The adoption number is real and it is specific. In a survey of more than 1,000 tax professionals published on 8 June 2026 by Blue J and CPA.com, 60% now use AI for tax research at least weekly, up from 33% the year before. Eighty-four percent said it saves time.
We are not your compliance adviser and none of this is advice. It is a description of published rules with the links attached, so you can hand it to the person who is.
So we went looking for the other half of that sentence. How many practices of two to twenty-five people have written down which tools are allowed, what may be pasted into them, and who checks the output.
We could not find it. Not a small number — no number.
Two statistics we are not going to repeat
There are two figures circulating in this space that would have made this post much easier to write. One says 46% of US accounting firms have inadvertently put confidential client information into public AI services. The other says fewer than 21% have formal governance over how AI is used, against more than 75% of staff using it.
We chased both. Neither leads back to a survey with a stated sample size, a field date or a described method — they sit on aggregator pages and a contributed column, each citing the last one. They may well be true. We are not going to be the fourth page repeating a number nobody can show you the sample for.
If somebody quotes you a governance statistic about small accounting firms, ask who was surveyed and when. We asked, and the answer ran out.
What is published, and what it does not cover
AICPA & CIMA surveyed 1,735 executives across eight geographies in the autumn of 2025 and published on 25 February 2026. Only 24–27% reported adequate AI-skilled talent, IT system readiness or regulatory preparedness. Among smaller organisations, fewer than one in five had the talent or the systems.
Grant Thornton's 2026 AI Impact Survey of nearly 1,000 business leaders, reported in the Journal of Accountancy, found 22% with a fully developed and implemented enterprise AI strategy, and 78% lacking confidence they could pass an independent AI governance audit within 90 days.
Both are good surveys. Neither is a survey of small practices. They are executives at organisations of every size and sector, and reading them as a statement about a six-person CPA firm is exactly the borrowing we just refused to do. They point in a direction. They do not measure your street.
Not measured: how many wrote anything down about it.
Available instead: surveys of organisations generally, quoted honestly.
What is not in doubt
The evidence about adoption is thin in places. The published position of the regulators is not thin at all, and it is short enough to read yourself.
FINRA Regulatory Notice 24-09, issued 27 June 2024, says the rules are technology-neutral and apply whether a firm builds the tool or buys it — explicitly including, in its own words, "through embedded features in existing third-party products." It states that it creates no new requirements.
That last part is the part worth sitting with. Nothing new was created, which means nothing was waiting to start.
On the adviser side, the SEC Division of Examinations published its fiscal year 2026 examination priorities on 17 November 2025. Two of the things examiners say they will look at: whether a firm has policies to monitor and supervise its use of AI, and whether what the firm says about its AI matches what the firm does.
There is already a settled case about the second one. On 18 March 2024 the SEC announced settled charges against two investment advisers over their AI claims. Neither was accused of using AI badly. They were accused of describing it inaccurately.
The question that sorts this
It is not are we allowed to use this. Nothing published says you are not. It is:
- What did we use, on what client work?
- Where did the text go, and where did the output end up?
- Who checked it, and how would we show that?
A practice that can answer those three has a governance story. A practice that cannot has the same exposure whether it used AI once or a thousand times, and it will find out which on somebody else's schedule.
What to do
Nobody has measured this for practices your size. So measure your own.
One line, sent to everyone, answered anonymously: have you used an AI tool on client work in the last month? Not to discipline anybody — say so in the same sentence, or you will get a number that is worse than no number. Somebody else can send it and count it. You need the total, not the names.
That is a morning, and at the end of it you have the figure this whole post says does not exist for firms like yours.
The cost of not having it is not a penalty. It is that every decision after this one — what to approve, what to buy, what to write down, what to tell a client who asks — gets made against a guess. That is free right now and expensive on the day somebody asks you to justify it.
If you would rather work through what the number means with someone who has had to answer it about their own system, that is what a first conversation is for.