Skip to main content

Industry guide

AI for Accountants

A calm, practical guide for an accounting practice: where AI genuinely speeds the drafting, summarising and research — and the bright lines it must never cross (the filing, the sign-off, and any figure you haven't checked). Plain English, honest limits, cited sources.

JR
Jon RossFounder, vocabotics — 15 years building safety-critical AIField guide · 5 min read · reviewed 2 July 2026
Verified by a human. Drafted with AI, verified by a human. Jon Ross, 2 Jul 2026
Living document. Reviewed 2 Jul 2026
Shipped & measuredDownload the guidePDF · 967 KB

Most AI advice for accountants is written by people who have never reconciled a VAT control account. This one is written for a working practice — the drafting, the summarising, the research, and the hard question of where a language model belongs nowhere near the work. No hype. No selling. Just where it saves you time, and the bright lines it must never cross.

The three bright lines

AI never does the filing, never provides the sign-off, and never hands you a figure to trust unchecked. It drafts the letter, summarises the guidance and lays out the working — you verify every number against the source and put your name to the result. Cross those lines and you have swapped a control for a risk.

What "AI" here actually means

An assistant you prompt in plain language — ChatGPT, Claude, Gemini, or a document tool built on them. It drafts, summarises, reformats and explains text. It is not your bookkeeping software, not HMRC's systems, and not a calculator you can trust blind: language models can produce a confident, wrong number. That single fact governs everything below. (New to these tools? Start with our Beginner's Guide to ChatGPT.)

Where it genuinely helps

  • Correspondence. Chaser letters, engagement wording, "here's what we need for your return" emails — drafted in seconds, edited to your house style, signed as yourself.
  • Summarising. Turn a fifteen-page guidance note or a rambling client email into a dated action list — then confirm the points you'll rely on.
  • Research, verified. Ask for an orientation on a rule, then follow every claim to HMRC or the standard before it reaches a client.
  • Working papers. Lay out a computation as a tidy schedule with the arithmetic shown — so a human can check each line.
  • Explaining to clients. Translate "input tax" or "payments on account" into plain English a client actually reads.

Drafting client letters, engagement notes and email

AI reliability: Go

Explaining a rule, then verifying against HMRC / the standard

AI reliability: Amber

Producing the figures on a return, or filing it

AI reliability: Stop

A real worked example: VAT, checked to source

AI can lay a VAT schedule out and do the plain arithmetic; you tie each figure to the ledger, apply the judgement (partial exemption, fuel scale charges, the reverse charge) and file it yourself. For a quarter with £48,250 standard-rated sales and £16,400 of purchases: output VAT is £9,650 (Box 1), input VAT £3,280 (Box 4), and the net due to HMRC is £6,370 (Box 5). The assistant can format that working paper in seconds — but every figure ties back to the day book, and the submission goes through your Making Tax Digital software, signed by you.

Where a human must stay

This is the part that matters most, so here it is straight.

  • The filing and the submission are yours. They go through compliant software; the responsibility to HMRC rests with the person who signs.
  • Never provide a figure you haven't verified. Models produce confident, wrong numbers. Every figure ties to source, in your systems.
  • The sign-off is a named human's. Judgement and accountability are the whole engagement — a chatbot cannot hold them.
  • Never paste client data into a public AI tool. Confidentiality and data-protection duties apply; use controlled, private tooling only.
The accuracy of a return and the duty to file it correctly rest with the person who signs it — AI is not a defence for an error. Keep AI use inside tooling you control and assess.
Technology & AI guidance for membersICAEW (as of July 2026)
Client financial data is exactly what you must not hand to an outside company casually; align AI use with a recognised risk framework and your data-protection duties before it touches client work.
AI Risk Management Framework (AI RMF 1.0)NIST (as of July 2026)

A safe first task, this week

Don't start with a client's numbers. Start with words. Take a piece of published HMRC guidance and ask the assistant to summarise it into a plain client note — then check it against the source. You'll feel the time saved and see, first-hand, where it drifts. Do that for a fortnight and you'll know exactly where an assistant belongs in your workflow: on the drafting and the explaining, never on the filing, the figures or the sign-off.

The honest bottom line

For a practice, AI is a fast junior, not a signatory. It'll clear a chunk of the drafting, summarising and explaining that eats your chargeable time, and help you make sense of dense guidance. It will not compute your returns, hold your judgement, or carry your responsibility to HMRC — those stay with you and your professional standard. Use it for the words, verify everything that matters against the source, and it earns its place without ever putting a client, or your practice, at risk.

Sources

  1. VAT: detailed information (returns, MTD, rates)HM Revenue & Customs · as of July 2026
  2. Making Tax Digital for VATHM Revenue & Customs · as of July 2026
  3. Self Assessment tax returns — deadlinesHM Revenue & Customs · as of July 2026
  4. Technology & AI guidance for membersICAEW · as of July 2026
  5. AI Risk Management Framework (AI RMF 1.0)NIST · as of July 2026
  6. Guidance for organisations (data protection)UK Information Commissioner's Office · as of July 2026

    We use cookies.