The medical fact-check workflow — 5 gates inside Content HQ

Why “the doctor will read it” isn’t a workflow

The default healthcare content review looks like this: writer drafts → editor polishes → doctor reads final → publish. It feels reasonable. It fails in three ways. The doctor is the bottleneck. The doctor reviews only the final, which means errors are already in derivative assets. And there’s no audit trail, so when a regulator or an aggrieved patient asks “who approved this and when,” the answer is “I think Dr. ___ saw it.”

A workflow distributes the load across roles, runs gates in parallel where possible, and produces a signed log. The output is the same content, with one-fifth the senior-clinician burden and a full audit trail.

The 5-gate workflow

Content HQ splits clinical review into five sequential gates. Each has a named role, a fixed question, and a time budget. The piece cannot advance until the prior gate logs an explicit yes/no with a comment.

Total active clinician time per 1,500-word piece: under 45 minutes split across two specialists. Total elapsed SLA: 72 hours.

The source-of-truth registry

Gate 1 is the discipline most clinics skip. Before a draft can be submitted, every clinical claim must be tagged with a registry ID. The registry is a per-client database of vetted clinical facts, each with: source title, source type (NMC guideline, ICMR statement, peer-reviewed paper, internal SOP), publication date, expiry date, and the lead clinician who endorsed it.

The registry replaces the writer’s ad-hoc Googling. If the registry has no entry for “recommended fasting window before LDL test in Indian adults,” the writer cannot insert a number — they raise a registry request, the clinical lead resolves it, and from then on every writer pulls from the same vetted entry.

Acceptable source types

• National Medical Commission clinical guidelines (current)
• ICMR statements and Indian disease-burden papers
• Peer-reviewed papers indexed on PubMed
• WHO documents
• Named-clinician original commentary (logged as “internal expert”)
• Government registries — CGHS, ABDM, CDSCO public lists

Not acceptable

• AI-generated summaries without traced primary sources
• Competitor clinic websites
• Lay press articles (use only as topic-trigger, never as fact source)
• Anonymous expert quotes from non-medical news outlets

Gate 3 — specialty review and the rotation roster

Specialty review must be done by a clinician with current practice in that subspecialty. A general physician cannot sign off on an IVF embryo grading article. Content HQ maintains a rotation roster: 2-4 named specialists per subspecialty per client, with capacity limits (e.g., 6 pieces/month each). The system auto-assigns the next available reviewer and sends a calendar block.

The roster prevents the “one heroic doctor reviews everything” failure mode. It also produces a real co-author byline — the reviewing clinician’s name appears on the published piece with their NMC registration number, satisfying E-E-A-T signal requirements.

Gate 4 — the compliance sweep

The compliance lead runs a fixed 12-item checklist:

1. No success-rate numbers without trailing source citation
2. No before/after imagery for surgical procedures (DCA red flag)
3. No “guaranteed,” “cure,” or absolutist language
4. No prescription-drug brand names in patient-facing content
5. NMC registration numbers present for every named clinician
6. Disclaimers present on disease-management content
7. No testimonials of named patients without DPDP-grade consent record
8. No competitor mentions
9. No price claims without effective-date stamp
10. No promotional terms inside doctor bios (DCA)
11. Image alt text includes no clinical claim that wasn’t fact-checked
12. Schema markup matches body claims

The audit trail

Every gate writes a row in the review log: timestamp, reviewer, decision (approve/return/escalate), comment, and any registry IDs added. The log is exported as a PDF attached to each piece. When a query lands — patient, regulator, internal QA — the answer is on a one-page audit sheet, not in someone’s memory.

The cardiology case — what informal review missed

An anonymised cardiology hospital ran 41 educational articles in 2025 with informal review. ICG’s 2026 onboarding audit re-ran the 5-gate workflow retrospectively against all 41 pieces. Findings:

• 14 pieces (34%) had at least one clinical claim that couldn’t be traced to an acceptable source
• 9 pieces had an outdated guideline reference (older than 24 months)
• 6 pieces had DCA-borderline absolutist language
• 3 pieces had an NMC registration number missing from a quoted clinician
• 1 piece had a dosing schedule that conflicted with the 2024 ICMR update

None of these were caught by “the doctor reads it.” All were caught by gate 1 + gate 4 on the rerun.

What to set up this week

• Stand up a source-of-truth registry — even a simple shared spreadsheet works for v1.
• Name 2-3 reviewing clinicians per specialty and book recurring calendar slots.
• Run the 12-item compliance checklist on your last 10 published pieces.
• Lock the publish button behind a signed-log artefact.

Related reading

• Content HQ product page
• E-E-A-T amplification per content piece
• AI-assist vs AI-generated healthcare content
• Content HQ performance scoring