How to Use AI to Write Better Client Emails in Your Accountancy Practice

Client correspondence is one of the biggest time drains in any accountancy practice. Information request letters, deadline reminders, fee explanations, query responses: these are tasks that are important enough to do well, repetitive enough to feel like a grind, and numerous enough to eat a significant portion of the working week. AI can change that.

This article is part of Runbook's complete guide to AI prompts for UK accountants. It covers how to use AI tools to write better client emails faster, which types of correspondence lend themselves most readily to AI drafting, what a practical workflow looks like, and where to apply caution. If you want to know whether your practice is set up to use AI effectively, the free AI Readiness Scorecard gives you a personalised picture in under five minutes.

Why client emails are the right place to start with AI

When practices first explore AI, the instinct is often to look for it to do something impressive: summarise accounts, analyse tax positions, or automate complex workflows. Those applications exist, but they are not where most practices find the easiest, most immediate gains. Client email drafting is.

The reason is structural. Writing a client email involves three distinct activities: deciding what to say (the professional judgement), working out how to say it (the communication skill), and putting it into well-formed sentences and paragraphs (the writing). AI cannot replace the first activity and is only an aid to the second. But it is genuinely useful for the third, and in a busy practice with fifty or a hundred client emails going out each week, reducing the time spent on the writing part adds up quickly.

A further advantage is that the risk profile is low. Every email drafted with AI should be reviewed before it is sent, which means errors are caught before they reach anyone. This is a meaningful distinction from AI use cases where the output goes directly into a workflow without a human checkpoint. For client correspondence, the accountant remains in control at the point that matters.

Practices that have already worked through their bookkeeping automation will recognise this pattern. As covered in our guide to using AI to automate bookkeeping tasks, the highest-value AI applications in a practice tend to be the ones that reduce time on repetitive written work while keeping qualified oversight in place. Client emails sit squarely in that category.

The email types AI handles best

Not all client correspondence is equally suited to AI drafting. The tasks where AI adds the most value are those that follow a recognisable structure, where the professional judgement has already been applied, and where the communication need is primarily one of expressing a clear message politely and precisely.

Information request letters

Chasing clients for missing documents, receipts, bank statements, or signed returns is a persistent and time-consuming task in most practices. The underlying message is straightforward, but writing it in a way that is polite, specific, and firm without being off-putting takes more effort than it should. AI drafts these well when given a clear list of what is needed and any relevant context about timing.

A typical prompt for this task might look like this:

The output will need checking and personalising, but the structural and tonal work is done.

Deadline reminder emails

Self assessment deadlines, VAT return submissions, payroll filing dates: UK accountancy practices send large volumes of deadline reminder correspondence across the year. These emails follow a predictable structure and benefit from being consistent in tone. AI produces reliable drafts for these tasks and, when given a clear prompt, will include the right level of urgency for the timeframe involved.

Covering letters for accounts and tax returns

The covering letter that accompanies a set of accounts or a completed tax return is a genuine time sink. It needs to be accurate, reference the right figures, explain what the client needs to do, and be clear about any outstanding matters. AI handles the structural and written elements of this task well when given the key facts. The accountant provides the numbers and the conclusions; the AI turns them into a professional letter.

Responses to routine client queries

Many client queries follow familiar patterns: questions about payment on account, queries about what records to keep, questions about whether a particular expense is allowable. Where the answer is settled and the professional has already formed a view, AI can draft the explanation clearly and in plain English, saving the time it takes to write the same type of response from scratch for the eighth time that month.

Meeting follow-up emails

A structured email summarising what was discussed and agreed in a client meeting, with clear next steps, is valuable but rarely prioritised because it takes time to write after the meeting has already taken up the diary. AI drafts these well from a brief set of notes. The output will not be perfect, but it is substantially faster than composing the email from scratch.

How to prompt AI for client correspondence

The quality of what an AI produces for client emails is almost entirely determined by the quality of the instruction you give it. This is the part that most people underinvest in when they start, and it is the reason they sometimes conclude that AI is not useful for this type of task. A vague prompt produces a generic email. A specific, well-constructed prompt produces something that requires only light editing before it goes out.

A good prompt for client correspondence contains five elements.

