How to brief an AI tool so it does genuinely useful work

Most disappointing AI output starts with a disappointing instruction. Ask a vague question and you get a vague, generic answer that could have been written for anyone. Brief it the way you would brief a capable contractor, with context, a clear task and an example of good work, and the results change completely.

Brief it like a capable new starter

Imagine a talented new starter on their first morning. They are well read, quick and keen, but they know nothing about your business, your customers or how you like things done. If you said "write something about our lamps", you would get something, but probably not what you wanted.

AI tools are in exactly that position every time you open a new conversation. Everything they need to do a good job has to be in the brief. The good news is that a few extra sentences make an enormous difference.

The five parts of a good brief

You do not need special phrases or tricks. A good brief covers five things:

Context. Who you are, what the business does and who your customers are.
Task. Exactly what you want produced, in one clear sentence.
Audience. Who will read or use the result, and what they already know.
Constraints. Length, tone, format, and anything to include or avoid.
Example. A sample of what good looks like, if you have one.

Not every request needs all five, but the more that is riding on the result, the more of them you should include.

A weak brief and a strong one

Here is a typical first attempt:

Write a product description for our desk lamp.

And here is the same request, properly briefed:

We run a small online shop selling home office furniture to people who work from home in the UK. Write a product description for our adjustable LED desk lamp. It has five brightness levels, three colour temperatures from warm to cool, a USB-C charging port in the base and a weighted base that stops it tipping. Our customers care about comfort during long working days more than about style. Write about 120 words in a friendly, plain-English tone, with short sentences and British spelling. Lead with the benefit of less eye strain, not the features. Avoid words like "revolutionary" or "game-changing". Here is a description we were pleased with for another product, so you can match the style: [paste example].

The first brief leaves the tool to guess everything, so it falls back on the most average lamp description imaginable. The second tells it who is buying, what matters to them, what the product actually does and what good writing looks like for this shop. The result will need far less editing.

Three more briefs worth borrowing

These follow the same five-part structure. Change the details to suit your business and keep the shape.

For ad headlines:

We sell refurbished office chairs online to people setting up home offices in the UK. Our customers worry about back pain and about buying second-hand. Write 15 headlines for a search advert, each under 30 characters. Give me five built around comfort, five around trust in refurbished quality and five phrased as a question the customer might be asking themselves. Avoid exclamation marks and words like "best" or "amazing". Here are our three best-performing headlines so far: [paste].

Why it works: it includes a real constraint (the character limit), real customer worries and three distinct angles, so you can test themes against each other rather than random lines.

For a customer reply:

You are helping me reply to a customer email. Our policy is that unopened items can be returned within 30 days, and we cover return postage for faulty items only. The customer's email is below. Write a reply that is warm and apologetic about the delay they experienced, explains the policy clearly in plain English and offers a next step. Keep it under 150 words. Do not promise anything that is not in the policy. [paste email]

Why it works: the policy is spelled out, so the tool cannot invent a more generous one, and there is an explicit instruction not to promise anything beyond it.

For a spreadsheet:

I have a sales export in Google Sheets. Column A is the order date, column B is the product name, column C is the quantity and column D is the order value. Write a formula that shows total order value by month for the last 12 months, and explain in two sentences how it works. Then suggest one chart that would make the trend easy to see.

Why it works: it describes the exact layout of the data and asks for a formula rather than an answer, so the spreadsheet does the arithmetic and the result can be trusted.

Give it your own material

The single biggest improvement most people can make is to paste in their own raw material. Customer reviews, the transcript of a sales call, last month's best-performing advert, your notes from a meeting, a spreadsheet export: all of it gives the tool something specific to work from rather than generic knowledge.

Customer reviews are particularly valuable for marketing copy. They contain the exact words real buyers use to describe their problems and what they love, and those words almost always outperform anything a marketer would invent.

Ask for options, not one answer

AI is fast, so make use of it. Instead of one headline, ask for fifteen, in three different angles: one set built on a problem, one on a result and one on curiosity. Instead of one email, ask for three versions at different lengths. You can then pick the strongest, combine the best parts of several, or spot an angle you would never have thought of.

Let it ask you questions first

For anything substantial, end your brief with: "Before you start, ask me any questions you need answered to do this well." The tool will usually come back with sensible questions about audience, tone or detail that you had not thought to include. Answering them takes a minute and saves several rounds of corrections.

Refine it in conversation

Treat the first answer as a draft, not a verdict. Specific feedback works best: "too formal, use shorter sentences, cut the second paragraph and make the opening about the customer". Vague feedback such as "make it better" produces vague changes.

It is also worth asking the tool to review its own work against your brief: "Check this draft against my instructions and list anything you have missed." It is often surprisingly honest about its own shortcomings.

Break big jobs into steps

Long, complicated requests tend to produce shallow results. Split them up instead. For a long article, ask for an outline first, agree it, then ask for one section at a time. For a marketing plan, start with research, then the plan, then the copy. Each step gets the tool's full attention, and you can correct the direction before it has gone too far the wrong way.

Check the work

AI tools can be wrong while sounding completely certain. Before anything goes public or into a decision, check:

facts, names and dates;
numbers and calculations, ideally in a spreadsheet rather than trusting the answer;
quotes and statistics, which can be invented;
links and sources, which may not exist.

If you ask a tool for sources, treat them as leads to follow up, not proof.

Build a library of briefs that work

When a brief produces something good, save it. A simple shared document of your best briefs, one for product descriptions, one for customer replies, one for ad headlines, means you never start from scratch, and anyone who helps you gets the same quality. Many AI tools also let you save standing instructions or set up custom assistants for a particular job, so your context about the business is included automatically every time.

In short

Better briefs, not better tools, are what separate people who find AI transformative from those who find it disappointing. Give it context, a clear task, constraints and an example, feed it your own material, and treat its first answer as a starting point. To see where AI earns its keep across the rest of a business, read the wider ways AI can help a small business. And if you want to get better at judging the copy it writes, start with the classic books on copywriting and persuasion.