Check your prompt before you send it

This prompt checker scores any prompt from 0 to 100 against 11 criteria that separate vague requests from prompts that get useful answers. Paste your text, see which ingredients are present and which are missing, and get a short tip for each gap. The analysis runs in your browser and your prompt is not stored.

🔒 Runs in your browser. Your text is not stored on our servers.

How to use it

  1. Paste your prompt

    Copy the prompt you plan to send to ChatGPT, Claude, Gemini or any other assistant and paste it into the box (up to 8,000 characters).

  2. Click Check my prompt

    The score ring, a grade from Weak to Excellent and the list of criteria appear instantly. Missing criteria are listed first, each with a tip.

  3. Fix the gaps or let the tool rewrite it

    Edit your prompt using the tips and check again, or click Rewrite it for me to get an Expert-level version built around your text.

The 11 criteria and how much each one weighs

Each criterion is either met or not, and the points of the criteria you meet add up to the score. The weights reflect how much each ingredient usually changes the answer:

CriterionPointsWhat the prompt tester looks for
Clear task14An explicit action verb: write, summarize, compare, translate, plan…
Context12Background such as "I am…", "our product…", "currently…"
Role10"You are…", "act as…" or a Role heading
Constraints10Limits: a number of words or items, "only", "avoid", "must"
Output format10Table, list, JSON, headings, paragraphs…
Audience8Who the result is for: beginners, customers, managers…
Structure8Line breaks, headings, numbered or bulleted parts, XML tags
Examples7"For example", "such as", a quoted or fenced sample
Tone & style7A named tone: formal, friendly, persuasive…
Clarity7Few vague words like "something", "stuff", "good"
Detail7Length between 40 and 900 words

A total of 85 or more is graded Excellent, 70–84 Good, 50–69 Needs work and below 50 Weak.

Before and after: a real example

A typical first draft looks like this:

Write something good about our new app for social media.

It scores low: there is a task verb, but no role, no context about the app, no audience, no format, no limits, and it contains two vague words. Here is the same request after following the tips:

You are a social media manager for a budgeting app.

Context: our app "Pocket" launched this week. It sorts expenses automatically and is free for students.

Task: write 3 Instagram captions announcing the launch for university students.

Requirements:
- Max 40 words each, friendly tone
- One clear call to action: download from the link in bio
- Avoid financial jargon

Format: a numbered list.

The rewrite meets almost every criterion and, more importantly, gives the AI enough to write captions you can actually post.

What the score can and cannot tell you

The checker detects whether the building blocks of a good prompt are present, using keyword and pattern matching in English, Spanish, German, French and Portuguese. It does not judge whether your idea is good, whether facts in the prompt are correct or how a specific model will respond. A prompt can score 100 and still ask for the wrong thing, and a short prompt with a low score can be perfectly fine for a quick question such as a definition.

Use the score as a checklist, not a grade on your work. The most valuable part is usually the list of missing items: if Context and Audience are flagged, adding two sentences about them will improve the answer more than polishing wording.

Rewrite it for me

The Rewrite it for me button takes your text and builds an Expert-level prompt around it, with role, context placeholder, requirements, an example slot and a quality checklist, in a neutral format that works with any assistant. When AI mode is available on this site, the local version is then replaced by a rewrite from a language model that keeps your intent and details. Either way, review it: delete what does not apply and fill in the [bracketed] parts.

If you want to target a specific model or choose tone, format and answer language, paste your idea into the AI prompt generator instead. To write instructions for a chatbot that must behave consistently over many conversations, use the system prompt generator.

Pro tips

  • Fix the highest-weight gaps first: Clear task, Context, Role, Constraints and Output format account for 56 of the 100 points.
  • Put each part of the prompt on its own line or under a short label. This alone meets the Structure criterion and makes the prompt easier for you to edit.
  • Replace vague words with numbers and names: "3 captions under 40 words" instead of "some short captions".
  • Paste a short sample of the style you want between triple quotes """…""" to satisfy Examples and steer tone at the same time.
  • Recheck after every edit; the score updates instantly, so you can see which change moved it.

Frequently asked questions

What is a good prompt score?

70 or more (Good) is enough for most tasks. Aim for 85+ (Excellent) when the output matters: client work, long documents, code or data analysis.

Does a higher score guarantee a better AI answer?

No. The score shows that the usual ingredients are present. The quality of the answer still depends on how accurate your context is, the model you use and whether the task is clear to a human reader too.

Why is my short prompt marked as weak?

Prompts under 40 words fail the Detail criterion and usually miss context, audience and format as well. For simple factual questions that is fine; for creative or professional work, add the missing parts.

Can I check prompts written in Spanish or other languages?

Yes. The criteria recognize common phrasing in English, Spanish, German, French and Portuguese. In other languages, structure, length and some format terms still work, but other criteria may be missed.

Is my prompt saved anywhere?

The check itself runs entirely in your browser. Only if AI mode is enabled and you click Rewrite it for me is your text sent to the configured AI provider to produce the improved version.

Related tools

Updated: