do AI assistants actually recommend clinics?

Yes, with a disclaimer in front. In our test on 5 September 2026 both ChatGPT and Perplexity opened with “this is not medical advice” and then named five to six specific providers for dental implants in Austin, telehealth for type 2 diabetes and a knee replacement in London. The two engines rarely agreed on the names. What they agreed on was the method: registries, reviews and studies first, the clinic’s site second.

what sources do the engines check before naming a clinic?

In our test: the National Joint Registry and CQC data for surgery in London, Peterson Health Technology Institute evaluations and a DOI-linked study for telehealth, Reddit, RealSelf and review feeds for dental implants, plus the clinics’ own pages once a clinic was already on the list. That is the layer a healthcare GEO program builds: the records the engines verify against, resolving to your clinic, in a form a crawler can read.

we have never worked with an agency that has no healthcare clients. why would we start with you?

You should ask that, and here is the honest answer: because the audit is free, measured and yours to keep whoever you hire. We publish our method and a 100-audit dataset; healthcare is where we are extending it, and the first clinics get the full audit at no cost while we do. If the audit shows something a specialist agency should fix, the list of specialist agencies is on this site, scored on the same criteria we score ourselves on.

is this compliant with HIPAA, GDPR and advertising rules for medical services?

The technical and entity work touches no patient data: it is schema, crawler access, credentials and public pages. Content is drafted for a clinician’s review and your compliance process before it goes live, and we do not publish outcome claims the practice cannot document. If your jurisdiction restricts how medical services can be advertised, that constraint goes into the brief before anything is written.

what does the healthcare bench actually consist of?

Right now: the live three-prompt test on this page, the product and credential checks that run in every one of our 160 audits, and the pattern from three law-firm and nine financial audits, where the engines treated credentials the same way. No completed healthcare audit, no healthcare client. The page will be updated the day that changes, with the date.

Healthcare SEO agency: GEO for clinics

009 Agency is a healthcare SEO agency built around Generative Engine Optimization: we make clinics, provider groups, telehealth companies and medtech brands visible, cited and chosen inside AI answers across ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek — and we re-measure it every two weeks against a logged baseline. Healthcare is the largest search demand of any vertical we track: more people look for a healthcare SEO agency each month than for one in legal, ecommerce and fintech combined.

It is also the one vertical on this site where we have no completed audit yet. We would rather say that than borrow a number. So instead of a bench, this page shows a live test: on 5 September 2026 we ran three patient prompts through ChatGPT and Perplexity, logged the answers verbatim, and read which sources each engine used to decide who to name. The pattern is the same one we see in every regulated category, with one healthcare-specific twist.

why clinics lose in AI answers specifically

The general failure pattern holds everywhere: known but not recommended. In our published dataset 93% of brands were described accurately by AI and only 4.5% were named when a buyer asked who to pick. Healthcare adds a twist: the engines do not just prefer verifiable sources, they say so to the patient, and they check against registers a clinic rarely thinks of as marketing.

the engine names the clinic it can verify, and tells the patient how it checked

Asked for a knee replacement in London, ChatGPT built its shortlist around the National Joint Registry: it cited named consultants’ NJR profiles, ODEP implant ratings and the CQC’s open data, then named King Edward VII’s, London Bridge Hospital, Cleveland Clinic London and Spire. Perplexity named Cleveland Clinic London, HCA, Cromwell, Fortius and the Harley Street knee clinics, and told the patient to prefer “named specialist teams rather than just marketing claims”. Neither engine cited a hospital’s marketing page as the reason to trust it. A clinic whose surgeons, outcomes and inspection record are not readable where the engine looks is competing on adjectives.

Reddit and peer-reviewed evaluations outrank your website

Asked for dental implants in Austin, ChatGPT’s first source was Reddit, four times, then RealSelf, a Birdeye review feed and the prosthodontists’ college. Asked for a telehealth provider for type 2 diabetes, it cited the Peterson Health Technology Institute’s evaluations five times, a DOI-linked study and Wikipedia, and named Teladoc, Virta, Omada and Dario on that basis. Perplexity used the clinics’ own sites more, but still leaned on Reddit, US News and the federal telehealth portal. In both prompts the engines opened with a disclaimer that this is not medical advice — and then recommended providers anyway. The recommendation is assembled from third-party evidence; the clinic’s site supplies the details once the clinic is already on the list.

