ChatGPT and AI for accountants: 12 tasks that are safe to hand over

The short answer: AI is safe to use around the numbers in an accounting practice, not in them. It is good at drafting client emails, summarising long documents, turning notes into file notes and explaining a report in plain words. It must not give tax, accounting or financial advice, and nothing it writes should reach a client or a return without a qualified person checking it.

This post is not tax or accounting advice, and neither is anything ChatGPT tells you. Check your professional body's guidance and your data-protection obligations before putting client information into any AI tool.

Before you start: two rules

1. Use a business plan with clear data terms. Free consumer apps may use what you type to improve their models. Business and team plans generally let you switch that off; read the terms for the plan you actually buy.
2. Keep identifiers out. Remove names, tax numbers, account numbers and anything else that identifies a client before you paste text in, unless your firm has approved the tool for that data.

12 tasks AI can safely draft

Each of these produces a draft that a person reads before it goes anywhere.

1. Chasing missing documents. "Write a friendly reminder asking for the March bank statements and two receipts, second reminder, firm but polite." This is the most repetitive email in most practices, and the easiest to check.
2. Engagement and onboarding emails. First-contact emails that explain what you need from a new client and when, written from your own checklist.
3. Summarising a letter from a tax authority. Paste the letter (with identifiers removed) and ask for a three-line summary and the deadline it mentions. Then read the original yourself; the summary is for triage, not for the file.
4. Explaining a report in plain words. "Explain this profit and loss summary to a café owner who is not a finance person, in five bullet points." You check that the explanation matches the numbers.
5. Turning call notes into a file note. Messy notes in, a tidy dated file note out. Add anything the AI left out.
6. Drafting a client newsletter. General updates about deadlines or changes, written from sources you give it. Every date and threshold gets checked against the official source before it goes out.
7. Writing process documentation. Turn "how we do month-end for retail clients" from your head into a step-by-step checklist a new hire can follow.
8. Cleaning up spreadsheet formulas. Ask it to explain or fix a formula. Test the result on known figures before trusting it.
9. Categorisation suggestions. Given a list of transaction descriptions, it can suggest categories for you to confirm. Your accounting software's own rules usually do this better, so treat it as a second opinion.
10. Meeting agendas and follow-ups. An agenda for a year-end review, then a follow-up email with the actions agreed.
11. Job adverts and interview questions. For hiring a junior bookkeeper, written from your own list of must-haves.
12. Marketing drafts. Website copy, a Google Business Profile description or a short post about your services, in your voice.

5 things AI must not do

• Give tax, accounting or financial advice. General AI tools state wrong thresholds, rates and deadlines with total confidence, and they do not know a client's full position.
• Decide how to treat a transaction. It can suggest; a qualified person decides.
• Produce figures that go into accounts or returns without being recalculated from the source records.
• Send anything to a client unreviewed. Every draft is read by a person first.
• Hold client records. Keep them in your accounting and practice-management systems, not in chat history.

How to set up the review step

The habit that makes AI safe in a practice is simple: AI drafts, a person approves. In practice that means drafts land in your email's drafts folder or a shared document rather than being sent, one named person checks each type of output, and anything with a number in it is checked against the source record rather than against the AI's own summary.

Which tools?

Most practices need very little: one general assistant on a business plan, plus the AI features already inside the accounting software they use. Our AI for accountants page lists a recommended stack with prices we verified and dated, and the bookkeeping assistant page breaks down what an AI setup can and cannot take off a bookkeeper's plate.

If you would rather have the whole thing worked out for your practice, with an opportunity map, a costed stack and a 90-day roadmap, that is what the Dapols AI consultant does. It recommends tools and review steps; it never gives tax or financial advice.