Writing effective AI prompts

28 July 2025

Last updated: 26 August 2025

David Menzies
Director of Practice, ICAS

Artificial Intelligence (AI) is quickly becoming a helpful assistant in the accountancy profession. From drafting emails to analysing data, AI tools can save time and improve productivity – but only if you ask the right questions and spend time ‘training’ the tools in question. The instructions you give an AI, known as AI prompts, determine how useful the AI’s response will be. In other words, better prompts lead to better answers. This guide explains what AI prompts are, why they matter, and how to construct them effectively.

What Are AI prompts?

AI prompts are the instructions or questions you give to a system, such as Co-pilot, Chat-GPT, Claude, Gemini, or other AI tools, to get a result. In practice, a prompt is just the input text you type or say to an AI tool like a chatbot or writing assistant. For example, asking “Give me a summary of the latest IFRS update” is a prompt instructing the AI what you want.

Think of it as instructing a colleague on what task you need done. The clearer and more specific your instruction, the closer the result will match your expectations. A prompt can be a question, a request, or a set of guidelines. It tells the AI what to do and how to do it. To make the result even better and more useful, if you want the answer in a certain style or focusing on certain details, you can include that in the prompt as well.

Why do AI prompts matter?

A well-constructed AI prompt yields a more useful AI response.

Good prompts are crucial because the AI’s output is only as good as the question or command you give it. A poorly worded prompt can lead to irrelevant answers, factual errors, or an overly generic response. On the other hand, a carefully crafted prompt can unlock extremely useful results, helping you work faster and smarter.

Consider that generative AI tools in accounting can do things like summarise complex tax legislation, draft client emails, analyse financial data, or even suggest business improvements. The catch is that the AI needs clear guidance on what exactly it must do. This skill of writing prompts – sometimes called prompt crafting or prompt engineering – is becoming crucial in the accounting profession.

In day-to-day work, using the right prompt can turn hours of work into minutes by getting quick, targeted answers from an AI. In short, learning to write good prompts can save you time, reduce frustration, and increase the value you get from AI.

But you don’t have to be a tech expert to write a great prompt. It doesn’t require programming or deep AI knowledge. In fact, prompt-writing is more about clarity and creativity – being clear on what you need, and how to ask for it. Qualities like specificity, clarity, and a bit of experimentation are far more important than technical jargon. This means anyone can master the art of prompting with a little practice.

Components of a good AI prompt

A well-constructed prompt usually has a few key elements:

Role: Tell the AI who it is by assigning it a role to play.

- Why it matters: When the AI “knows” who it is supposed to be it can tailor the response appropriately.

- Example: "You are a tax advisor to UK owner managed businesses" or “You are an insolvency practitioner in the UK specialising in personal insolvency.”
Context: Set the scene. Tell the AI the context or what perspective to take.

- Why it matters: Context guides the AI to produce an answer at the right level of detail and tone.

- Example: “I want to share with a client...” or “We are writing to clients to update them on changes to…”
Mission: State exactly what you need. This is the core instruction of your prompt. If you aren’t clear about what you want, the AI might give you a wandering or off-target answer.

- Why it matters: AI can’t infer your true intent from a vague question. Being up front about your aim focuses the AI’s efforts.

- Example: “Summarise the key changes in UK corporate tax law for 2025 that would impact a small business.” or “Summarise the meeting held with the client to send to them in an email.”
Specific details: Provide relevant details and scope. The best prompts include the specific information or context you care about.

- Why it matters: AI works best with concrete information – it can zero in on what’s relevant. If you leave it too open-ended, you might get a very general answer.
- Example: “Give me five budgeting tips for a freelance consultant in the UK who struggles with irregular income.” Here you’ve specified the audience (freelance consultant), the location (UK, which might affect tax or currency context), and the format (five tips). The AI now has clear boundaries for its answer.
Format and tone: Explain how you want the answer delivered. If you prefer the style, length, or format of the answer, include that in your prompt.

- Why it matters: The AI can present information in many ways – a bullet list, a formal letter, an executive summary, etc. If you specify a format, you’ll get an output that’s easier to use immediately. Tone is important if the output is client-facing.
- Example: If you need a brief overview, you might say “Explain the changes in a short, bulleted list.” Or for tone, “Draft a polite and professional email to a client explaining…”.
Example: To help make the output even better, give a template, model or previous output on which to base its response.

