The Thomson Reuters Institute surveyed more than 1,500 professionals across 27 countries for its 2026 AI in Professional Services report, published in February. Two findings from it belong together.
The Policy Has the Wrong Reader
More than half of corporate legal and tax departments want the firms they engage to use AI. Fewer than one third know whether their firms do.
Most AI policies in accounting firms are written as internal documents. Approved tools, prohibited uses, a line about not pasting client data into consumer products, a sign-off sheet in the staff handbook.
All of that is necessary. None of it addresses a client who has formed a view about AI and no way to check it against reality.
The same report found 40% of firm respondents had received conflicting instructions from clients about AI use. Not objections. Conflicting instructions, which is what happens when several people at a client have opinions and nobody has been given a document to react to.
That is a communication vacuum, and vacuums get filled by assumption. Some clients assume you are using AI on everything. Some assume you are not using it at all. Both are making decisions about your firm on that basis.
Adoption Was Never the Bottleneck
It is worth being clear that this is not a story about reluctance.
Organisation-wide AI use in the same survey nearly doubled to 40% in 2026, from 22% the year before. Fifteen percent of organisations had already adopted agentic AI tools and 53% were actively planning or considering it.
What has not kept pace is measurement and confidence. Only 18% of organisations track the return on investment of their AI tools at all, and most of the tracking that does exist follows operational metrics rather than business impact. Separately, the 2025 PCPS National Management of an Accounting Practice Survey found just 16% of firms were very confident in their ability to adapt to AI and automation over the following three years.
So the profession is adopting quickly, measuring rarely, and feeling uncertain. A policy written in that state tends to be defensive, and a defensive policy is exactly the one you cannot show a client.
Why the Banned Tools List Ages So Badly
Here is the practical failure mode. Most first-draft AI policies are lists: these tools are approved, these are prohibited.
That document is obsolete within a quarter. A tool on the approved list ships a feature that changes where data goes. A prohibited tool becomes the default assistant inside software the firm already pays for. Somebody discovers a new product and there is no process for deciding about it, so they either use it quietly or ask a partner who guesses.
A policy built on tool names has to be rewritten every time the market moves, which is why so many of them are dated last year and quietly ignored.
Write the Kind That Survives
A policy built on decisions rather than products holds up, and it has the useful property of being showable. Four questions do most of the work.
What categories of client data may be processed by an outside system, and under what terms. Who reviews AI-assisted output before it leaves the firm, named by role rather than by person. What clients are told, when, and in what document. And how any of it is recorded, so the answer to what did we do on this engagement is not a memory.
Notice that none of those name a vendor. They survive the next product launch, and they answer the questions the survey says clients are actually asking.
The Record Is the Part That Gets Skipped
The fourth question is the one firms most often leave until something goes wrong, and it is the only one that cannot be answered retroactively.
MetaWurks covers part of it. Documents ingested into the platform are not used to train models and are not exposed to other users, so client material stays inside the firm's control while an accountant queries it in plain English. Role based access controls decide who can open which client's records, and audit logs record who opened what and when, which is the artefact that turns a policy statement into something demonstrable.
It does not write the policy and it does not decide what clients should be told. Those are the firm's judgments, and they are cheap to make and expensive to postpone.
More than half of your clients want you using this technology. Most of them have no idea whether you are. That gap is not closed by a better internal document. It is closed by one you are willing to send them.
Join the Conversation
If a client asked tomorrow whether AI touched their work this year, is there a document you could send, or would you have to compose an answer?