The Study Guild

Can UK accountants use ChatGPT with client data?

Short answer

Only after checking. UK accountants may use AI tools with client data where the firm has established where that data goes, has a lawful basis under UK GDPR, and maintains supervision proportionate to the work. Consumer accounts on default settings usually fail that test, so the decision must be made and documented deliberately.

What do the rules actually require?

Three things: a lawful basis for the processing under UK GDPR, a contractual position with the vendor that protects confidentiality, and supervision proportionate to the risk of the task. None of these are new obligations. What is new is that a tool now moves client data somewhere you did not previously have to think about.

Professional confidentiality obligations are not suspended because a task is delegated to software. If you would not forward a client's management accounts to an unknown third party without checking the terms, the same reasoning applies to pasting them into a chat window.

The practical test we teach: could you explain to that client, accurately and without embarrassment, where their data went and who could see it?

Sources

Does it matter whether you use a consumer or business account?

Substantially. Business and enterprise tiers of the major AI tools generally offer contractual commitments on data retention and training that consumer tiers do not, and provide administrative controls. The difference is contractual rather than technical, which is precisely why it needs reading rather than assuming.

Vendors also change these terms. A position you verified eighteen months ago may no longer hold, and you will not be notified in a way anyone reads.

Build a review date into the firm's AI policy rather than treating the assessment as a one-off.

What should never go into a general-purpose AI tool?

Identifiable client data where you have not verified the vendor's terms, anything covered by a confidentiality undertaking that names permitted recipients, and material subject to legal privilege. Anonymising properly is harder than it looks — small-company financial detail is often identifying on its own.

Is confidentiality the biggest risk, or is it accuracy?

Confidentiality is the risk that ends careers, but accuracy is the risk that occurs daily. Models fabricate references to tax guidance confidently and make arithmetic errors on long documents quietly. Both failure modes look like competent work, which is why a checking control matters more than a better prompt.

PatternWhat goes wrongWhat to do instead
Consumer ChatGPT, default settingsTraining and retention terms usually fail a client-data testUse a business tier after reading the DPA, or do not paste client data
"Anonymised" management accountsTurnover plus sector plus region often identifies a small companyTest anonymisation; do not assume removing the name is enough
Unchecked tax citationsConfident fabrications of HMRC or ICAEW referencesVerify every authority against the primary source before it leaves the firm
Unsupervised junior useThe tool looks like a senior reviewSupervision proportionate to the task, named in the file

Who should skip accountancy-AI training?

Anyone wanting to learn bookkeeping from scratch. Anyone hunting a list of tool recommendations rather than a method for judging tools. Anyone whose firm has already banned AI. Anyone who is not in, or entering, UK accountancy — the whole design assumes that professional context and those duties.

SituationWhat to buyWhat to skip
Consumer ChatGPT on a shared loginStop. Read the terms, or do not paste client dataPrompt courses that ignore the data destination
Firm wants a documented methodProfession-specific training that starts from confidentialityGeneric ChatGPT certificates
You wanted to leave accountancyA different career, honestlyAn AI-for-accountants programme

The solicitors answer is the sibling: same structure, privilege instead of tax citations. If you cannot explain to a named client where their file went, the tool is not ready for live work, however good the prompt.

Membership and a live cohort are different products. Do not buy a library priced like a qualification, and do not buy a cohort if you will not submit work to be marked.

Common questions

Does ICAEW prohibit using AI?

No. Professional bodies have issued guidance rather than prohibitions, and the guidance broadly restates existing obligations on confidentiality, competence and supervision. Check your own body's current guidance directly — it is updated periodically.

Is anonymising client data enough?

Sometimes, but genuine anonymisation is harder than removing the name. Turnover, sector and region together can identify a small company. Treat anonymisation as a control to be tested, not assumed.

Do we need to tell clients we use AI?

There is no blanket requirement, but engagement letters and privacy notices should be accurate about how their data is processed. If your notice does not currently cover it, it needs updating.

Learn this properly

If you want to do this as a career rather than read about it, these live programmes cover it in depth.

Starts 3 Feb 2027Intermediate

AI for UK Accountants

Redesign your practice workflow around AI tools that actually hold up under client confidentiality, ICAEW obligations and a real audit trail.

Not for: Anyone wanting to learn bookkeeping or accounting from scratch

Live hours
16h live
Duration and weekly load
8 weeks · ~5h/week
Cohort size
max 20

£499or from £83.16/mo, interest-free

More on AI for Professionals