How to write a good AI prompt: a practical guide

A practical, step-by-step guide to writing AI prompts that get useful answers from ChatGPT, Claude and Gemini: role, context, task, format, and how to check the result.

By the ThinkWithPrompt editorial team at Synottic · Updated 11 October 2026

A good prompt is not clever wording. It is a short brief that gives the model what a capable colleague would need to do the job well: who they are acting as, the situation, the exact task and what the finished result should look like. Most disappointing answers come from leaving one of those out, not from the model being unable to help.

1. Start with the outcome, not the topic

“Marketing email” is a topic. “Write a 120-word email inviting existing customers to a product webinar on 14 November, with one clear call to action” is an outcome. Before you write anything else, finish the sentence “When this works, I will have…”. That sentence usually becomes your task line.

2. Give the model a role

A role sets the vocabulary, priorities and level of expertise the model writes from. “Act as an experienced B2B copywriter” produces different choices from “Act as a customer support lead”, even for the same task. Keep it to one plain line; inflated personas (“world-class expert with 30 years of experience”) add nothing.

3. Add the context only you know

The model does not know your audience, your constraints or what has already been tried. Context is where you supply them:

Audience: who will read or use the output, and what they already know.
Situation: why this is needed now and any relevant background.
Constraints: length, tone, things to avoid, facts that must be included.
Source material: notes, data or a draft to work from, pasted in full.

This is why every prompt in our library uses highlighted [BLANKS]: they mark exactly the context you need to supply for the prompt to fit your situation.

4. State the task precisely

Use a direct verb and one job per prompt: write, summarise, compare, critique, plan. If the work has stages, list them as steps. If there are criteria a good answer must meet, say so, for example “rank the options by cost, then by time to implement”.

5. Specify the format

Say what the result should look like: a table with named columns, five bullet points, a 200-word summary followed by risks, an email with a subject line. A clear format makes answers easier to scan, easier to compare and much easier to reuse.

6. Ask the model to check its work

For anything that matters, add a short verification step: “Before you answer, check the draft against my requirements and list any assumptions I should confirm.” This catches missing requirements and makes guesses visible instead of hiding them inside confident prose. It is the idea behind the RCTFVV framework.

7. Iterate in the same conversation

Treat the first answer as a draft. Point to what is wrong (“the tone is too formal for this audience; shorten the second section by half”) rather than starting again. Once a prompt reliably gives you what you need, save it so you never have to rewrite it.

A before-and-after example

Before: “Write a project update.”

After:

Act as a project manager writing for busy senior stakeholders. Context: Project: [PROJECT NAME]. Audience: [WHO WILL READ IT]. This week: [WHAT HAPPENED]. Risks or blockers: [RISKS]. Task: Write a status update that leads with overall status (on track, at risk or off track), then progress, risks with owners, and decisions needed. Format: Under 200 words. A one-line status at the top, then three short sections with bullet points.

The second version is longer, but every line removes a guess the model would otherwise have had to make.

Common mistakes

Several jobs in one prompt. Split “research, outline and write” into steps, or into separate prompts.
No audience. The same content written for a CFO and for a new hire should look very different.
Missing source material. If you want a summary of your notes, paste your notes; do not describe them.
Trusting facts and figures. Models can state wrong numbers, quotes and citations fluently. Check anything you will rely on.
Pasting what you shouldn’t. Leave out confidential, personal or regulated data unless your organisation has approved the tool for it.

A quick checklist

Does the prompt say who the model should act as?
Does it include the context only you know?
Is the task one clear job with a direct verb?
Is the output format specified?
For important work, does it ask the model to check itself and flag assumptions?

Frequently asked questions

How do I write a prompt for AI?

Write a prompt for AI in four parts. Give the AI a role (who it should act as), the context only you know (audience, situation, constraints, source material), one clear task with a direct verb, and the format you want back (a table, five bullet points, a 150-word email). For anything important, add a final line asking it to check the draft against your requirements and list its assumptions.

How long should an AI prompt be?

As long as it needs to be to remove guesswork, and no longer. A quick rewrite can be one or two lines; a report, plan or analysis usually needs a short paragraph of context plus the source material pasted in full. Length that adds real detail improves answers; filler such as inflated personas does not.

What is the most common prompt-writing mistake?

Leaving out context. The model doesn't know your audience, constraints or what has already been tried, so it fills the gaps with generic assumptions. Adding two or three lines about who the output is for and what it must include usually improves the answer more than any clever wording.

Prompts that follow this structure

Compare options with a decision matrixScores options against weighted criteria so the trade-offs and the best choice are explicit.
Turn rough notes into a polished emailConverts fragmented notes into a clear, natural email ready to send.
Build a presentation storyline, slide by slideDesigns the narrative first — central message, evidence, and one message per slide.
Summarize meeting notes or a transcriptTurns raw notes or a transcript into a clean summary with decisions, actions, and open questions.
Create a new-hire onboarding planA 30-60-90 onboarding plan and welcome pack for a new employee.
Write an employee announcement or internal messageClear, human internal communications — changes, news, policies, and benefits.