How AI talks about finding and registering with a GP in England

Published: 15 September 2026. Last updated: 15 September 2026.

A person trying to join a GP surgery needs a different answer from someone booking at a surgery they already belong to. This study makes that distinction visible in AI-search measurement. We track 75 patient questions focused on England, including London-specific searches. A snapshot captured on 15 September 2026 contains 572 answers across ChatGPT, Gemini and Google AI Overview. Every service in the selected 14-service panel is scored against that same answer denominator. The NHS App is the most frequently named service, appearing in 258 answers, or 45.1%. Even this leading service appears in fewer than half of the measured answers.

In short

• The NHS App appears in 258 of 572 answers, or 45.1%.
• NHS 111 follows at 23.3%, then Patient Access at 7.7%.
• 8 of 14 tracked services are named. The full-panel median naming rate is 1.6%.
• NHS App naming varies by task: 90.5% in the booking-route group and 36.2% in the registration group.
• The separate source snapshot records 730 retrievals for nhs.uk. Retrieval events aren't a count of unique answers.

What we measured

This is a selected panel of services discussed around GP access, from national NHS services to booking and consultation software. It isn't a census of providers or a ranking of clinical quality. The measurement records what the engines return. A name appearing in a question doesn't earn a service credit in the result.

The public measurement receipt contains the service counts, definitions, anonymous full-panel distribution, question-group totals and source retrieval records. The capture date is 15 September 2026. We don't claim a trend across several days or a change since an earlier measurement. The figures describe the saved snapshot.

Naming rate means the number of answers that name a service at least once, divided by 572. Naming the same service repeatedly within an answer still counts as one named answer. Percentages are rounded to one decimal place. Answers can name several services, so the percentages aren't portions of a pie and shouldn't be added into a market-share total.

A separate source report records retrieval events. A source can be retrieved repeatedly, including within the same answer. Its answer denominator wasn't preserved in a form we can reconcile, so this report publishes event counts only. We don't divide those events by 572 or describe them as the percentage of people who saw a source.

The service names are measurement labels. They identify the services observed in the snapshot, rather than certifying current product availability, ownership or suitability for any individual patient. Official links below explain what each service actually offers.

Which services appear in the answers?

The table includes every service with a positive naming count in the measured panel, ordered by named answers. It deliberately compares presence in the discussion. NHS 111, an NHS app and a practice's consultation system serve different needs; they aren't interchangeable products a patient can freely select at every surgery.

NHS App

NHS 111

Patient Access

eConsult

myGP

AccuRx

PATCHS

Livi (now Kry)

The full distribution is steep. Across all 14 services, the median naming rate is 1.6%. Six services have no recorded naming in this snapshot; their identities aren't published. The eight positive rows above are the complete positive distribution, so there isn't an unseen group just below the table. Livi's single named answer sits at its lower edge. Its current official UK website redirects to Kry, while the measurement label remains Livi to preserve what was counted. The measurement receipt gives the complete anonymous count distribution for independent calculation.

The stronger conclusion is that the leading name still leaves 54.9% of answers without a mention. That doesn't represent a gap the NHS App should fill everywhere. It shows why a broad GP-access score needs to be read alongside the patient's task.

Registration and appointment booking are different jobs

The NHS App's official guidance says users must already be registered with a GP surgery in England or the Isle of Man. Its help pages also explain how an existing user can change GP surgery in the app. Both facts matter. An absolute claim that the app cannot support registration would be wrong. NHS App eligibility and changing GP surgery.

For someone seeking their first registration, the NHS website provides a route to find a surgery and submit an online application where offered. Someone already registered and moving surgery can follow the app's changing-surgery route. Someone seeking an appointment needs the booking or request route their surgery provides. These are different starting conditions, even when people use similar words to describe them.

The distinction is practical. A page headed “register online” should explain whether it means creating an app account, joining a GP surgery or connecting an existing patient record. A page headed “book a GP” should explain whether the next step is selecting an appointment or submitting a request for the practice to assess. Those explanations give people a usable route and give search systems precise material to retrieve.

We haven't scored the correctness of every answer. A mention of the NHS App in a registration answer can be appropriate context, a comparison, a warning about eligibility or advice to an already-registered person. The aggregate count can't distinguish those uses. Establishing an error rate would require an answer-by-answer assessment against a defined rubric and the patient's stated circumstances.

That's why the measured finding is about naming and task differences. It doesn't establish that AI routinely sends unregistered patients into the wrong flow, and it doesn't measure how many people abandoned a registration after following an answer.

The same service appears differently across patient tasks

The questions cover finding a surgery, registering, access barriers, getting an appointment and choosing a booking route. These group totals describe the collected answers, not the volume of real-world patient searches. The number of answers in a group depends on the measurement design and engine returns.

For booking-route questions, the NHS App appears in 105 of 116 answers, or 90.5%. For registration questions, it appears in 42 of 116, or 36.2%. The equal group denominators make this particular comparison straightforward. They don't show that either group has more patient demand or that one engine gives better advice.

