Free tool

Prompt checker: see what your prompt is missing

Paste any ChatGPT, Claude or Gemini prompt below and this free tool scores it on the six things that make AI answers better, a role, context, a clear task, a format, constraints and an example or tone, then shows exactly which ones you are missing and how to fix them.

72 / 4000 characters

46

Needs work

out of 100

  1. Role Missing

    No role assigned, so the model answers as a generic assistant.

    Add one line at the top: "You are a ___ (job title) at a ___ (kind of business or context)."

  2. Context Partly

    Has a little background, but mostly speaks in generalities.

    Add a specific: a name, a number, a date, or what has already happened.

  3. Task Present

    Has a clear action verb, so the model knows what to do.

    Keep it. One sentence, one verb, one ask is exactly right.

  4. Format Partly

    Names a type of output or a vague length ("short"), but no exact structure.

    Turn it into a number or a shape: a word count, a sentence count, bullet points, or a table.

  5. Constraints and check Missing

    No boundaries and no instruction to flag uncertainty.

    Add a limit ("don't invent numbers") and a check ("if unsure, say so instead of guessing").

  6. Examples or tone Partly

    Names a tone, but gives no example of what that sounds like.

    Add a short example or a quote in the voice you want.

Want to see a real AI rewrite a prompt instead of a checklist scoring it? Watch the live prompt makeover demo →

Improved structure

Your prompt, organized under the same six headings, with a fill-in-the-blank line wherever something is missing. This is a scaffold built from simple rules, not an AI rewrite. Copy it, fill in the blanks yourself, and paste the result into ChatGPT or Claude. 

Role:
Add: You are a ___ (job title or expertise) at a ___ (company or context).

Context:
Partly there. Strengthen it: For context: ___ (names, numbers, what already happened, and who this is for).

Task:
Write a marketing email for our spring sale. Keep it short and friendly.

Format:
Partly there. Strengthen it: Format the answer as ___ (bullet points, a table, an email, JSON) in about ___ (a word or sentence count).

Constraints and check:
Add: Do not ___. If you are unsure about something, ___ instead of guessing.

Examples or tone:
Partly there. Strengthen it: Tone: ___ (friendly, formal, direct). For example: "___"

How this works

This tool never calls an AI model. It reads your prompt for the wording patterns below and adds up a score out of 100. It is a checklist based on wording, not a judgment of quality, so treat the score as a prompt to strengthen specific parts, not as a verdict.

  • Role (16 points). Looks for a direct assignment like "you are a ..." or "act as a ...". A hint like "write like a pirate" earns half credit.
  • Context (20 points). Looks for at least two of: a number, a name, background words like "because" or "last month", or simply enough length to carry real detail.
  • Task (24 points). Looks for one clear action verb (write, compare, summarize) or a direct question. Four or more different action verbs, or several separate questions, reads as too many asks bundled together.
  • Format (16 points). Looks for an exact shape: a word or sentence count, bullet points, a table, JSON, or a subject line. Naming only the type of document (an email, a report) or a vague length ("short") earns half credit.
  • Constraints and check (16 points). Looks for a boundary ("don't", "must", "only") and a self-check instruction ("if you are unsure, say so"). Both together earn full credit, either alone earns half.
  • Examples or tone (8 points). Looks for a named tone (friendly, formal, direct) and an example or quote to anchor it. Either alone earns half credit.

Because this is pattern matching on wording and not meaning, it has real limits. A prompt that hits every keyword without making much sense can still score well, and a well-written prompt phrased in an unusual way can score lower than it deserves. Use the result as a checklist to strengthen your own prompt, not as the final word on whether it will work.

Frequently asked questions

What makes a good ChatGPT or Claude prompt?

A good prompt usually covers six things: a role for the model to play, real context about your situation, one clear task, a format for the answer, a constraint or a check for uncertainty, and a tone or example to anchor the style. A quick question does not need all six, but the more that is riding on the answer, the more of them are worth including.

Why does giving the AI a role make a difference?

A role like "you are a tax accountant" or "act as a skeptical editor" tells the model which knowledge, vocabulary and tone to lean on before it writes a single word, instead of defaulting to a generic, average answer. It is one line to add, and it is one of the fastest ways to turn a generic answer into a specific one.

How long should a prompt be?

As long as it takes to give a stranger everything they would need to do the task well, and no longer. A one-line prompt is fine for something trivial. Once real work rides on the answer, most prompts are too short rather than too long: they are missing context, a format or a constraint, not padded with extra words.

Does this work for Claude, Gemini and other AI tools too, not just ChatGPT?

Yes. Role, context, task, format, constraints and tone are habits for writing to any chat AI, not a ChatGPT-specific trick. The same six checks apply whether you paste the result into ChatGPT, Claude, Gemini or anything else.

Related

Want the next one first?

Tools like the prompt checker start in the community, where I post the prompts and workflows I actually use with clients.

Join the community

Want it done for your business?

Twenty free minutes. I will tell you where AI fits in your business, or that it does not yet.

Book a free 20-minute call