How to Write Clear Instructions for an AI Assistant

AI assistants can draft, summarise, explain and organise information quickly, but the quality of what they produce depends heavily on what you ask for. A vague request tends to produce a vague answer. Using a simple, repeatable structure makes results more useful and, just as importantly, easier to check.

Start with context

An assistant only knows what you tell it in the conversation, along with whatever general knowledge it has. It does not automatically know your audience, your organisation or what you have already tried.

Begin with a few sentences of background. Explain who the work is for, what situation it relates to and any relevant details. For example, "I run a small accountancy practice and need to explain a new client portal to clients who are not comfortable with technology" gives far more to work with than "write about our portal".

State the task precisely

Next, say exactly what you want done. Use a clear action, such as draft, summarise, compare, list, rewrite or check. If there are several tasks, number them so each one can be handled and reviewed separately.

Be specific about scope. "Summarise this report" leaves many questions open. "Summarise this report in five bullet points for a manager who has not read it, focusing on decisions that need to be made" leaves far fewer.

Set constraints and boundaries

Constraints help the assistant avoid answers that are technically correct but unusable. They also make it clear what you do not want.

Useful constraints often include:

• Length, such as a word range or number of points

• Tone and reading level, such as formal, friendly or suitable for non-specialists

• Spelling conventions, such as British English

• Material to use or avoid, such as "only use the information in the text below"

• Things to leave out, such as speculation, jargon or invented examples

If accuracy matters, say so directly. Asking the assistant to rely only on the material you provide, and to say when something is uncertain or missing, reduces the chance of confident-sounding errors.

Describe the output you want

Tell the assistant what the finished result should look like. You might ask for a short email with a subject line, a table-free list of steps, a paragraph suitable for a website or a set of headings with notes beneath each.

Where possible, give a brief example of the format or style you are after. Even a single sample sentence can steer tone and structure more effectively than a long description. If the output will be pasted somewhere else, mention that, so the formatting suits its destination.

Make the results checkable

The most useful requests produce answers you can verify. Ask the assistant to separate facts drawn from your material from its own suggestions, or to point to which part of the source text supports each statement.

You can also ask it to list any assumptions it made. This makes gaps visible, so you can correct them rather than discovering them later. For calculations, ask to see the working, not just the final figure.

Remember that AI assistants can make mistakes, including stating incorrect facts with apparent confidence. Treat the output as a draft. Check names, numbers, dates and any claims before relying on them, particularly for anything that will be published, sent to customers or used in a decision.

Refine rather than restart

If the first response is not quite right, it is usually quicker to give specific feedback than to start again. Point to what worked and what did not: "Keep the structure, but make the tone less formal and remove the second paragraph." Small, targeted adjustments tend to converge on a good result.

If you find yourself writing similar requests regularly, save your structure as a template. Over time, you will learn which pieces of context and which constraints make the biggest difference for your type of work.

A simple structure to reuse

Putting these elements together gives a pattern you can apply to almost any request: context, task, constraints, output format and a request to show assumptions or sources. It does not need to be long. A few clear sentences under each heading is often enough.

The bottom line

Clear instructions lead to clearer, more reliable results. By explaining the context, defining the task, setting boundaries, describing the output and asking for checkable answers, you make an AI assistant far more useful and its mistakes far easier to catch. The assistant can save time, but responsibility for checking the final result still sits with you.