Most accountants who have tried AI followed the same path. They opened ChatGPT, typed something vague like "help me with a tax return," received a generic and largely useless response, and concluded the technology was overhyped. The problem was not the tool. The problem was the prompt.
A well-written prompt turns a general-purpose AI tool into something that behaves like a reasonably competent junior who already knows your practice area, your client type, and the tone you expect. A poorly written prompt produces the digital equivalent of a work experience student guessing what you want.
This guide covers everything a UK accountancy practice needs to know about AI prompts: what they are, how to write them well, which areas of your practice benefit most, and how to build an internal prompt library that your whole team can use. Whether you are a sole practitioner or a practice manager overseeing a team of twenty, the principles are the same.
This is part of Runbook's complete series on AI for UK accountancy practices. If you are still deciding whether to introduce AI tools at your firm at all, the companion article on AI implementation for UK accountancy practices is the right starting point.
Disclaimer: Nothing in this article constitutes legal or data protection advice. Your practice should take qualified advice on its specific obligations before processing client information through any AI tool. Where example prompts are given in this article, client data or confidential information should never be entered into an AI tool before your data protection adviser or legal advisers have confirmed it is compliant to do so.
1. What is an AI prompt (and why does it matter)?
An AI prompt is the instruction you give to a tool like ChatGPT, Microsoft Copilot, or Claude. It is the text you type into the chat window. The quality of the response you receive is almost entirely determined by the quality of that instruction.
Think of it this way. If you asked a new starter at your practice to "sort out the client files," you would get a blank stare or, worse, a creative interpretation of what "sort out" means. If you said "file the signed engagement letters for the ten clients listed on this spreadsheet into their respective folders on the shared drive, surname first, and flag any that are missing a signature," you would get the result you wanted.
AI tools work the same way. The more specific and contextual your instruction, the more useful the output. This is not a limitation of the technology. It is how language models are designed to work: they predict the most helpful response based on what you give them.
For accountancy practices, this matters because the tasks where AI can help (drafting emails, summarising documents, generating first-draft reports, creating template letters) all depend on clear input. A two-minute investment in writing a better prompt can save ten or fifteen minutes on the task itself. That maths adds up quickly across a team.
Before diving in, it is worth checking whether your practice is ready to start using AI in a structured way. Our free AI Readiness Scorecard takes five minutes and gives you a clear starting point based on your firm's size and current workflows.
2. The anatomy of a good prompt for accountancy work
Through testing hundreds of prompts across real UK accountancy tasks, a consistent pattern emerges. The prompts that produce usable first drafts almost always contain the same five elements.
Role
Tell the AI what role it is playing. "You are an experienced UK practice accountant writing to a small business client" sets a completely different baseline than no role instruction at all. The role shapes the vocabulary, the level of formality, and the assumptions the AI makes about what the reader knows.
Task
State exactly what you want produced. Not "write something about VAT" but "draft a 120-word email explaining to a client why their flat rate percentage has changed and what the financial impact will be for the next quarter." Specificity eliminates ambiguity.
Context
Provide the details that a knowledgeable human would need to do the task well. Client name, company type, the specific situation, any relevant deadlines, and the relationship history. The AI cannot read your mind or your practice management system. What you leave out, it will either omit or invent.
Tone and format
A short phrase is enough. "Professional but approachable" produces noticeably different output from "formal and cautious." If you need a specific format (bullet points, numbered steps, a table), say so. If you need a word limit, state it. These constraints are among the most effective levers you have.
Guardrails
Tell the AI what not to do. "Do not invent any figures." "Do not reference US tax law." "Do not use the word delve." Negative instructions are surprisingly powerful and prevent the most common failure modes.
You are a UK practice accountant writing to a sole trader client.
Draft an email to James Carter, who runs a small plumbing business. He has not yet submitted his Self Assessment records for the year ended 5 April 2026. We need his bank statements, a summary of business expenses, and mileage records. The filing deadline is 31 January 2027 but we need records by 31 July 2026 to begin preparation.
Tone: friendly but clear. He is a good client but tends to leave things late.
Keep it under 130 words. Do not invent any figures. Do not reference any specific tax amounts.
That prompt takes roughly sixty seconds to write. The resulting draft will need minor adjustments at most.
Skip the learning curve. 50 tested prompts covering client emails, tax work, meeting notes, and more. Instant download.
3. Where prompts save the most time in a UK practice
Not every task in an accountancy practice benefits equally from AI. The sweet spot is work that is repetitive in structure, moderate in complexity, and low in regulatory risk. Here are the areas where UK practices are seeing the most consistent time savings.
