Verified against ChatGPT · 2026-08-14
Run a health content draft through a claim-by-claim QA pass before it goes anywhere near publish
Audits a draft piece of health or wellness content claim by claim, separating what's supported by cited sources from what's asserted without one, and flagging anything that reads as diagnostic or prescriptive for a clinician to review before it ships.
The prompt
Ready to copy — highlighted parts are example details you can swap.
You are running a quality-assurance pass on a draft piece of health or wellness content before it goes to a clinician for final review. Your job is to find every problem, not to make the draft sound more finished than it is.
DRAFT CONTENT
A 600-word blog post about magnesium and sleep quality, including the line 'studies show magnesium significantly improves sleep for most adults' with no citation attached.
SOURCES OR CITATIONS PROVIDED
One small 2012 study on magnesium supplementation and self-reported sleep quality in older adults with insomnia, provided as a link.
INTENDED PUBLICATION CONTEXT
Public-facing wellness blog on a supplement retailer's website.
AUDIENCE
General adult consumers researching sleep aids, no assumed medical background.
PROCESS
Go through the draft claim by claim. For each factual or clinical claim, mark it as: supported by a provided source (cite which one), asserted without a provided source, or written in a way that's vague enough to be unverifiable as stated — vague claims are still a problem, since "studies show" without a specific source is functionally an unsupported claim wearing a disguise. Separately flag any sentence that reads as personalized medical advice, a specific treatment recommendation, or a diagnostic statement ("if you have X symptom, you likely have Y") rather than general information, since that category needs to be rewritten as general education framed for a reader to discuss with their own clinician, not asserted as applicable to any individual reader. Check whether the content anywhere implies certainty inconsistent with the strength of the underlying evidence (a single small study written up as settled fact, a common but unproven folk claim stated as established) and flag the mismatch specifically. Do not fix these problems yourself in this pass — this is a QA report, not a rewrite, so the person who wrote the draft or the clinician reviewing it makes the actual edit decisions.
WHAT NOT TO DO
Do not add a citation of your own to fill a gap you found, and do not soften your assessment of a flagged claim because the surrounding writing is otherwise good — a QA pass that pulls punches to be polite defeats its own purpose.
MANDATORY DISCLAIMER
Open the QA report with a note stating that this pass checks internal consistency, sourcing, and tone against the material provided, that it is not itself a clinical fact-check, and that final medical accuracy sign-off must come from a qualified clinician before publication regardless of how this report reads.
OUTPUT FORMAT
1. The disclaimer.
2. A claim-by-claim table: claim (quoted), status (supported / unsourced / vague-unverifiable), note.
3. A separate list of any sentences flagged as personalized-advice or diagnostic-sounding language, with the exact quote.
4. A short overall risk summary: how many high-priority issues before this should go to clinical review.Customize
Optional — swap in your own details for the highlighted parts above.
Why this works
The specific trap in health content QA is a vague-but-confident claim like "studies show" that sounds sourced without actually citing anything checkable — treating that phrasing as its own flagged category, distinct from a plainly unsourced claim, matters because a lazy content-QA pass would only catch claims with zero supporting language and miss the more common pattern of hedge-language doing the work of a citation without any of its accountability. Separating "supported," "unsourced," and "vague-unverifiable" into three distinct statuses rather than a binary sourced/unsourced check gives whoever reviews the report a much more actionable picture, since a vague claim and a flatly unsupported one require slightly different fixes (add a real citation versus rewrite the claim entirely). Flagging sentences that read as personalized advice or diagnostic framing as a wholly separate category from sourcing issues addresses a distinct risk: a claim can be perfectly well-sourced in general and still be dangerous if it's phrased in a way that lets an individual reader apply it to themselves as a diagnosis ("if you have X, you likely have Y"), which is a framing problem, not a citation problem, and needs to be caught even when the underlying fact is accurate. The instruction against the model fixing the draft itself, and against softening flagged issues to be polite, keeps this pass functioning as a QA gate rather than a co-authoring pass — a QA report that quietly resolves issues on the writer's behalf, or downgrades a real problem because the prose reads well, stops being a reliable gate before it reaches clinical review. The disclaimer is explicit that this pass is not itself a clinical fact-check, because internal consistency and sourcing checks are necessary but not sufficient — only a qualified clinician can confirm the underlying medical accuracy the sourcing check assumes.
What you get back
Note: this checks sourcing, consistency, and framing only — it is not a clinical fact-check; final sign-off must come from a qualified clinician. Claim: "studies show magnesium significantly improves sleep for most adults" — status: vague-unverifiable (no specific source named in the sentence, and the one provided source is a small study on older adults with insomnia, not "most adults" generally). Flagged as overreaching relative to the underlying evidence.
Verified against
ChatGPT GPT-5.1 · 2026-08-14
Changelog
- 2026-08-14 — Initial publish, verified against ChatGPT GPT-5.1.
Need this built into your business?
If a prompt isn't enough — what Scult builds, built and maintained for you — that's Scult's day job.
EXPLORE WHAT SCULT BUILDS
