AI policy for accountants: a one-page template for UK practices

16 November 2026·The AI Ledger editorial team
AI policy for accountants: a one-page template for UK practices

A workable AI policy for accountants fits on one page and needs about 10 clauses: which tools are approved, what client data may go into them, who checks the output, and what happens when something goes wrong. Below is a template you can copy today, followed by the reason each clause is there, so you can adapt it rather than adopt it blindly.

Most small practices do not need a 20-page AI governance framework. They need a short, clear rule set that staff will actually read, and that a partner can point to if a client or regulator asks how AI is used. If you are still working out what AI is good for in the first place, start with our free lesson on what AI can and cannot do in an accounting practice, then come back to the policy.

This guide is general information, not legal advice. Check the final policy against your engagement letters, your professional body's requirements and, if in doubt, your own adviser.

Why a small practice needs an AI policy now

Staff are already using AI, whether or not the practice has approved it. The risk is rarely a dramatic failure. It is quieter than that: a junior pastes a client's bank statement into a free chatbot to "tidy it up", or a draft letter goes out with a confident but wrong statement about a relief.

A one-page policy does three jobs:

  1. It tells staff which tools are safe to use and which are not.
  2. It sets a minimum standard of review before anything AI-assisted reaches a client.
  3. It gives the practice a written record that it thought about confidentiality and data protection, which matters under UK GDPR and under the confidentiality principle in accountancy ethics codes.

The template

Copy the block below into a document, replace the bracketed text and delete any clause that does not apply. Keep it to one page if you can.

[PRACTICE NAME] - AI USE POLICY
Version [1.0]  |  Approved by [partner name]  |  Date [dd/mm/yyyy]  |  Review by [dd/mm/yyyy]

1. PURPOSE
   This policy sets out how staff may use artificial intelligence (AI) tools in
   client and practice work. It applies to partners, staff, contractors and
   anyone working on practice systems.

2. APPROVED TOOLS
   Only the tools on the approved list may be used for practice work:
   - [Accounting software AI features, e.g. in-ledger categorisation and assistants]
   - [Receipt and document capture tool]
   - [Business-plan general assistant, e.g. under a company workspace account]
   Any new tool must be approved by [named person] before use.

3. PERSONAL AND FREE ACCOUNTS
   Personal or free AI accounts must not be used for client work, even with
   training settings switched off.

4. CLIENT DATA
   - Never enter client data into a tool that is not on the approved list.
   - In approved general assistants, remove names, addresses, UTRs, NI numbers,
     bank details and account numbers unless the task genuinely needs them.
   - Never enter passwords, government gateway credentials or full bank logins.

5. HUMAN REVIEW
   Every AI-assisted output that reaches a client, HMRC or Companies House must
   be reviewed by a competent person who takes responsibility for it, exactly as
   if they had written it themselves.

6. TAX AND TECHNICAL POSITIONS
   AI may help research and draft, but no tax treatment, relief, deadline or
   figure may be relied on without checking it against a primary source
   (legislation, HMRC guidance or the practice's technical resources).

7. TRANSPARENCY WITH CLIENTS
   We will tell clients, in our engagement letter or privacy notice, that we use
   AI tools to support our work and that a person reviews all advice.

8. RECORDS
   Where AI materially shaped a piece of advice, note it on the file: the tool,
   the date and who reviewed the output.

9. INCIDENTS
   If client data is entered into an unapproved tool, or an AI error reaches a
   client, report it to [named person] the same day. We will assess whether it
   is a personal data breach that must be reported.

10. TRAINING AND REVIEW
   All staff will read this policy and complete basic AI training. This policy
   will be reviewed every [6/12] months or when we add a new tool.

Signed: ____________________   Name: ____________________   Date: __________

Why each clause matters

1. Purpose

A short scope statement stops the "I did not think it applied to me" problem. Naming contractors matters because many small practices use freelance bookkeepers or outsourced payroll help who work on practice systems.

2. Approved tools

This is the clause that does most of the work. An approved list turns a vague instruction ("be careful with AI") into a clear one ("use these, not those"). In most UK practices the list will start with the AI already inside the accounting software you pay for, such as Xero's AI features, Intuit Intelligence in QuickBooks or Sage Copilot, because that data is already inside a system covered by your existing supplier terms. Specialist capture tools come next. A general assistant goes on the list only under a business plan with proper terms.

Naming one person who approves new tools keeps the list current and stops tools arriving by stealth.

3. Personal and free accounts

Consumer AI plans are built for individuals. Their terms, data controls and retention rules are different from business plans, and the practice has no contract or admin control over a member of staff's personal account. Even where a consumer plan lets you switch off training, the account still belongs to the individual, not the firm. A blanket rule is simpler than a judgement call each time. Our guide on whether it is safe to put client data into AI compares what the main vendors say about consumer and business plans.

