A hospital with six years of blogging usually has somewhere between a few hundred and a few thousand health pages. Nobody has read them all since they were published. Some were written by an agency, some by a consultant's assistant, some are leftovers from a migration. A few make claims that the hospital's own compliance lead would not approve today.
This article is for marketing leads, SEO managers and content heads who have been asked to "clean up the health content" and need a defensible way to do it. It gives you a scorecard for six risk dimensions, a method for scoring a sample of 200 pages, and a triage matrix that turns scores plus Search Console data into one of four actions: keep, rewrite, merge or remove. It is a working template. The clinical judgements stay with the clinicians who review the flagged pages.
The reason to score at all, rather than just delete thin pages, is that the pages that pose the most risk are often not the thin ones. A long, well-ranked article with an unreferenced dosage statement or an implied cure is more dangerous than a short outdated one. Google's guidance on helpful, reliable content and its Search Quality Rater Guidelines both put health among topics where trust matters most, and an audit should look for trust problems, not only for length.
Scope: what counts as a YMYL health page
"Your Money or Your Life" content is content that could affect a person's health, safety, finances or wellbeing. For a hospital site that means condition pages, symptom pages, treatment and procedure pages, drug and test pages, cost pages, and any blog post that gives guidance on health decisions. It does not usually include a careers page or a parking map.
Start from a crawl of the whole site, filter to the clinical sections, and then draw your 200. If the site has fewer than 200 such pages, score all of them. If it has thousands, take a stratified sample, with proportionate numbers from each specialty and each content type, plus every page in the top tier by clicks. The aim is a picture of the whole site you can act on, not a perfect census.
Our overview of E-E-A-T for healthcare websites explains the framework; this article is the operational template that follows from it.
The scorecard: six dimensions
Score each page from 0 to 2 on each dimension, where 0 is a clear problem, 1 is partial and 2 is sound. Define each level in writing before anyone scores a page, so two auditors would reach the same number.
| Dimension | 0 | 1 | 2 |
|---|---|---|---|
| 1. Claims | Specific clinical claims with no support, or claims beyond the evidence | Mostly supported, with some vague or generalised statements | Every clinical claim is qualified, supported and within scope |
| 2. Sources | No sources, or only weak ones (news, brand pages) | Some primary sources, not mapped to claims | Primary sources mapped to claims, current and checkable |
| 3. Reviewer | No named author or reviewer | Author named, no clinical review shown | Named clinical reviewer with credentials and a review date |
| 4. Dates | No date, or a date that conflicts with the content | Published date only, or old with no review | Published and reviewed dates visible and plausible |
| 5. Outcome promises | Promises, guarantees or comparative superiority | Ambiguous language such as "best results" | None; limits and uncertainty acknowledged |
| 6. Intent fit | Page does not answer the query it ranks for | Partly answers, with padding or tangents | Answers the intent and routes the reader to a sensible next step |
A seventh column of facts, not scores, goes beside them: URL, specialty, page type, word count, clicks and impressions over a fixed period, linked conversions if tracked, the number of other pages targeting the same query, and the date of the last substantive edit.
Critical flags that override the total
Some findings should trigger action no matter how well the page scores elsewhere. Give them their own column:
- A guarantee, cure claim or implied outcome.
- A dosage, self-medication or self-treatment instruction without a cited, approved source.
- Content that contradicts the hospital's current clinical practice.
- A patient identifiable from text, image or testimonial without clear consent.
- A statement that appears to breach advertising rules for doctors or hospitals. In India the NMC advertising guidance is the reference we use with clients; confirm against the regulator's own text.
- A named doctor who has left the institution, still shown as the reviewer or treating physician.
A critical flag sends the page to the clinical lead and compliance lead for a decision within an agreed period, and the page is edited or taken down while that happens if the flag is serious.
Running the audit on 200 pages
Set up. Build a spreadsheet with one row per page and the columns above. Pull Search Console data for a fixed period (the longest you have, up to 16 months), and join it to the crawl by canonical URL. The Performance report can export queries and pages, and the Search Console API can export larger sets.
Calibrate. Have two auditors score the same ten pages independently. Compare the results, discuss every disagreement and tighten the definitions. Repeat until scores agree on nearly all pages. Skipping calibration is the most common reason audits cannot be defended later.
Score. At roughly ten minutes a page, 200 pages is around 33 hours of work, so divide it between two auditors and give a clinician a defined slice of flagged pages, not the whole set. Score the first read quickly, and record a note for every 0.
Verify. Pick twenty scored pages at random and have a second person re-score them. If the totals differ by more than a point on many, recalibrate.
Review with clinicians. Send the critical-flag pages and the lowest-scoring pages to the relevant specialty reviewer. Bundle them by specialty so a consultant reviews related pages together.




