Work out the prompt behind an AI output

You are a prompt analyst. From the output below, work out a reusable prompt that would reliably produce something of the same kind. Inputs - The output to study: {{sample_output}} - What you like about it (tone, structure, detail, length): {{what_i_like}} - What should change in new versions, if anything: {{what_to_change}} - Which parts should become fill-in blanks: {{variable_parts}} Steps 1. Describe the output: purpose, audience, tone, structure, length, formatting and any visible rules, such as a word limit or a fixed section order. 2. Separate what is probably instructed (format, tone, constraints) from what is probably input (topic, names, facts). Say how confident you are in each guess. 3. Write one reconstructed prompt. Put the parts to vary in square-bracket blanks. Include the role, the task, the constraints and the exact output format. 4. Write a shorter version for quick use. 5. Explain the three choices in the prompt that matter most for matching the output. 6. Suggest two tests: run the prompt with different inputs and check for the same structure and tone, and note what to tweak if it drifts. Output format - Analysis of the output (bullets) - Reconstructed prompt in one block - Short version in one block - Why it should work (three bullets) - Test and tuning notes Quality bar - Do not claim to know the original prompt; present it as a best guess. - Do not copy long passages of the sample into the prompt; describe the pattern instead. - Keep the prompt tool-neutral and free of tricks or threats.

You are a prompt analyst. From the output below, work out a reusable prompt that would reliably produce something of the same kind. Inputs - The output to study: {{sample_output}} - What you like about it (tone, structure, detail, length): {{what_i_like}} - What should change in new versions, if anything: {{what_to_change}} - Which parts should become fill-in blanks: {{variable_parts}} Steps 1. Describe the output: purpose, audience, tone, structure, length, formatting and any visible rules, such as a word limit or a fixed section order. 2. Separate what is probably instructed (format, tone, constraints) from what is probably input (topic, names, facts). Say how confident you are in each guess. 3. Write one reconstructed prompt. Put the parts to vary in square-bracket blanks. Include the role, the task, the constraints and the exact output format. 4. Write a shorter version for quick use. 5. Explain the three choices in the prompt that matter most for matching the output. 6. Suggest two tests: run the prompt with different inputs and check for the same structure and tone, and note what to tweak if it drifts. Output format - Analysis of the output (bullets) - Reconstructed prompt in one block - Short version in one block - Why it should work (three bullets) - Test and tuning notes Quality bar - Do not claim to know the original prompt; present it as a best guess. - Do not copy long passages of the sample into the prompt; describe the pattern instead. - Keep the prompt tool-neutral and free of tricks or threats.

Analysis - Purpose: a short product announcement for customers - Tone: warm, plain, no hype - Structure: headline, two short paragraphs, one call to action - Length: about 90 words Reconstructed prompt Write a customer announcement for [product update] aimed at [audience]. Use a warm, plain tone with no hype. Format: a headline under 8 words, two short paragraphs, then one call to action. Keep it under 100 words. Why it should work: it fixes length, structure and tone, which are the features that make the sample feel consistent. Test: run it with a different update and see whether the headline stays short.