Skip to content

Make a prompt clearer before you send it.

Paste a prompt you’re about to give ChatGPT, Claude, Copilot or Gemini. It checks it against those four companies’ own guides to writing prompts, and spots personal details before they go anywhere.

Free. It all runs in your browser. This page sends nothing you type to us or anyone else.

The prompt enhancer

The checks need JavaScript switched on. They’re listed below, so you can run them yourself.

Spellcheck is off in this box. Browser add-ons that read what you type, such as grammar checkers, can still see it, as they can on any page.

What it checks, and who says so.

Each check is something at least one of the four companies says in its own guide to prompts, and under each one is who says it. The tool only looks for what’s missing, so whether the prompt is any good is still your call.

  1. Says what you want done

    It says what you want done, such as summarise or compare.

    • Microsoft “Goal: What do you want Copilot to do?”
    • Google “Tell Gemini what you need it to do”
  2. Gives the background

    It says what this is for and why. The AI knows nothing about your work unless you say.

    • Anthropic “a brilliant but new employee who lacks context on your norms and workflows”
    • Microsoft “Context: Why do you need it, and who is involved?”
  3. Says who it’s for

    It says who will read or use the answer.

    • Microsoft, in its example prompt: “Our audience is professionals who work in a hybrid environment”
  4. Points to what to work from

    It points to what to work from, or has it pasted in below the instructions.

    • Microsoft “Source: What information or samples should Copilot use?”
    • Google “Use your documents.”
  5. Says what the answer should look like

    It says how long the answer should be and what shape it should take.

    • OpenAI “Be specific … about the desired context, outcome, length, format, style”
    • Anthropic “Be specific about the output format and constraints you want.”
  6. Keeps the instructions apart from the material

    On a long prompt, the text to work on is marked off from the instructions.

    • OpenAI “use ### or """ to separate the instruction and context”
  7. Avoids vague words

    It says what it means, such as “under 120 words” where it might have said “short”.

    • OpenAI “Reduce ‘fluffy’ and imprecise descriptions”
  8. Says what to do, as well as what not to

    Where it says what to avoid, it also says what it would like instead.

    • OpenAI “Instead of just saying what not to do, say what to do instead”
  9. Leaves out personal details

    Email addresses, phone numbers, card and account numbers, tax file and Medicare numbers, street addresses, dates of birth and passwords. This one’s ours, from our never-paste list.

The four guides

Want your team writing prompts like this?

Prompting that works is part of every session: a weak prompt from the room, fixed together, then everyone writes one for their own work.