- Why it matters: It can reduce the number of further prompts or refinements that might be needed.
- Example: "Here's an example of a report that we send to clients. Use this structure for this report." Or “Here’s the firm's tone of voice and style guide to use.”
Constraints or additional guidance: Mention any dos or don’ts. In some cases, you might want to tell the AI to focus on or avoid something.

- Why it matters: This helps prevent unwanted content and keeps the answer on track.
- Example: “Without using technical jargon, in about 2-3 sentences, …” Here you’ve constrained jargon (to keep it simple) and length (so the answer is concise). Or “Use UK English spelling and grammar” to help save reformatting the response.

Putting all this together you can build a very effective prompt.

Example: “[Role] I am a senior manager in an audit firm. You are an audit AI assistant. [Format] In two to three paragraphs [Objective] explain the concept of going concern, [Tone] in plain English as I might to a junior team member, [Additional guidance] focusing on how it affects a company’s financial statements. List 3 - 4 example indicators that a company might be a going concern risk. [Constraints] Use references to UK accounting and auditing standards. Use UK English spelling and grammar. Don't make anything up.”

Such a prompt gives the AI a clear mission and boundaries, increasing the chance of an accurate answer on the first try.

Other tips and common issues to avoid

Some additional tips and common mistakes include:

Avoid vague requests: Clarity is king. Write your prompt in plain language and be as specific as necessary. Vague or overly generic requests will produce broad, unfocused answers. If you need something specific, state it clearly (as shown in the examples above). Remember, the AI cannot read your mind or guess what you really want, so spell it out.
Mind your language (literally): Double-check grammar and wording in your prompt. While AI models are pretty good at understanding imperfect input, a typo or ambiguous phrasing can confuse the system. For example, “advise” vs “advice” or a missing “no” can flip the meaning. Also, use terms the AI is likely to know. Stick to standard accounting terminology. If the question is complex, try breaking it into simpler sentences.
Provide feedback: Don’t worry if you don’t get a perfect answer on the first try. It’s common to tweak a prompt after seeing the first response. Think of it as a conversation: You can ask the AI to clarify or provide more detail on something it answered, or you can rephrase your prompt and try again if the answer wasn't quite what you wanted. For example, you can follow up with “Thanks, now give me an example,” or “Please make it shorter.” Iteratively refining your prompt can lead to a much better final answer. It’s not “one and done” – feel free to prompt again with adjustments.
Ask for multiple options: If you are looking for ideas or creative output, you can ask the AI for several alternatives at once. For instance, “Give me three alternative wordings for this conclusion sentence” or “Suggest two different approaches to solve this tax issue.” This way, you can pick the best of the AI’s suggestions. It’s sometime useful to get multiple drafts – e.g., “provide three versions of a client letter” – to compare and choose the one that fits best.
Avoid confidential information: Never put sensitive client or firm data into a prompt when using public (open) AI services. Remember that what you input might be stored or used to train AI models. Treat an AI like you would any external service - don’t share private financial figures, personal data, or anything confidential. If you need analysis on confidential data, use specialised enterprise AI tools that guarantee privacy, or anonymise the information first. Always follow your organisation’s data protection guidelines.
Verify outputs: AI can sometimes produce authoritative-sounding answers that are incorrect or incomplete. This is often called a “hallucination” in AI jargon – the system might make up a fact that looks plausible. Some of this can be avoided by putting constraints in the prompt, for example "Only use the information and sources provided, do not make anything up." Always apply professional scepticism. Use AI outputs as a starting point, not absolute truth. Double-check important facts and calculations in any AI-generated content. If the AI drafts a memo or report for you, review it carefully as you would any work produced by a more junior member of staff. Make sure conclusions are supported and the advice complies with regulations and ethical standards. Ultimately, you as the professional are responsible for the final output.

Conclusion

AI tools have enormous potential to lighten the load – from quickly crunching through background research to generating first drafts of documents or emails. The prompt is the key to unlocking that potential. Think of writing AI prompts as a new extension of your communication skills. Just as you tailor your questions when consulting a colleague or an expert, tailoring your prompts when consulting an AI makes a world of difference.

With practice, you can learn to construct prompts that are clear, specific, and contextual, providing results that are accurate and useful. In the collaborative future of accounting – where human expertise and AI assistance work together – prompt crafting is a skill that will set you apart. It enables you to harness AI efficiently while keeping the quality and integrity of your work. By mastering how to ask, you empower the AI to deliver the answers you need, amplifying your own capabilities and those of your firm.