How it works

Paste the prompt you use now, exactly as you use it, and say what's wrong with the results if you know.
Pick the AI you use (ChatGPT, Claude, Gemini or any), then click Improve my prompt.
Fill in the [PLACEHOLDERS], paste the improved prompt into your AI, and compare the result with what you got before.

Examples

Sample results for the inputs shown, so you can see the format and quality before you try it.

Example 1: input

Your current prompt: write a blog post about remote work

Which AI will you use? Any AI (ChatGPT, Claude, Gemini)

What's wrong with the results now? Too generic

Example 1: output

IMPROVED PROMPT Write a blog post about [SPECIFIC ANGLE ON REMOTE WORK] for [AUDIENCE]. The goal is for readers to [WHAT READERS SHOULD DO OR THINK AFTER READING]. Use these points and examples as the backbone of the post, and treat them as the source of truth: [YOUR POINTS, EXAMPLES, DATA] Method: 1. Open with a specific situation or problem the audience recognizes, not a general statement about remote work. 2. Make 3 to 4 points, each backed by one of my examples or data points. 3. For each point, give one thing the reader can do this week. 4. End with a short, practical conclusion, not a summary of every point. Format: 800 to 1,000 words, a headline, short subheadings, paragraphs of 2 to 4 sentences, plain language. Don't invent statistics, studies or quotes. If a point needs evidence I didn't give, write [NEEDS SOURCE] instead. Before answering, check that every section contains something specific to [AUDIENCE] rather than advice that applies to anyone. WHAT CHANGED - Added an angle, audience and goal; "remote work" alone is why the results were generic. - Added a slot for your own points and examples, so the post has material the AI can't invent. - Added a four-step structure that requires a concrete action per point. - Banned invented statistics, which generic posts tend to fill gaps with. FILL IN [SPECIFIC ANGLE ON REMOTE WORK] — for example, running meetings across time zones [AUDIENCE] — who reads your blog, such as first-time managers [WHAT READERS SHOULD DO OR THINK AFTER READING] — the one takeaway [YOUR POINTS, EXAMPLES, DATA] — notes, stories or numbers from your own experience

Example 2: input

Your current prompt: Summarize these meeting notes. Be detailed but keep it short. Include action items. Also make it sound professional and friendly and put it in a table.

Which AI will you use? Claude

What's wrong with the results now? It ignores the action items half the time

Example 2: output

IMPROVED PROMPT Summarize the meeting notes in <notes></notes> for people who missed the meeting and need to know what was decided and who is doing what. <notes> [PASTE MEETING NOTES] </notes> Write the summary in this order: 1. Summary: 3 to 5 bullets covering the decisions made and why, in plain, friendly, professional language. 2. Action items: a table with columns Action | Owner | Due date. Include every action item in the notes. If an owner or date isn't in the notes, write "not stated" rather than guessing. 3. Open questions: anything raised but not resolved, or "None". Use only what is in <notes>. Don't add decisions or owners that aren't there. Before answering, check that every action item mentioned anywhere in <notes> appears in the table. WHAT CHANGED - Put the action items in their own table with a final check, which fixes them being skipped. - Resolved "detailed but short": the summary is short bullets, and detail lives in the action table. - Wrapped the notes in <notes> tags, which Claude handles well for pasted material. - Added "not stated" for missing owners and dates, so gaps don't get filled with guesses. FILL IN [PASTE MEETING NOTES] — the raw notes or transcript

Example 3: input

Your current prompt: You are a data analyst. Analyze this sales data and tell me insights.

Which AI will you use? ChatGPT

Example 3: output

IMPROVED PROMPT Act as a sales analyst preparing a one-page readout for [AUDIENCE, E.G. THE SALES DIRECTOR]. I've uploaded [FILE NAME], which contains [WHAT EACH ROW IS AND THE KEY COLUMNS]. Use the uploaded file as the only source; calculate with code rather than estimating. 1. Check the data first: date range, row count, missing or odd values. List any problems before the analysis. 2. Compare [PERIOD] with [COMPARISON PERIOD] for total revenue, deals won and average deal size. 3. Break revenue down by [REGION, REP OR PRODUCT] and name the three biggest changes, with the numbers. 4. For each change, give the most likely explanation the data supports, and say what the data can't tell us. Format: a short summary (3 bullets), then a table of the key numbers, then the three changes with one recommended next step each. Before answering, check that every number you quote was calculated from the file. WHAT CHANGED - Named the audience and the decision, since "insights" without a reader produces a list of trivia. - Asked for a data check first, which catches gaps before they turn into wrong conclusions. - Told ChatGPT to calculate with code on the uploaded file instead of estimating numbers. - Required each insight to separate what the data shows from what it can't show. FILL IN [AUDIENCE, E.G. THE SALES DIRECTOR] — who will read the analysis [FILE NAME] — the file you upload [WHAT EACH ROW IS AND THE KEY COLUMNS] — for example, one row per deal with date, rep, region, amount, stage [PERIOD] and [COMPARISON PERIOD] — for example, Q3 2026 and Q2 2026 [REGION, REP OR PRODUCT] — how to break the numbers down

Tips for better results

Say what's going wrong. "It ignores the action items" gets a targeted fix; without it, the improver can only guess at the problem.
Paste your real prompt, typos and all. The improver keeps every requirement you wrote and resolves contradictions instead of dropping them.
Save the improved version as a template. Most of the gain comes from reusing a prompt that already has the structure, not from rewriting it each time.
Test old and new on the same input. If the new one isn't clearly better, add the specific failure to the problem box and run it again.

FAQ

Is this prompt improver free?

Yes. Enter your email once to use it (you'll also get Something Big, our free weekly AI newsletter, and you can unsubscribe anytime). There's no account and no credit card.

How is this different from the AI prompt generator?

The prompt generator writes a new prompt from a description of your task. The prompt improver starts from a prompt you already have, keeps what you wrote, and fixes what's making the results weak.

What does it usually change?

Most weak prompts are missing the audience, the source material, the output format, or rules against making things up. The improver adds those, resolves contradictions, and explains each change.

Does it work for ChatGPT, Claude and Gemini?

Yes. Pick the AI you use and it follows that model's conventions, such as XML tags around pasted material for Claude. Or pick Any AI for wording that works everywhere.

More free AI tools

Related prompts

Prefer to use ChatGPT or Claude directly? These free prompts do similar jobs.