You ask a chatbot for help with a piece of work. The reply lands in seconds and it is… fine. Slightly bland, a little generic, not quite what you had in mind. Most people shrug, blame the tool and go back to doing it themselves. The prompt is usually the problem.

Prompting is not a dark art. It is closer to briefing a capable freelancer you have never met: quick, willing and completely unaware of your circumstances. Most weak answers come from a handful of avoidable habits. Here are seven of them, with fixes you can use today.

1. Writing a question when you needed to write a brief

"Tell me about email marketing" is a question. It invites a summary of everything and nothing. A brief does the opposite: it narrows the job until a decent answer is almost inevitable.

Compare it with: "Write a 200-word introduction for a newsletter aimed at independent café owners. The goal is to persuade them to try a loyalty card scheme. Plain English, no exclamation marks, no jargon." Same tool, very different result.

A workable brief usually answers a few simple questions:

• Who is this for, and what do they already know?
• What should it achieve — inform, persuade, summarise, help me decide?
• How long, and in what shape? Bullet points, a table, three options, one paragraph?
• What tone do you want, and what should it avoid?

You do not need all four every time. But if you cannot answer them yourself, the chatbot cannot either.

2. Assuming it knows your context

Chatbots have no idea about your team, your deadline, your earlier decisions or the email you are replying to. They will fill the gap with something plausible, which is how you end up with advice for a multinational when you run a two-person studio.

The fix is boring and effective: paste the material. Drop in the customer complaint, the meeting notes, the paragraph you are rewriting, the constraints you are working within. If a detail would change the answer, include it. "Rewrite this so it sounds less defensive, but keep the apology and the refund offer" works precisely because the source text is sitting right there.

3. Building up to the request instead of leading with it

Long, gentle openings — "I was wondering whether you might be able to help me with something, it's about a project I'm working on, and I thought maybe…" — bury the task. By the time the actual request arrives, the framing has muddled it.

Lead with the instruction, then supply the context. "Summarise the notes below in five bullet points for a project update. Audience: senior managers who have not seen the project before." One sentence of task, one of context.

Politeness is fine. Hedging is not. "Maybe something about pricing?" gives the tool nothing to aim at.

4. Chasing a weak first draft with a weak follow-up

The first answer is a starting point, not a verdict. Most people reply with "make it better", which is the prompt equivalent of shrugging. Nothing changes, because nothing was specified.

Useful follow-ups carry a specific instruction:

• "Cut it to 120 words without losing the third point."
• "The second paragraph is too formal — rewrite just that one."
• "Give me three alternative openings, each with a different tone."
• "Keep the structure, but replace every example with one from retail."

Two or three of those will improve a draft more than ten hopeful nudges. If a follow-up is not working after a couple of attempts, the original prompt was probably too vague — go back and rebuild it.

5. Never saying what a good answer looks like

If you do not describe the shape of the answer, you get the default: tidy paragraphs, a neutral tone, a short summary at the end. That is rarely what you need.

Say it out loud. "A table with three columns." "A numbered list of risks, each with a one-line mitigation." "No more than 150 words." "Write it as an email I can send without editing." Format instructions are cheap to give and save a lot of reformatting later. If you are unsure, ask for two versions in different formats and pick the better one.

6. Trusting fluent sentences as though they were checked facts

Chatbots write confidently whether or not they are right. Names, dates, figures, quotations and references can be produced smoothly and be completely wrong. Fluent is not the same as accurate.

Treat anything factual as a claim to verify, especially if it is going into something public or being used to make a decision. Ask where a figure came from, then check it yourself. For anything with legal, medical or financial weight, use the chatbot to organise your thinking and then speak to a qualified professional before acting on it.

7. Using one endless conversation for everything

A long thread carries every earlier instruction with it. Halfway through, you are arguing with a summary of a summary, and details from an abandoned approach keep bleeding into the current one.

Start a fresh conversation for each new task and paste in only the context that matters. If you are refining one piece of work, staying in the same thread is fine — just say clearly what has changed: "Ignore the earlier version. We are now writing for a different audience."

A short routine worth keeping

If you take nothing else from this, borrow the following and use it every time:

1. Write the instruction first, in one sentence.
2. Add the context: audience, purpose, constraints, source material.
3. Describe the format and the length you want.
4. Read the answer, then send one specific follow-up rather than a vague one.
5. Check any fact, figure or quotation before you use it.

That is perhaps two extra minutes of typing. In return, you get answers that need editing instead of rewriting — and you stop blaming the tool for a briefing you never quite gave.

Photo: Airam Dato-on / Pexels