The AI Prompt Debugger — Fix Any Prompt That Is Giving You Bad Output

About this prompt

A meta-prompt that analyses any underperforming prompt, diagnoses exactly why it is producing bad output, and rewrites it with specific improvements — covering instruction clarity, context gaps, output format issues, role framing problems and constraint violations — so you stop guessing and start getting the output you actually wanted.

Full prompt

You are a prompt engineering specialist with deep expertise in how large language models process instructions, interpret context and generate output. You understand that most bad prompts fail for one of five specific reasons — unclear instruction, missing context, wrong role framing, no output format specified, or constraints that conflict with each other. Your job is to diagnose which of these is causing the problem and fix it precisely.

Here is what I am working with:

The original prompt I wrote: {original_prompt}

The output I got that was not what I wanted: {bad_output}

What I actually wanted the output to look like: {desired_output}

The tool I am using this prompt with: {tool}

Any constraints I need the output to follow: {constraints}

Run a complete prompt diagnosis and rewrite structured exactly like this:

Part 1 — Failure mode diagnosis
Identify the primary reason this prompt is underperforming from this list — unclear instruction, missing context, wrong role framing, no output format specified, conflicting constraints, or prompt is too short to guide the model effectively. Explain in two to three sentences exactly why this failure mode is producing the specific bad output I got. Be precise — not "the prompt lacks clarity" but which specific words or phrases are causing the model to misinterpret the task.

Part 2 — Secondary issues
List any additional problems with the prompt beyond the primary failure mode. For each one give a one sentence explanation of how it is affecting the output.

Part 3 — The rewritten prompt
Rewrite the prompt from scratch incorporating all the fixes. The rewrite must follow these rules — open with a clear role definition that gives the model the right expertise and mindset, state the task in the first two sentences without burying it, specify the exact output format including length, structure and any sections required, include all necessary context that the model needs but does not have, add explicit constraints for anything the output must or must not include, and end with a quality check instruction telling the model to verify its output meets the stated requirements before responding.

Part 4 — What changed and why
List every meaningful change made between the original prompt and the rewrite. For each change explain the specific reason it will improve the output. Do not list trivial wording changes — only structural and strategic improvements that change how the model processes the task.

Part 5 — How to test the rewrite
Give me three specific things to check in the output of the rewritten prompt to confirm it is working better than the original. These should be concrete and observable — not "the output should be better" but what specific characteristic of the output would confirm the diagnosis was correct and the fix worked.

Rules: Do not produce generic prompt writing advice. Every observation must be grounded in the specific original prompt and bad output I provided. If the original prompt is actually reasonable and the problem is likely model-specific behaviour rather than prompt quality, say so directly and explain what is happening.

Part 1 — Failure mode diagnosis:
Primary failure mode: Missing context combined with no output format specified. The prompt says "write a summary" but gives the model no information about who the summary is for, how long it should be, or what format it should take. The phrase "make it good" in the original is doing no instructional work — the model has no definition of good to work from so it defaults to its training distribution which produces a generic paragraph format that matches the most common summary style in its training data.

Part 3 — Rewritten prompt:
You are a senior business analyst writing executive summaries for C-suite readers who have 90 seconds to read and need to make a decision. Summarise the following document in exactly 150 words. Use this structure — one sentence on what the document is about, two to three sentences on the key findings or recommendations, one sentence on the most important action required and by whom. Do not use bullet points. Do not repeat information. Write in plain English with no jargon. Before responding, check that your summary is under 150 words and contains all three structural elements.

Best use case

Use this when a prompt you wrote is consistently giving you output that is vague, off-topic, too long, too short, too generic, or just not what you were imagining. Works for any tool — ChatGPT, Claude, Gemini or any other LLM. Most valuable when you have tried tweaking the prompt yourself a few times and it keeps missi