Home › Topics › Prompting

Prompting that works

How to brief an AI tool the way you would brief a capable new colleague, so you spend less time rewriting what it gives you.

Guides in Prompting

About this topic

Prompting has picked up a reputation as a secret art, full of special phrases and elaborate templates. In practice it is closer to a skill most of us already have: explaining a job clearly to someone who has never seen our work before.

The difference with AI is that it will never ask a clarifying question unless you invite it to. It fills every gap with the most typical guess. Leave out the audience and it writes for everyone. Leave out the length and it picks one. Leave out what you already tried and it suggests the obvious. Much of the frustration people feel with AI output comes from those silent guesses.

The guides in this topic focus on closing those gaps. You will not find lists of magic words here. You will find ways to structure a request, examples of weak and strong briefs side by side, and techniques for fixing output that is close but not quite right.

If you only read one piece from this section, make it Writing Prompts That Actually Work. It covers the core habits in one sitting.

Key ideas

A prompt is a brief

Write it for a smart stranger who knows nothing about your situation. If they could not do the task well with what you wrote, the model probably cannot either.

Paste the source, do not describe it

Models are far more reliable when they transform material you provide than when they produce facts from memory.

Examples beat adjectives

Two sample paragraphs in the style you want will get closer than a list of words describing that style.

Constraints are a gift

Word limits, a fixed format and things to avoid all narrow the guessing and tend to improve the result.

The first answer is a draft

Treat the reply as a starting point and give specific feedback, the way you would with a colleague's first attempt.

Start fresh when a chat goes wrong

A long conversation full of corrections can carry confusion forward. A new chat with a better brief is often quicker.

A five-minute habit for better prompts

You do not need a template library to improve. Before you send any request that matters, run through these questions. After a few weeks they become automatic and take seconds.

What exactly do I want back? Name the output: an email, a table, five headline options, a one-paragraph summary. Vague requests get vague shapes.

Who is it for? A summary for your manager and a summary for a new customer should look very different. Say who will read it and what they already know.

What does the model need to know that it cannot guess? Background, deadlines, previous decisions, the tone your organisation uses. Paste the relevant material rather than paraphrasing it.

What should it avoid? Jargon, invented figures, a sales tone, recommendations you have already rejected. One line of limits saves a round of edits.

How will I check it? Decide in advance which parts you will verify. If accuracy matters, ask the model to quote the source text behind each point so checking is quick.

When the answer comes back, resist the urge to say "make it better". Say what is wrong: too long, too formal, missing the pricing point, assumes the reader is technical. Precise feedback gets precise fixes. The productivity topic shows how to turn prompts that work into reusable routines.

Common myths

MythThere is a perfect prompt for every task.

In practiceGood prompts depend on your situation, audience and material. A template is a starting point, never the finished brief.

MythTyping in all capitals makes the model pay attention.

In practiceClear placement and plain wording work better than shouting. Put the most important instruction on its own line, near the end.

MythPrompting will soon be unnecessary.

In practiceModels keep getting better at guessing, but they still cannot know your context unless you share it. Explaining a task clearly will stay useful.

MythLong prompts always beat short ones.

In practiceRelevant detail helps and padding hurts. A short brief that captures what matters beats a long one full of boilerplate.

Prompting glossary

Plain-English definitions for terms you will meet in this topic.

Zero-shot prompt
A request with no examples. Fine for simple tasks, weaker when you need a particular format or style.
Few-shot prompt
A request that includes a handful of examples of the output you want. One of the most reliable ways to control style and structure.
System prompt
Standing instructions set before a conversation starts, often by the app maker. Some tools let you write your own as custom instructions.
Role prompt
Asking the model to take a perspective, such as a sceptical editor. Useful for tone and focus, but it adds no real expertise.
Prompt chaining
Breaking a large job into steps and feeding each result into the next request. Easier to check than one huge prompt.
Grounding
Tying the answer to material you supply, such as a report or policy, rather than the model's general memory.
Output format
The shape you ask for: bullet points, a table, a fixed set of headings. Stating it up front reduces reformatting later.

Questions about prompting

Should I save my prompts?

Yes, especially for tasks you repeat. A plain document of briefs that worked, with a line on when to use each, becomes very valuable within a few months.

Do prompts work the same in every AI tool?

Mostly. Clear briefs travel well between tools. You may need small adjustments for length limits or formatting quirks.

What if the model ignores part of my instructions?

Move the ignored instruction to its own line near the end, or point out the miss in a follow-up message and ask it to fix only that part.

Is it worth paying for prompt packs?

Rarely. Most are generic templates you could write yourself in a few minutes, and copying them teaches you little about why they work.

Other topics

Learning AI from scratch

For people who want to understand AI well enough to use it with confidence, without a maths degree or a coding bootcamp.

Tools for real work

How to judge AI tools by the jobs they do for you, not by their launch videos, and how to keep your setup small and your data safe.

Careers and skills

A calm, practical look at how AI is changing work, written for people who would rather make a plan than read another prediction.

Productivity and automation

How to find the repetitive parts of your week, hand them to AI safely and keep the time you win back.