1. A role or context

Tell the AI what perspective to write from. "You are writing on behalf of a UK accountancy practice" is sufficient for most purposes. If your firm has a particular sector specialism or the client is in a specific industry, include that too.

2. A specific task

Be precise about what the email needs to do. "Write a chaser email for missing documents" is weaker than "Write an email requesting three months of bank statements and a copy of the signed lease agreement, ahead of a 30 June year-end." The second version gives the AI enough to work with.

3. Relevant context

Include the information the AI needs: what is being requested or communicated, any relevant deadlines, the nature of the relationship (new client, long-standing client), and any specific points that need to be made. You do not need to include identifiable client information for most drafting tasks. The type of situation, the documents needed, and the timing are usually sufficient.

4. Tone and format instructions

Specify how the email should sound. "Professional but approachable, British English, no more than 200 words" gives the AI clear parameters. If your practice has a house style, describe it. If certain phrases are off-brand, say so.

5. Guardrails

Where relevant, tell the AI what not to do. "Do not include specific tax advice or figures" is a sensible guardrail for a general chaser email. "Do not use jargon the client is unlikely to understand" helps ensure the output is readable for a non-accountant audience.

For a broader grounding in how to structure prompts across all types of accountancy work, the complete guide to AI prompts for UK accountants covers the principles in detail with worked examples.

A practical workflow for your practice

The most efficient way to use AI for client correspondence is not to write a new prompt from scratch each time, but to build a small library of tested prompts for the email types your practice sends most often. Once you have a prompt that reliably produces a good draft for, say, a document request letter or a self assessment reminder, you can reuse it with minor adjustments for each new instance.

A simple workflow for getting started looks like this.

1. Identify your five most common email types. Look at your sent folder or ask your team which correspondence takes the most time to write. These are your starting points.
2. Write a prompt for each one. Use the five-element structure above. Test it two or three times and adjust until the output is consistently close to what you would write yourself.
3. Save the prompts somewhere accessible. A shared document, a note in your practice management system, or a dedicated prompt library works. The goal is that anyone on the team can find and use the prompt without having to create it again.
4. Review every output before sending. This is not optional. AI drafts are starting points, not finished emails. Check for accuracy, tone, and anything the prompt did not capture about the specific client or situation.
5. Iterate. When a prompt produces something that needs significant editing, update the prompt rather than accepting that the task is one where AI does not help. Usually, the issue is in the prompt, not the tool.

This workflow requires a small upfront investment in building the prompt library, but it pays back quickly. A team of five people each sending ten client emails per day, with each email taking an average of eight minutes to draft, represents roughly 3.3 hours of writing time daily. If AI reduces the drafting time on half of those emails by half, the saving is over 40 hours per month.

Data protection: what to check before you start

Before using any AI tool to draft client correspondence, data protection is an essential consideration for UK accountancy practices.

For drafting emails using non-identifiable context, such as a description of the situation, the documents needed, or the type of client, the GDPR risk is lower because you are not processing personal data. That said, practices should still consider confidentiality obligations, their own internal AI policy, and the current terms and default settings of whichever tool they are using. Terms vary by provider and can change, so it is worth checking rather than assuming.

If you intend to include identifiable client information in your prompts, such as a client's name, their financial figures, their National Insurance number, or any other personal data, you should use a business-grade tool with appropriate contractual and privacy controls in place. Free-tier terms and data-use policies vary across providers, so you cannot rely on a blanket assumption about how any particular tool handles inputs.

Business plans from the major providers typically offer data processing agreements, stronger privacy commitments, admin controls, and default protections against using inputs for model training. ChatGPT Business and Enterprise, Microsoft 365 Copilot or Copilot services with enterprise data protection, and Anthropic's Claude Team or Enterprise plans are examples of this tier. Having the right plan in place is an important step, but it does not make a practice automatically compliant. Your own governance, internal policies, and data protection obligations remain your responsibility.

For most day-to-day drafting tasks, the simplest approach is to write prompts that describe the situation rather than naming the client. "A small retail business, sole trader, year-end 31 March, missing three months of bank statements" gives an AI enough to produce a useful draft without requiring any identifiable data to be included.

Frequently asked questions