Skip to content
ceeme

By sector

Who is liable for what AI says about our hospital?

Care Trust Ops is the measurement for hospitals, clinics and care homes: ceeme tracks what AI assistants say about your services, departments and practical information, reports every gap with reality as a finding with its source, and routes every correction through an approval flow with an audit trail.

Measure your own company → Rather talk it through first

Who in our organisation needs to own this?

In practice three people together: communications spots it first, the board decides, and the DPO looks on the moment data is involved. That is not bureaucracy — it is why the approval flow sits in the code rather than in an understanding.

WhoWho decidesWhat keeps them up at night
Hospital, clinic or care groupCommunications and the board together, with the DPO alongsideMedical claims, privacy, and what happens when it goes wrong at the weekend
Residential care homeThe director, together with whoever handles communicationsResidents' privacy, and a family forming its picture elsewhere

What exactly can go wrong in an AI answer about care?

Four risks, and the first is the strangest: an engine can say something about you that exists nowhere. Not quoted, not copied — assembled from two half-facts. That is measurable, because we keep the answer and the sources underneath it, and the gap between those two IS the finding.

DriverWhat ceeme measures underneathWhat stays dark without a connection
Hallucination riskinvented and outdated facts about you, counted by typeWe see what an engine invents about you, and we follow the media coverage around it through the press feed.
Compliance / deontology riskclaims your sector does not allow you to makeYour professional body's rulepack is not fully formalised yet; what is there is strict, not complete.
Governance riskwho may publish what, and whether it can be undoneWe show who may do what. Enforcing it per channel needs an operator flag and an adapter.
Reputation riskmisquotes, sliding reviews and where they come fromReviews and social are read live. Print and paywalled media are not in there.
Local trustGBP, reviews and whether your details match everywhereWe read your Google profile. Changing anything in it needs a GBP connection.

Do you look inside our patient records or our systems?

No, and that is not a matter of settings. What we read is what an AI assistant reads too: your public pages, your Google profile, the press, the guides and the reviews. From your systems there is at most a read connection to the web statistics, and that is optional.

SourceKind
GBPpublic — we read it, you connect nothing
press and newspublic — we read it, you connect nothing
reviews and directoriespublic — we read it, you connect nothing
LinkedInpublic — we read it, you connect nothing
GA4connected — only after your approval, read-only unless agreed otherwise
GSCconnected — only after your approval, read-only unless agreed otherwise
CMS (read→scoped)connected — only after your approval, read-only unless agreed otherwise
GBP (read)connected — only after your approval, read-only unless agreed otherwise
GA4 (optional)connected — only after your approval, read-only unless agreed otherwise

Do you write medical content for us?

No. If an engine says something medically wrong, we trace it to the source and tackle that — a page being misread, a guide with an outdated department name. The medically correct wording comes from you, because that is where it belongs.

RuleKind
Not a single sentence that could read as a medical claim or a promise of treatment, not even in passing.hard rule — the build refuses output that crosses it
The measurement runs on public sources. No path exists that brings personal data from your practice into a measurement.hard rule — the build refuses output that crosses it
Every finding, every approval and every change is logged with time and person, and can be read back afterwards.hard rule — the build refuses output that crosses it
We do not write medical content. If something medically wrong is circulating, we tackle the source and not the text.forbidden — we do not do this, not even on request
Patient data is not read, not stored and not processed, in any route.forbidden — we do not do this, not even on request
Residents' data stays entirely outside the measurement — no connection, no import, no setting that enables it.hard rule — the build refuses output that crosses it
Nothing goes out without the person authorised to sign off at your organisation having seen it — the flow sits in the code, not in an understanding.hard rule — the build refuses output that crosses it
Every claim about the care you provide must trace back to something you confirmed yourselves.hard rule — the build refuses output that crosses it
Residents' names, files and data are not read, not stored and not processed.forbidden — we do not do this, not even on request
Not a single sentence promising an outcome or a quality of care, not even if that sentence already sits on your own site.forbidden — we do not do this, not even on request

How does approval work when something wrong is circulating?

A finding arrives with its source and a proposal. Communications assesses, the board approves, and only then does anything go out. For urgent cases there is a shorter path with the same steps — faster, not fewer. Every step is logged, and there is always a way back.

Frequently asked questions

Can you force ChatGPT to correct our waiting times?

Nobody can, and anyone promising it is selling something they do not own. What does work is changing the source the engine leans on. If your own page shows a waiting time nobody maintains any more, that is the fix — and the answer follows on its own, next time the engine looks.

How quickly do we know when something serious is circulating?

As fast as the measurement runs, and you set that frequency. What you do not get today is full media coverage: reviews and social are read live, print and paywalled media are not. The table above says so, because an alerting system that acts as if it sees everything is more dangerous than no alerting at all.

We already have a spokesperson protocol. Does this clash with it?

It should fit inside it, not sit beside it. The approval flow is configurable to whoever may sign off at your organisation, and nothing goes out without that person seeing it. What we add is not a second protocol but the piece that was missing: seeing what is said about you in a place where no journalist calls.

Read on

Free, no account and no card.