Prompt Checker

Paste a prompt and see how ChatGPT, Claude or Gemini will read it: a score out of 100, what each part is missing, and one button to fix it. Free, no sign-up.

0 / 6,000

Want a better version straight away? Open the Prompt Optimizer

Related prompt tools

A weak prompt and its scorecard

An authored example: a one-line cold email request as the checker scores it, then the same request after the fixes, scored again.

A cold email: before

ChatGPT34/100
Before
write a cold email to a gym owner about our booking software, make it convincing

Example score

Overall 34/100 · Needs work

  • Clarity6/10

    The task is clear, but 'convincing' does not say what the email should get the reader to do.

  • Context2/10

    Nothing about the software, the gym or why this owner would care.

  • Constraints2/10

    No length, tone or limits, so the model will write a long generic pitch.

  • Output format3/10

    It is clear an email is wanted, but not whether to include a subject line or sign-off.

  • Examples4/10

    A short email rarely needs a sample, but a line in your own voice would help it sound like you.

  • Context: Add what the software does, the problem it solves for gyms and one proof point.
  • Constraints: Ask for under 120 words, a plain friendly tone and no hype words.
  • Output format: Ask for a subject line, the body and a one-line sign-off, nothing else.
  • Examples: Paste one sentence you would write yourself so the tone matches.
  • Clarity: Name the one action you want: reply, book a call or start a trial.

A cold email: after

ChatGPT84/100
After
Write a cold email to the owner of an independent gym, introducing [SOFTWARE NAME], our class booking software. Context: - What it does: members book and cancel classes from their phone; the gym sees bookings and no-shows in one place. - Why it matters to them: fewer empty spots and less time answering booking messages. - Proof: [ONE RESULT, SUCH AS "a 3-studio gym cut no-shows by a third"]. Requirements: - Under 120 words, plain and friendly, no hype words. - One clear ask: a 15-minute call this week. - Mention the gym by name: [GYM NAME]. Output: a subject line under 50 characters, the email body, then a one-line sign-off from [YOUR NAME].

Example score

Overall 84/100 · Good, a few gaps

Examples is the lowest at 7: a short email rarely needs a sample, but one line in your own voice would still lift it.

  • Clarity9/10
  • Context8/10
  • Constraints9/10
  • Output format9/10
  • Examples7/10

What the checker scores

Five parts decide whether a model answers well on the first try. Each gets a score from 0 to 10, a reason and one edit.

01
Clarity
Is the task stated once, early and in specific words, with nothing contradictory?
02
Context
Does the model get the audience, the purpose and the material it needs to tailor the answer?
03
Constraints
Are length, tone, scope and what to avoid spelled out instead of left to chance?
04
Output format
Does the prompt say exactly what the answer should look like: sections, a list, a table or JSON?
05
Examples
Is there a sample of the wanted result or style where one would help, or is the task simple enough without?

How to use the prompt checker

Three steps from a prompt you are unsure about to one you know is ready.

  1. 11. Paste your promptThe message or system prompt you are about to send, up to 6,000 characters.
  2. 22. Pick the assistantChoose ChatGPT, Claude, Gemini, DeepSeek, Grok and Any assistant, so the score reflects how that assistant reads prompts.
  3. 33. Read the score, then fix itEach dimension gets a score, a reason and one edit. Make the edits yourself, or press Fix it to have the optimizer rewrite the prompt with those notes.

Frequently asked questions

What people ask before they paste a prompt.

1What does a prompt checker do?
It reads a prompt you wrote and tells you where the model will have to guess. You get a score for each part that decides the quality of the answer, a one-line reason for each score and a concrete edit to make. It does not run the prompt or write a new one unless you ask it to with Fix it.
2How is the score worked out?
The checker rates clarity, context, constraints, output format and examples from 0 to 10 each, then gives an overall score out of 100 for how likely the prompt is to get a good answer on the first try. A part the task does not need, such as examples for a simple factual question, is not held against you.
3Is it free?
Yes. You get 10 free runs a day on each device without an account, and 20 a day with a free account. The checker shares that allowance with the other prompt tools, and paid plans remove the daily limit.
4What does Fix it do?
It opens the Prompt Optimizer with your prompt and assistant already filled in, the goal set to match your weakest dimension, and the checker's notes attached. Press Optimize there to get a rewritten prompt with a note on each change. That run counts as one more free run.
5Why does the assistant I pick change the score?
Assistants read prompts differently. Claude follows XML-tagged sections well, Gemini does best with the task in the first sentence, and DeepSeek works better without role play. The checker judges the prompt against the habits of the assistant you pick, and Any assistant uses a plain structure that suits all of them.
6Does a high score mean the answer will be good?
It means the model has what it needs to do the job: a clear task, the context, the limits and the shape of the answer. The facts you put in still matter, so a prompt with the wrong details can score well and still get a wrong answer. Treat the score as a checklist, not a promise.
7Do you keep the prompts I paste?
The prompt is sent to our AI model provider to be scored and is not shown to anyone else. Nothing is saved on your account.

Check a prompt before you send it

Free, 10 a day, no sign-up. Paste it, pick the assistant, read the score.

Paste a prompt