Client correspondence
This is the single most productive area for most practices. Document request emails, deadline reminders, fee explanations, cover notes for tax returns, and routine follow-ups all follow predictable patterns. A team of ten can recover several hours per week by using prompt templates for these tasks alone.
For a full walkthrough of how to write and use AI prompts for client emails, with worked examples covering the most common email types and a practical workflow for building a prompt library, see our dedicated guide to using AI to write better client emails in your accountancy practice.
Report writing and management commentary
First drafts of management accounts commentary, year-end summaries, and client-facing narrative reports are excellent candidates. The AI handles the structure and phrasing; your accountant provides the judgement and the numbers.
Engagement letters and standard documents
Generating first drafts of engagement letters, terms of business, and other template-based documents. The AI produces the structure and boilerplate; you review, adjust, and add practice-specific terms. For a practical walkthrough of this specific task, including worked prompts for both limited company and sole trader clients, a section-by-section breakdown of what a well-drafted engagement letter needs to cover, and a review checklist to work through before any letter goes out, see our dedicated guide to using AI to draft engagement letters.
Tax return cover letters
The covering letter that accompanies a filed tax return is one of the most formulaic documents in accountancy. Client name, return type, key figures, payment dates, deadlines. AI handles these in seconds with the right prompt.
Tax return season also generates significant written work beyond the cover letter itself: information request emails before the return can begin, polite but persistent chasing when clients are slow to respond, and plain-English explanations of what the figures mean. Each of those tasks has its own prompt requirements. Our guide to AI prompts for tax return work covers all of them with ready-to-use prompts you can adapt immediately.
Internal documentation
Staff handbook updates, process notes, internal memos, and training materials. These tasks often get pushed down the priority list because they do not generate revenue, but they matter for practice management. AI brings the effort cost low enough that they actually get done.
Client onboarding
Welcome emails, initial information request checklists, and onboarding workflow documentation. A well-prompted AI can produce a complete onboarding email sequence in minutes.
Where to start: If you are new to AI prompts, begin with client correspondence. It is the highest-volume, lowest-risk area and the one where time savings are most immediately visible. One well-tested document request template, used consistently across the team, is worth more than ten experimental prompts across ten different task types.
4. Worked examples: prompts you can use today
Below are three prompts covering different practice areas. Each follows the five-element structure outlined in Section 2. You can copy and adapt these immediately.
Example 1: Self Assessment deadline reminder
You are a UK practice accountant writing to individual tax clients.
Draft a batch email to be sent to all personal tax clients who have not yet submitted their records for the 2025/26 tax year. The Self Assessment filing deadline is 31 January 2027 for online returns. We need their records by 30 September 2026 to allow adequate preparation time.
Mention: P60 or equivalent income confirmation, bank interest certificates, rental income details (if applicable), Gift Aid records, and pension contribution statements.
Tone: professional, helpful, not nagging. This is a first reminder, not a chase.
Keep under 180 words. Sign off as "The team at [Practice Name]." Do not invent any tax figures or thresholds.
Example 2: Management accounts commentary
You are a UK accountant preparing management accounts commentary for a client.
The client is a limited company operating three coffee shops in Birmingham. Revenue for the quarter ended 31 March 2026 was £187,400, up 8% on the same period last year. Gross margin has fallen from 62% to 58%, driven by increased supply costs. Staff costs are up 12% due to the April 2025 National Living Wage increase. Net profit before tax was £14,200, down from £19,800 in the prior year quarter.
Draft a 200-word commentary summarising performance and highlighting the two key issues (margin compression and staff costs). Use plain language the business owner will understand. Do not offer advisory recommendations. Do not invent any additional financial data.
Example 3: New client welcome email
You are a UK practice accountant welcoming a new client.
Draft a welcome email to Rachel Edwards, the director of a newly incorporated limited company called Bright Arc Consulting Ltd. She has just signed our engagement letter for annual accounts preparation, corporation tax, and payroll for one employee (herself). Her year-end is 31 March.
Include: a thank you, confirmation of the services we will provide, what she needs to send us in the first month (bank account details, incorporation documents from Companies House, HMRC UTR once received), and how to get in touch with questions.
Tone: warm, clear, reassuring. She is a first-time director and unfamiliar with accountancy processes.
Keep under 200 words. Do not give tax advice. Sign off with a first name only.