the credential layer is machine-unreadable in the same way it is in law and finance

We have not audited a clinic yet, but we have audited three law firms and nine financial companies, and the engines treated credentials the same way in all of them: a licence as an image, a bar number missing, a founder whose record resolved to a dead page produced a warning in the answer, not a footnote. Healthcare has more registers than either — medical boards, NJR, CQC, NPI, HIPAA notices, accreditation bodies — and the engines already read them. The audit’s first job is to check which of yours resolve.

what the shelf looked like on 5 september 2026

three patient prompts · ChatGPT gpt-5.5 and Perplexity sonar-pro, web search on, logged verbatim
the patient asked	who was named	what the answer was built from
dental implants, Austin	ChatGPT: Council Oak Perio, Austin Oral Surgery, Central Texas Oral Surgery, Shoal Creek Prosthodontics, AIDM. Perplexity: ClearChoice, Union Dental Implant Center, Austin Implants & Periodontics, Austin Implant Clinic, Bedrock Dentistry. No overlap.	ChatGPT: Reddit ×4, RealSelf, Birdeye reviews, the prosthodontists’ college. Perplexity: the clinics’ own implant pages, Reddit ×2, DexKnows.
telehealth for type 2 diabetes, US	ChatGPT: Teladoc/Livongo, Virta, Omada, DarioHealth. Perplexity: Virta, Teladoc, Vida, Amwell, Steady Health, LifeMD.	ChatGPT: PHTI evaluations ×5, a DOI-linked study, Wikipedia. Perplexity: telehealth.hhs.gov, US News, GoodRx, CB Insights, Teladoc’s own site.
private knee replacement, London	ChatGPT: King Edward VII’s, London Bridge / HCA, Cleveland Clinic London, Spire, with named consultants. Perplexity: Cleveland Clinic London, HCA hospitals, Cromwell, Fortius, Harley Street Specialist Hospital.	ChatGPT: National Joint Registry surgeon profiles, CQC open data, ODEP, the hospitals’ own pages. Perplexity: the hospitals’ pages ×7, Circle Health, two knee-specialist sites.

Three things repeat. The two engines rarely name the same providers, so a clinic can be on one shelf and off the other. ChatGPT decides from registries, reviews and studies before it reads a clinic’s site; Perplexity reads the site but ranks by what the site can prove. And every answer carried a “verify credentials” instruction to the patient, which is the engine telling you what it will check next.

what the first 90 days look like

The order is the same as everywhere, because it has to be: readable, then verifiable, then quotable.

Audit. The prompts your patients actually ask — by procedure, by location, by condition and by insurer — run live through the engines and logged verbatim, plus crawler checks with real AI user-agents and a pass over every register the engines used above. You get a health score, what the engines currently say about your clinic, and the three most expensive gaps. Complimentary, and for the first healthcare clients it is also how we build the bench, which we will say to your face. what is inside →
Make the credentials machine-readable. Physician and MedicalOrganization records with licences, board certifications, NPI or GMC numbers and accreditations; one canonical domain per clinic; the review platforms and registers the engines read resolving to you. Shipped as files your web vendor applies. technical GEO →
Publish what gets quoted. One page per procedure and situation, written to be lifted in a paragraph and signed by the clinician who performs it, reviewed for compliance before it goes live. Outcomes and volumes where you can state them, because that is what the registries the engines cite contain.
Get onto the lists. The review platforms, specialist directories, condition communities and evaluations the engines actually cited in your specialty and city — identified from your own panel, not from a generic outreach list.
Re-measure. Same panel, every review, reported as a delta. Series, not screenshots.

proof, and its limits

There is no healthcare result on this page, because there is no healthcare client yet. The measured result we can point to is from another regulated vertical: a legal client went from 0/5 to 4/5 engines naming it on its category prompt by day 17 (July 2026), with the log published. The mechanics that moved it — readable credentials, one entity, a page per specific question, placement in the sources the engines cite — are the ones the engines used in the table above. The first healthcare audit we complete will replace this paragraph.

one thing we will not do

We will not write medical claims. Every page that describes a procedure, an outcome or a condition is drafted for a named clinician to review and sign, and nothing goes live under an anonymous byline or with an outcome figure the practice cannot back. The engines above cited registries and peer-reviewed evaluations over marketing pages for a reason; a claim that fails against those sources produces a warning to the patient, not a recommendation.

healthcare SEO agency: GEO for clinics and medtech