4. Client data

Most AI tasks in a practice do not need personal identifiers. A model can redraft a letter about a late filing penalty without knowing the client's name or UTR. Stripping identifiers by default reduces the harm if something does go wrong, and fits the data minimisation idea in UK GDPR.

The ban on credentials is absolute. There is no task where an AI tool needs a client's Government Gateway login.

5. Human review

AI tools produce fluent text, and fluent text is easy to trust. The review clause makes the reviewer responsible for the output as if they had written it. That is the standard a client and your professional body will hold you to anyway, so the policy simply says it out loud.

6. Tax and technical positions

General assistants can state rules confidently and get them wrong, cite thresholds from an earlier tax year, or merge two reliefs into one. This clause does not ban AI from technical work. It says AI is a starting point for research, and the primary source is the authority.

7. Transparency with clients

Clients increasingly ask whether their accountant uses AI. A line in the engagement letter or privacy notice answers the question before it is asked, and it supports your transparency duties under data protection law. It also sets the expectation that a person, not a model, stands behind the advice.

8. Records

You do not need to log every time someone uses AI to fix a spreadsheet formula. You do want a note on the file where AI shaped advice, so that if the advice is challenged later you can show who checked it and when.

9. Incidents

Mistakes will happen. The aim of this clause is speed: someone pastes data into the wrong tool, tells the named person the same day, and the practice decides calmly whether it is a personal data breach. Under UK GDPR, some breaches must be reported to the ICO within 72 hours of the practice becoming aware of them, so a same-day internal report keeps that window open.

10. Training and review

AI tools change quickly, and so do vendor terms. A review date written into the policy means it is looked at again, rather than quietly going out of date. Basic training ensures everyone understands what the tools are good at and where they fail.

How to roll the policy out in a week

DayTaskWho
1List every AI tool staff currently use, including personal accountsPractice manager
2Decide which go on the approved list; check business terms for eachPartner
3Fill in the template and remove clauses that do not applyPartner
4Add a line about AI to the engagement letter and privacy noticePartner
5Short team meeting: walk through the policy and answer questionsEveryone

If you are not sure which tools belong on the approved list, the AI tool finder narrows options by practice size and software.

Common mistakes to avoid

  • Writing a ban you cannot enforce. "No AI at all" pushes use underground. An approved list with clear limits works better.
  • Copying a large company's policy. A 30-clause corporate policy will not be read in a five-person practice.
  • Forgetting the in-ledger tools. Staff often do not think of categorisation suggestions or built-in assistants as "AI". They still need human review.
  • Never reviewing it. Vendor terms and plan names change. Put the review date in the diary.

For a structured way to train the team on the policy and the tools behind it, see our AI course for accountants.

This is covered in lesson 1 of our free-to-start course, Practical AI for UK accounting practices: read the free lesson.

Frequently asked questions

Does a sole practitioner need a written AI policy?

It is still worth having. A one-page policy is quick to write, helps you make consistent decisions, and gives you something to show a client or your professional body if asked how you use AI. If you take on staff or contractors later, it is already in place.

Can staff use ChatGPT or other general assistants under this policy?

Only if the practice approves a specific plan. Most practices that allow a general assistant do so under a business or team plan with a data processing agreement and admin controls, not personal accounts. Clause 3 rules out free and personal accounts for client work.

Do we need to tell clients we use AI?

There is no single rule that says you must use a particular form of words, but transparency is a core idea in data protection law and clients are increasingly asking. A short line in the engagement letter or privacy notice is a sensible, low-effort step.

Who should approve new AI tools?

One named person, usually a partner or practice manager. Having a single approver keeps the list consistent and means someone actually reads the supplier's terms before a tool is used with client data.

How often should the policy be reviewed?

Every 6 to 12 months, and whenever you add a new tool or a vendor makes a significant change to its terms. Put the review date in the policy itself.

Does the policy cover AI features inside Xero, QuickBooks or Sage?

Yes. In-ledger AI still produces suggestions that need checking, such as bank categorisation or drafted messages. The advantage is that the data is already held by a supplier you have terms with, so these tools are usually the easiest to approve.

What counts as an incident under clause 9?

Any client data entered into an unapproved tool, credentials entered anywhere, or an AI error that reaches a client, HMRC or Companies House. Report it internally the same day so the practice can decide whether it is a reportable personal data breach.

Keep reading

Free: a practice AI policy template

A ready-to-adapt policy for using AI tools with client data, in Word and PDF. Free with the 5-minute Friday brief on AI tools for accountants.

We email you the template and the Friday brief. Unsubscribe in one click.