# Prompt Ambiguity Auditor

Reviews an existing prompt for ambiguous wording and unstated assumptions that cause inconsistent AI outputs, then rewrites it for precision.

## Prompt

You are a prompt engineering expert who specializes in finding the exact wording that causes AI models to produce inconsistent or unexpected outputs.

Prompt to audit:
{{prompt_to_audit}}
What you want the output to consistently look like: {{desired_output_description}}
What's currently going wrong (if you've tested it): {{observed_problems}}

Do the following:
1. Identify every phrase in the prompt that could reasonably be interpreted more than one way, and explain the specific alternative interpretations possible.
2. Identify any unstated assumptions the prompt relies on (things you know but didn't tell the model, like expected length, format, tone, or audience).
3. Flag any instructions that could conflict with each other if followed literally.
4. Rewrite the full prompt to remove the ambiguity, adding explicit constraints where needed (format, length, tone, structure) without making it unnecessarily long.
5. Explain in one sentence per fix why each specific change should produce more consistent output.

## Best for

Prompt engineers, developers, and power users building reusable prompts or AI workflows who need consistent, predictable output instead of results that vary wildly between runs.

## Compatible tools

- Claude
- ChatGPT
- Gemini

## How to use

- Paste the exact prompt you're currently using, along with what you want the output to consistently look like.
- Share specific examples of what's gone wrong if you've already tested the prompt.
- Review the flagged ambiguities and unstated assumptions to understand your own prompt-writing blind spots.
- Test the rewritten prompt across several runs to confirm consistency has improved.

## Customization tips

- If you're building a prompt for a specific AI model, mention it, since different models interpret ambiguous instructions differently.
- For prompts used in production systems, ask for a version with explicit output format constraints (like JSON) if structured output is required.
- Run this audit again after any significant prompt edit to catch new ambiguity introduced by the changes.

## Example input

Prompt: 'Write a summary of this article.' Desired output: A consistent 3-bullet summary highlighting the main argument, key evidence, and implication. Observed problems: Sometimes get one paragraph, sometimes 5 bullets, sometimes includes opinion.

## Example output

Flagged ambiguity in 'summary' (no length or format specified), an unstated assumption that bullets are wanted, no instruction to stay neutral/factual, a rewritten prompt specifying exactly 3 bullet points covering main argument, key evidence, and implication, with an explicit instruction to remain neutral and avoid inserting opinion, and reasoning for each fix tied to the specific inconsistency observed.