These three examples cover a significant proportion of the written output most practices produce. Adapt the client details, adjust the constraints, and they are ready to use.
5. How to build a prompt library for your firm
Individual prompts are useful. A prompt library is transformational. The difference is the shift from one person experimenting to a whole team operating from a shared, tested set of instructions.
A prompt library does not need to be complicated. At its simplest, it is a shared document (or folder) containing your best prompts, organised by task type, with notes on how to customise each one. Here is a practical approach.
Start with your highest-volume tasks
Identify the five to ten email and document types your practice produces most often. Missing document requests, deadline reminders, fee quotes, onboarding emails, and meeting follow-ups are common starting points. Write a prompt template for each.
Use a consistent format
Every prompt in the library should follow the same structure: role, task, context (with placeholders for the details that change each time), tone, format, and guardrails. Consistency makes the library intuitive. Anyone on the team can pick up any prompt and use it without additional explanation.
Store it where your team already works
The best prompt library is the one people actually open. If your team lives in Microsoft 365, use a shared OneNote notebook or a Word document on SharePoint. If you use Google Workspace, a shared Google Doc works fine. If you use practice management software with a document library, put it there. Do not create a new system for this.
Review and improve monthly
Ask your team at the end of each month: which prompts are you using most? Which ones need rewriting? What new task types should we add? A prompt library that is reviewed regularly will steadily improve. One that is written once and forgotten will be abandoned within weeks.
Skip the setup: Our AI Prompt Pack for UK Accountancy Practices includes 50 tested prompts covering client emails, HMRC correspondence, meeting notes, internal documents, and more. Copy them into your shared folder and your team can start using them this week.
Covers tool selection, data policy, staff rollout, and client communication. Step-by-step format designed for UK practices of 5 to 50 staff. Instant download.
6. Common mistakes and how to avoid them
There are common mistakes when it comes to AI adoption. Knowing them in advance saves considerable time and frustration.
Writing prompts that are too vague
"Write an email about a tax return" will produce something generic and unhelpful. Every missing detail is a gap the AI will fill with assumptions, and those assumptions are frequently wrong. The fix is simple: include the specific client context, the specific purpose, and the specific constraints every time.
Trusting the output without review
AI produces confident, fluent text. It also occasionally fabricates deadlines, invents HMRC thresholds, and hallucinates details that do not exist. Every AI-generated document must be reviewed by a qualified person before it leaves the practice. This is not a trust issue with the technology; it is standard quality assurance, the same as reviewing a junior's work.
Starting too ambitiously
Practices that try to use AI for everything on day one almost always give up within a fortnight. Start with one task type. Get comfortable with it. Build confidence across the team. Then expand. The practices that succeed with AI are the ones that treat adoption as a gradual process, not a one-off project.
Ignoring tone and length instructions
Without explicit tone and length instructions, AI defaults to verbose and generic. Adding "keep under 130 words" and "professional but warm" takes five seconds and improves the output dramatically. These are the highest-leverage additions you can make to any prompt.
Using consumer-tier accounts for client work
Free and consumer versions of AI tools often have weaker data protection terms than business accounts. Some use your inputs to train their models. If you are processing anything that touches client data, you need to be on a business-tier plan and you need to have read the provider's data processing agreement. More on this in the next section.
7. Data protection considerations for UK practices
This is the area where UK practices rightly exercise the most caution, and where most generic AI guides are weakest. The legal framework that governs how you process personal data through AI tools is the UK GDPR and the Data Protection Act 2018. It is enforceable, and it applies to AI.
Runbook does not provide legal or compliance advice. What we can do is set out the questions you should be discussing with your data protection adviser before your team starts using AI prompts that involve client information.
- What happens to the data you enter? Does the AI provider store your prompts? Are they used for model training? Where are the servers located? The answers differ significantly between providers and between account tiers.
- Do you have a lawful basis? Processing client data through a third-party AI tool requires a lawful basis under UK GDPR. Legitimate interests is the most commonly cited, but this needs to be assessed and documented.
- Is your privacy notice up to date? If you have started using AI in your workflows, your clients should know. Reviewing your privacy notice is a sensible early step.
- Can you minimise the data in prompts? Using "Client A" instead of a real name, omitting UTR numbers, and stripping out identifiable details where possible reduces your data protection exposure without significantly affecting the quality of the AI output.
- Are you using a business-tier account? Most AI providers offer stronger data protection commitments on their business plans, including no-training guarantees and data processing agreements.
