a model built around places that do not exist. But accounting is among the fields most exposed to it.
The AI-briefed client
Picture the owner of a small business in Queens who forwards her accountant a ChatGPT screenshot: change how the business is registered and save thousands in tax every year. On paper it works. For her business it does not, because New York City’s own rules and the salary she would have to pay herself would eat most of the savings. Her accountant now has to explain all of that without sounding defensive, so she does not make an expensive mistake. And some clients in her place start to wonder whether the chatbot knows something their accountant does not.
Cases like hers are becoming routine. OpenAI has said that tax questions put to ChatGPT in the US roughly quadrupled between the first quarter of 2025 and the first quarter of 2026, while noting itself that the tool is not a substitute for professional advice. Adobe found that 26% of US workers planned to use AI to file their taxes in 2026, up from 11% in 2024. Clients are arriving better armed and less informed at the same time, and the work of checking what they bring falls on the firm.
Why accounting is especially exposed
Tax rules are public, but how they apply depends on things a general-purpose model does not see: where the client is, how the business is set up, which tax year it is, what was filed before. Thresholds, deadlines and credits move every year, and a model can confidently apply last year’s figure. The models are also less consistent than a single good answer suggests. In TaxBench’s 2026 evaluation of 16 frontier models, the best one answered tax-knowledge questions correctly 84% of the time on a first attempt, but only 42% of the time when it had to get the same question right across five attempts.
The cost of those misses lands on the professional, who has to find the flaw in the model’s reasoning, explain it and defend the correction. That creates a trust paradox: the more polished the chatbot’s plan, the more a correction can look like an accountant being defensive or out of date. Most firms do not bill for this work, yet the liability for what gets filed stays with them. For a profession already short of experienced people, it is one more load on teams that are already stretched.
Five ways to handle it
- Set expectations at onboarding. Tell clients how the firm treats AI-generated advice and who is accountable for what gets filed. That removes the grey zone where work is done but not billed, or where a practitioner ends up owning a decision that was never theirs.
- Treat the client’s AI questions as a signal. Ask what they asked the chatbot and what it told them. The questions people type into a model are often the worries they hesitate to raise with their accountant.
- Show how you checked. Point to the rule or publication behind your conclusion instead of dismissing the chatbot’s plan. It makes the answer easier to accept, and it shows the client where the model fell short.
- Keep a library of recurring misconceptions. The same wrong answers come back again and again. Write standard responses once and share them before the questions arrive.
- Use AI in your own operations. Every step above takes time a fully manual firm does not have. Automating the routine work is what frees accountants for these conversations.
What changes for service firms
Information keeps getting cheaper, and professional firms are no longer the gatekeepers of it. In theory anyone can now look up how to do what they used to pay for. What clients still pay for is time and peace of mind. Accounting is a trust layer in the financial system, and as clients increasingly arrive with a plan already in hand, the accountant’s value moves from explaining to verifying and deciding. The same is true of law firms, consultancies and advisory practices, and that kind of trust is built by people.
The real test is whether a firm gives its accountants the time for that relationship work. In a profession short of people, AI is part of the answer: it takes on the routine work and leaves the judgment and the decisions to the professionals. That is also how we run the practices that join Numica. Clients will always have a second opinion now. The firm’s job is to be the one they trust.