Interactive tool · free
Prompt critic
Paste a prompt you already use and see what is clear, what is missing, and a ready line to fill each gap.
Private by design Your answers are processed in your browser only. They are not sent to an AI model, not stored, and never put into a link.
Paste it as it is, in Arabic, English or a mix. Remove names, phone numbers and client data first.
A structural checklist, not an AI quality score. The tool looks for signs of each element (words, labels, lists), so it can miss unusual phrasing, and a prompt that ticks every box can still ask for the wrong thing.
Analysis
Paste a prompt into the box to see the analysis. The check looks for eight elements:
- Goal Why you want the result and what you will do with it.
- Context Who you are, the situation, what has been tried.
- Input material The text or data the model should rely on.
- Audience Who the result is for.
- Constraints Length, tone, language and what to avoid.
- Output format The shape of the output: table, bullets, email…
- Examples A short sample of what you want.
- Verification and quality How to handle what it doesn't know, and how you will judge the result.
How does the check work?
- Goal: a purpose phrase ("so that", "in order to", "بهدف", "من أجل"…) or a "Goal:" line. A task verb on its own (write, summarise, اكتب…) counts as "could improve".
- Context: a "Context:" line, or at least two background signals (I am…, our company…, we sell…, currently…, you are…). One signal means "could improve".
- Input: text between """ or ``` markers, or under a "Text:" line. Mentioning material without separating it means "could improve". Text between markers is treated as material and not searched for instructions.
- Audience: "audience", "readers", "الجمهور" or an "Audience:" line = strong; "for our clients", "للعملاء" = could improve.
- Constraints: four kinds are counted: length, tone, language or dialect, and rules about what to include or avoid. Two or more = strong, one = could improve.
- Output format: a format word (table, bullets, JSON, جدول…) plus a detail (a number, named columns, a "Format:" line) = strong; the word alone = could improve.
- Examples: an "Example:" line or a quoted sample of at least 15 characters = strong; "for example" or "such as" alone = could improve. Optional.
- Verification: "don't invent", "rely only on", "criteria", "cite", "لا تختلق"… = strong; one weaker signal (sources, check, تحقّق) = could improve.
- Warnings: under 12 words; three or more different task verbs; "help me with", "something", or "ساعدني" with no clear verb; an email, a number of 8–15 digits, or a long card or account number; over 120 words with no line breaks or labels; brackets left unfilled.
What the tool does
The tool checks the structure of a prompt you already have, in Arabic, English or a mix. It looks for signs of eight elements that make a request clear (goal, context, input material, audience, constraints, output format, examples and verification) and sorts them into strong, missing and could improve.
Each gap comes with a labelled line you can insert with one click, then complete the part in brackets. It also flags common problems: a prompt that is too short, several tasks crammed into one request, vague wording like "help me with something", and email addresses or phone numbers that should come out before you paste. It is a structural checklist, not a judgement of answer quality.
Who it's for
- People who use AI tools daily and keep getting generic answers
- Teams building a shared prompt library who want one standard for it
- Trainers and participants in AI workshops reviewing their exercises
- Anyone who would rather improve an old prompt than start from scratch
How to use it
- Paste your prompt as it is, after removing any personal or confidential data.
- Read the analysis: warnings, then what is missing, what could improve and what is strong.
- Click "Insert this line" for each element you need, then fill in the brackets.
- Copy the improved prompt, try it, and compare the result with what you used to get.
Privacy
The tool is structured logic running in your page: rules, questions and decision tables I wrote myself, not an AI model. Nothing you type leaves your device; the only thing recorded is that the tool was used, with no content.
The tool looks for words, labels and structure; it does not understand meaning. It can miss an element phrased in an unusual way, or count a word used in another sense. A prompt that ticks all eight boxes can still ask for the wrong thing: the real test is the result you get.