Reconstructed artifact
This is not a verbatim copy of the company prompt. It documents the editorial checks and workflow I remember building from recurring feedback.
1. Search-intent intro
Does the introduction answer the reader’s core question quickly enough? Flag broad openings, long scene-setting and generic statements that delay the useful answer.
Typical instruction
Refine the intro so it directly answers the reader’s search intent before expanding the context.
2. Structure and flow
Check whether sections follow a logical progression and whether transitions make the relationship between ideas clear.
3. Specificity and clarity
Flag vague or fluffy phrasing. Ask what the sentence actually means, what detail is missing and whether the point can be made more concrete.
Typical instruction
Tighten this phrase and make the explanation more specific.
4. Examples, facts and support
Identify claims that need evidence and sections that would be stronger with an example.
Typical instructions
- Add an example to strengthen this point.
- Add a fact, claim or source here.
5. Writer judgment
The AI output was a review layer, not a rewrite command. Writers were expected to assess the flagged issues, preserve the brief, verify factual information and only accept changes that improved the article.
6. Self-edit before submission
After the refinement pass, the writer completed a final self-editing checklist before the draft came to me for editorial review.