Trust Health Score and coverage
What the number means, why coverage is separate, and when the score is suppressed.
What the number is
A deterministic, versioned summary of what the public record currently says about your clinic. The same inputs and the same formula version always produce the same number.
Next to the number you will always see the formula version that produced it. A change of version is a documented event, not a silent recalculation.
What the number is not
It is not a clinical quality measure and must never be presented as one. It measures public reputation, which is a different thing from how well a clinic treats patients. Presenting it as a quality rating breaches the Acceptable Use Policy.
What goes into the number
Five parts, each measured from the public record:
- Ratings — the star ratings your sources report, weighted by how many items each rating is drawn from, with every source's own scale divided out so a 4-out-of-5 and an 8-out-of-10 count the same.
- Sentiment — how positive or negative the writing itself is.
- Negative themes — the share of discussion that is complaint.
- Replies to reviews — how often the clinic responds, where the source reports it.
- Spread across sources — whether your standing rests on one listing or several.
How much people have written is deliberately not one of them. Volume tells us how much of the picture we have, not how good the picture is, so it belongs to coverage instead. Mixing the two would make both numbers move together and neither explainable.
Each part carries a weight. The exact weight applied to your score is shown next to it in the product, on every reading, alongside the formula version — so what you see is what was actually computed rather than a document that may have drifted from it.
A part we could not measure carries no weight. Its share is spread across the parts we could. If a source stops reporting reply rates, that must not look like a clinic that stopped replying.
Coverage Confidence
Reported separately, never folded into the score. It is built from how much has been written, how many of the sources we asked actually answered, how many distinct sources contributed, how recent the reading is, how certain we are the profiles are yours, whether the languages could be identified, whether the reading could be pinned to your location, and whether any source was listed twice.
It is graded High, Medium, Low, or Insufficient, and the grade is computed without ever looking at the score. Two clinics with identical evidence and opposite ratings grade identically.
When no score is shown
When coverage is Insufficient, the numeric score is suppressed entirely.
It is not estimated and it is not shown with a caveat. A clinic with two
reviews must never look mathematically comparable to one with two thousand.
Three conditions withhold the number outright, whatever else is true:
- Fewer than 15 public items exist. Below that, one unhappy patient moves the number more than the clinic's actual standing does.
- No source we asked returned anything.
- Fewer than a third of the sources we asked returned anything — scoring what came back would describe a fraction of the picture as though it were all of it.
A reading also loses freshness over 30 days, after which it contributes nothing to coverage until a new one is taken.
Where the AI is, and is not
A model classifies public items into themes and writes summaries. It does not produce, adjust, weight, or override the number. That separation is what makes the score reproducible — see the published Methodology and AI Disclosure.
Reading a movement
Before acting on a change, check three things: how many items the period contains, whether the formula version is the same as the period you are comparing against, and which themes moved. The evidence behind each theme is one click away.
Was this page helpful?