Many practices start by using AI for structure and tone only, entering the type of communication and the general scenario without including any client-identifiable information. This approach sidesteps the data protection questions entirely while still delivering significant time savings. Once a formal AI usage policy is in place, the scope can expand.
For a structured approach to these questions, our AI Implementation Checklist includes a dedicated GDPR section with a questions framework built specifically for UK accountancy practices.
Important: Nothing in this article constitutes legal or data protection advice. Your practice should take qualified advice on its specific obligations before processing client information through any AI tool.
8. Next steps
If you have read this far, you now have a solid grounding in how AI prompts work, where they save time in a UK practice, how to write them well, and what to watch out for. The question is what to do next.
The simplest starting point is to pick one task. Choose the highest-volume, lowest-risk email type your practice sends (document requests are usually the best candidate), write a prompt using the five-element structure from Section 2, and try it. Spend one week testing it. Ask a colleague to try it. Adjust the prompt based on what you learn. That single week will tell you more about the practical value of AI for your practice than any amount of reading.
From there, the path is straightforward: add more prompt templates, build them into a shared library, agree a review process, and expand to new task types as your team's confidence grows.
The full approach to selecting tools, writing a data policy, and managing a practice-wide rollout is covered in the Runbook guide to AI implementation for UK accountancy practices. For help choosing between AI tools, see the best AI tools for UK accountants.
The free AI Readiness Scorecard takes five minutes and gives you a personalised starting point based on your practice's size, tools, and workflows.
Frequently asked questions
What is an AI prompt?
An AI prompt is the instruction you type into a tool like ChatGPT, Microsoft Copilot, or Claude. The quality of the response you receive is almost entirely determined by the quality of that instruction. The more specific and contextual your prompt, the more useful the output.
How do I write a good AI prompt for accountancy work?
A good prompt for accountancy work contains five elements: a role (telling the AI what perspective to write from), a specific task, context including client details and any relevant deadlines, tone and format instructions, and guardrails telling the AI what not to do or invent. All five work together to produce usable first drafts rather than generic filler.
Is it safe to use AI prompts with client data?
This depends on the AI tool and the account tier you are using. Free tiers generally do not include the data processing agreements required under UK GDPR for processing client personal data. Business-tier accounts from providers such as OpenAI and Anthropic include data processing agreements and commitments that your data is not used for model training. Your practice should take qualified data protection advice before processing client information through any AI tool. Runbook does not provide legal advice.
What is a prompt library?
A prompt library is a shared collection of tested AI prompts, organised by task type, that your whole team can use. It is the difference between one person experimenting and the whole practice benefiting from consistent, tested instructions. At its simplest, it is a shared document containing your best prompts with placeholders for the details that change each time.
What tasks benefit most from AI prompts in an accountancy practice?
Client correspondence (document requests, deadline reminders, fee explanations), management accounts commentary, engagement letters, tax return cover letters, internal documentation, and client onboarding emails are the highest-value areas for most UK practices. The sweet spot is work that is repetitive in structure, moderate in complexity, and low in regulatory risk.
Do I need specialist AI tools, or will ChatGPT do?
For most of the tasks covered in this guide, a well-prompted general-purpose tool such as ChatGPT on a paid Team plan, or Claude Pro, is sufficient. Specialist accountancy AI tools exist for specific tasks such as document extraction and bookkeeping automation, but for drafting, summarising, and template-based written work, general-purpose tools deliver strong results when prompted well. For an overview of the available options, see our guide to the best AI tools for UK accountants.
How long does it take to see time savings from AI prompts?
Most practices see measurable time savings within the first week of consistent use on a single task type. The compounding effect comes from building a prompt library: once a team of ten is all working from the same tested templates, the aggregate time saving becomes significant. Practices that take a structured approach rather than ad hoc experimentation typically see sustainable adoption within four to six weeks.
- 20 ChatGPT Prompts for UK Accountants 20 ready-to-use prompts covering client emails, HMRC letters, meeting notes, and tax work. Copy, paste, and adapt.
- How to Use AI to Write Better Client Emails in Your Accountancy Practice Practical prompts, a step-by-step workflow, and data protection guidance for the highest-volume writing task in most practices.
- The Best AI Prompts for Tax Return Work: A UK Accountant's Guide Ready-to-use prompts covering cover letters, information requests, client chasing, and query responses for every stage of tax return season.
- Using AI to Draft Engagement Letters: Prompts and Examples for UK Accountants Worked prompts for limited company and sole trader engagements, a breakdown of what every letter needs to cover, and a pre-send review checklist.