The access-barrier group provides a useful check on broad claims. It contains 117 answers, with the NHS App named in two. A report that applied the overall 45.1% rate to every access problem would hide that difference. Operators should inspect the group that matches the work their service can actually perform before choosing pages to improve.

An existing-patient service should make its appointment and practice-linking instructions easy to find. A registration service should explain joining, eligibility and the next step after submitting details. An information service should help readers distinguish the options and point to an appropriate official route. The measurement doesn't give every service the same job.

Which sources were retrieved?

The following table shows recorded retrieval events from the separate source snapshot. It ranks the ten highest-count domains in that saved source report. The type column is an editorial description of the source, not an assessment of its medical authority. A large count shows repeated retrieval in this record; it doesn't prove influence on a particular sentence.

First, official guidance deserves the first content check. The three leading domains are NHS sources. Where a service explains registration eligibility or an appointment pathway, link the relevant rule beside the claim and check that the operational instructions agree with it. A logo or a broad “NHS approved” phrase isn't an adequate substitute for the specific source.

Second, practice websites belong in the research queue. nhsgp.net and gpathand.nhs.uk are practice sites, so calling every NHS-related domain an institution would conceal useful differences. Read the retrieved page's task, opening answer and next step. A clear practice page offers a more useful comparison for local content than a national policy document covering a different purpose.

Third, community material warrants careful reading. Reddit records 34 retrievals, while YouTube records 32 elsewhere in the source export. Neither count establishes that posting there changes an AI answer. Use these sources to identify recurring confusion, then check the exact question against official guidance and repair the corresponding explanation on the appropriate owned page.

What this means for GP service operators in England

Start with the moment at which someone can use the service. A person arriving from an AI answer shouldn't need to infer whether they must already be registered, live inside a catchment area or use a particular practice. Put the applicable conditions beside the action. Then test the route with the same starting information the question gives.

A useful registration page answers the practical question early and links to the live application route. A useful appointment page distinguishes a request from a confirmed slot. A useful practice finder explains what its listings can establish and what the patient still needs to confirm with the surgery. Each page has a job that can be tested directly.

For the measurement itself, keep three records separate: a service named in an answer, a website retrieved as a source and a completed action on the service's own site. Our AI-search measurement methodology explains how to keep those events distinct. A rise in naming doesn't by itself demonstrate more registrations, and a retrieved source doesn't prove that its operator was recommended.

Next, use the source table to choose what to read. If an NHS page is repeatedly retrieved, check the rule it supplies. If a practice page is retrieved, examine the location, task and instruction it covers. If a discussion is retrieved, identify the unresolved question. This is a research queue with a specific reason to inspect each page.

A useful contribution to a local information page might be an accurate explanation of which practices accept online registration, checked against their current routes. A generic request to add a company name offers little value. The relevant test is whether the proposed information helps someone complete the task the page exists to support.

Finally, assess content at the answer level before commissioning more of it. Our guide to how ChatGPT chooses sources provides the wider retrieval context. Compare the exact question, the answer and its links, then record the missing or unclear fact. A precise correction to an existing page can be more useful than another broad article about digital healthcare.

The free AI visibility diagnostic offers a separate first check of your website.

What this report can and cannot establish

This is directional evidence about a defined panel and snapshot. Its bounded questions are:

• Which of the tracked services were named in the saved answers?
• How often did each name appear against the shared answer denominator?
• How did NHS App naming differ across the measured patient tasks?
• Which domains accumulated recorded retrieval events in the separate source report?
• Which official service rules help interpret those observations?

It cannot establish clinical quality, incremental revenue, completed registrations, appointment availability or a causal link between publishing a page and receiving a mention. The selected services don't represent every possible route to care. The pooled results also don't provide a platform-by-platform accuracy comparison. We don't infer individual patient behaviour from generated answers.

The source report has an additional boundary: its event counts can be reproduced, but its percentage denominator cannot be established from the saved export. Publishing only the counts preserves the useful evidence without attaching a false precision to it. Future refreshes should preserve distinct-answer source counts alongside retrieval events.

For an operator, the next step is to split the findings by service fit, inspect the exact answer and its sources, and correct the weakest factual explanation. Chasing one blended ranking encourages a service to answer questions it was never designed to solve.

Sources and measurement record

• Schmitdy: public measurement receipt. Counts, definitions, anonymous distribution, question groups and source records for the 15 September 2026 snapshot.
• NHS: NHS App. Official service identity and scope.
• NHS: who can use the NHS App. Existing GP registration requirement.
• NHS: viewing or changing your GP surgery. The app's changing-surgery route.
• NHS: register with a GP surgery. Registration routes and requirements in England.
• NHS 111 online. Official service identity and purpose.
• Patient Access. Official GP appointment and prescription service description.
• eConsult. Official digital consultation and triage service description.
• myGP. Official app features and practice access information.
• Accurx. Official patient communication service description; measured under the label AccuRx.
• Patchs. Official patient service entry point.
• Kry UK. Current destination of the Livi UK website; the study preserves its recorded Livi label.