Prompt Quality Checker
Score a ChatGPT, Claude or Gemini prompt from 0 to 100 and get specific fixes for vague words, missing format, conflicts, shouting and unfilled placeholders.
The interactive tool loads here when JavaScript runs.
About this toolWhat is the Prompt Quality Checker?
Paste a prompt and the Prompt Quality Checker scores it from 0 to 100 and lists what is missing or risky, each with a reason and a concrete fix. It catches the problems that make answers inconsistent: no clear task, no output format, vague words such as "some" and "good", missing audience or purpose, instructions that contradict each other, ALL-CAPS warnings, long lists of "don't", unfilled template placeholders like [NAME] or {{topic}}, and requests for facts with no permission to say "I don't know".
The checks are deterministic rules that run in your browser. Nothing is sent to an AI model, so the same prompt always gets the same score and you can check confidential prompts. A rewrite skeleton wraps your text in XML sections and adds a bracketed slot for each missing piece, ready to fill in.
- Score from 0 to 100 with Strong, Good, Needs work and Weak grades
- Seventeen rule-based checks with evidence, reason and fix
- Rewrite skeleton in XML sections with slots for missing parts
- Calms ALL-CAPS emphasis while keeping acronyms
- Weak and strong sample prompts
- No AI call: deterministic and private
Updated
How to use the Prompt Quality Checker
- Paste your prompt, or load the weak or strong sample to see how scoring works.
- Read the score and the findings, which are sorted from high to low severity.
- Each finding shows the words that triggered it, why it matters and how to fix it.
- Use the rewrite skeleton: fill in each bracketed slot, then paste it back to re-check.
- Test the improved prompt on a few real inputs in your model of choice.
Worked examples
The weak sample prompt
Input
"You are an expert. Write me some good content about our product for [COMPANY] etc. Make it short but also very detailed... DO NOT use jargon!!! NEVER mention competitors..."
Output
Score 19 of 100 (Weak): conflicting instructions, unfilled placeholder, no format, shouting, facts without "I don't know", vague words, no audience, negative-only rules
"Short but also very detailed" forces the model to choose; [COMPANY] reaches it as literal text; asking for market statistics with no escape hatch invites invented figures.
The strong sample prompt
Input
Role, <context> with product and page, <instructions> with audience and 120-word limit, <facts>, <output_format>
Output
Score 100 (Strong), no findings
It says what to do, for whom, from which facts, at what length and in what shape, and allows the model to leave out facts it does not have.
Background and insight
The rules are drawn from published prompting guidance rather than invented scoring. Anthropic's guide for Claude advises being specific about format and constraints, explaining the motivation behind an instruction, using examples, telling the model what to do instead of what not to do, and separating parts of a long prompt with XML tags. For current models it also warns that aggressive wording such as "CRITICAL: You MUST" can make the model over-apply a rule, and suggests calmer phrasing. Its hallucination guidance lists giving the model explicit permission to say "I don't know" as a basic technique.
Each finding costs points by severity: 18 for high (the model will probably go wrong), 10 for medium (output will be inconsistent) and 5 for low (an easy improvement). The score is a checklist, not a prediction of answer quality. A one-line question to a chat assistant can be perfectly good and still lose points for missing audience or length; read the reasons and ignore rules that do not apply to your case.
Heuristics have limits. The checker looks for words and patterns, so it cannot tell whether your examples are correct, whether the task is achievable or whether the facts you supplied are true. Testing a prompt on several real inputs remains the only reliable check.
Common use cases
Before shipping a prompt in an app
Catch conflicts, placeholders and missing format rules before a system prompt reaches thousands of users.
Reviewing a team's prompt library
Apply the same checklist to shared templates so they meet a consistent standard.
Debugging inconsistent answers
When output varies from run to run, the cause is often a missing format, a conflict or a vague word.
Learning prompt writing
See which habits matter and why, with a fix for each one.
Tips and limitations
- Treat the score as a checklist. A short, precise question can be excellent without an audience or a length.
- Placeholders are high severity because the model sees "[NAME]" literally and will either invent a name or comment on the bracket.
- Acronyms such as JSON, API and CSV are not counted as shouting; emphasis words such as MUST and NEVER are.
- If a vague word is used precisely ("some" in "some of the rows are blank"), ignore that finding.
- The checker cannot judge facts or examples. Always run the prompt on real inputs.
- Everything runs in your browser; prompts are not stored or sent.
Frequently asked questions
How is the prompt score calculated?
It starts at 100 and subtracts 18 for each high-severity finding, 10 for medium and 5 for low, with a floor of 0. The weights reflect how likely each problem is to change the answer.
Does the prompt checker use AI to grade my prompt?
No. It uses fixed rules that run in your browser, so results are repeatable and nothing is sent to a model or a server.
Why does the checker flag ALL-CAPS words like MUST and NEVER?
Current models follow instructions closely, and shouted rules tend to be over-applied. Anthropic suggests normal wording plus the reason for the rule. Acronyms are not flagged.
Why should a prompt let the model say "I don't know"?
When a prompt asks for figures, dates or sources and gives no way out, a model is more likely to fill gaps with plausible inventions. Explicit permission to admit uncertainty reduces that.
Can a prompt score 100 and still give bad answers?
Yes. The checks cover structure and wording, not whether the task is possible, the examples are right or the supplied facts are true. Test on real inputs.
What does the rewrite skeleton do?
It places your prompt inside <instructions>, lowercases shouted words, and adds sections such as <context>, <constraints> and <output_format> with bracketed hints for whatever the checker found missing.
Is there an in-depth guide to the Prompt Quality Checker?
Yes. “How to Write Better AI Prompts: Structure, Examples and Format” covers the background, how it works, common pitfalls and worked examples. Read the full guide.
Related Tools You May Like
Free, in-browser tools that pair well with this one.