AI Writing Review Checklist: Facts, Voice, and Sources
HumanizerKit | Version 2026-10-14
https://humanizerkit.com/guides/ai-writing-review-checklist

Reuse and adaptation permitted with attribution to HumanizerKit.
This workflow aid does not certify accuracy, originality, or authorship.

Draft / URL: ____________________
Reviewer: ____________________
Review date: ____________________
Audience and purpose: ____________________

1. Purpose and audience
Start by recording who the draft is for and what they should understand or do next. For a US audience, use consistent American spelling and an appropriate tone, but preserve source spellings in proper names and quotations.
[ ] The main point appears early.
[ ] The reader has enough context to take the next step.
[ ] Examples fit the intended audience and are labeled if hypothetical.
[ ] The draft avoids unsupported urgency, superlatives, and guarantees.

2. Facts and meaning
Compare the edited draft with the original and then compare factual claims with their sources. Those are two different checks: a rewrite can preserve the original meaning and still repeat a false claim.
[ ] Names, dates, quantities, units, and product limits match verified sources.
[ ] Qualifications and uncertainty remain intact.
[ ] Quotations match the source exactly and have an attribution.
[ ] No statistic, result, testimonial, or personal experience has been invented.
[ ] A second reviewer checks consequential claims when the context calls for it.

3. Sources and attribution
Open each reference rather than relying on a plausible-looking title. Confirm that it supports the specific sentence where it is used. A paper about an earlier detector or a different writing sample cannot establish the current performance of your tool.
[ ] Each citation resolves to a real source.
[ ] The source supports the claim, not just the general topic.
[ ] Publication date, study conditions, and relevant limitations are clear.
[ ] Borrowed language is quoted or paraphrased accurately with attribution.
[ ] Any source-overlap report has been reviewed in context.

4. Voice and readability
Read the draft aloud or have someone unfamiliar with it summarize the main point. A sentence that sounds polished can still conceal a missing explanation. Keep necessary technical terms, but explain them for readers who may not know them.
[ ] Each paragraph has a clear purpose.
[ ] Vague phrases are replaced with concrete actions or supported detail.
[ ] Sentence length varies naturally with the ideas.
[ ] The tone fits the author and publication.
[ ] The final revision preserves the intended meaning.

5. Detection and disclosure
If a detector was used, record the tool, date, input length, and whether the input was truncated. Keep the result separate from your judgment about the draft. A high or low score does not determine who wrote it.
[ ] Any AI detector result is treated as uncertain review information.
[ ] No accusation or authorship claim rests only on a detector score.
[ ] AI assistance is disclosed where the relevant policy requires it.
[ ] Private drafts and personal information are handled with permission.
[ ] Source notes and relevant revision history are retained.

6. Leave a short review record
Record the reviewer, review date, changes made, and unresolved questions. A useful note is specific: “Verified the three pricing figures against the plan page; removed an unsupported adoption statistic; client must confirm the launch date.” Avoid a blanket “AI checked” label that says nothing about the review.
Revisit the record when underlying information changes. If your article includes prices, availability, or product capabilities, link to the current source and update the claim when you learn it is no longer accurate.

Changes made: ____________________
Unresolved questions: ____________________