Patients ask AI before they book. What should a US clinic check in AI answers?

Patients are asking AI first

A growing share of US patients ask an AI assistant about their health before they call a clinic. In KFF's tracking poll of 1,343 US adults, fielded 24 February to 2 March 2026, 32% said they had used AI for health information or advice in the past year.¹ KFF is a health policy research organisation, and its question covers health information in general, not choosing a provider.

Vendor research points the same way for provider choice. rater8 sells reputation management to medical practices. Its patient choice survey, published in June 2026, reports that AI tools have become a larger influence on how patients pick a doctor than in the previous edition.² It is a vendor survey with a stake in the answer, so we describe its direction rather than its figures.

For a clinic group, the practical question is narrow: when a patient asks an AI engine about your clinic, or about clinics like yours, is the answer right? This post is a checklist for answering it.

What do patients ask AI about a clinic?

They ask the questions that decide whether to book. Clinical questions matter too, but for a clinic's visibility the high-stakes facts are operational, and they change more often than most websites are updated.

• Insurance: which plans are accepted, at which location, and whether that changed this year.
• New patients: whether each clinician or location is accepting them, and how long the wait is.
• Hours and locations: weekend and evening hours, addresses, parking, telehealth options.
• Conditions and services: what each location treats, and what it refers elsewhere.
• Clinicians: names, specialties, credentials and languages spoken.
• Comparison: how you compare with nearby practices, often drawn from reviews.

Example prompts are plain. "Which dermatology practices near me take Medicare and see new patients?" "Is there an urgent care open on Sunday evening that treats children?" "Does a given orthopedic group accept my plan?" Each one depends on facts that a clinic controls but rarely publishes in one consistent place.

A quarterly AI answer check for a clinic group

1. List the facts

   Insurance, new-patient status, hours, services and clinicians, per location.

2. Write the prompts

   Ten to twenty per location, in the words patients use.

3. Run them on six engines

   On more than one day, because answers vary between runs.

4. Log every error

   Wrong, missing or outdated, with the date and the engine.

5. Trace and fix the source

   Directory, insurer listing, review site or your own page.

6. Recheck next quarter

   And after any change in plans or locations.

Why accuracy is the first risk

Health answers can be wrong in ways that matter. In January 2026, Google removed AI Overviews for some liver-test questions after a Guardian investigation found misleading summaries, including test ranges given without context.³ Google said its own clinicians found much of the information was not inaccurate, but the summaries came down. That case concerned clinical information, not clinic details. It still shows how a confident summary can get a health fact wrong, and how slowly a fix can follow.

For a clinic, the more common errors are mundane and costly. An answer lists a plan you stopped accepting, sends patients to a closed location, or says a clinician is taking new patients when the list closed months ago. Each one creates a wasted call, a frustrated patient or a complaint.

Our view: treat an AI answer about your clinic like a directory listing you do not control. You cannot edit it directly, but you can find and fix the sources it was built from.

How often do health searches show AI Overviews?

Often. In an Ahrefs study of 146 million desktop search results from September 2025, data now ten months old, AI Overviews appeared on 44.1% of medical queries, against 20.5% of results overall.⁴ Ahrefs sells SEO tools, and the study does not state a geography.

That means a patient searching Google for a condition or a treatment option is likely to see an AI summary before any clinic's website. Whether your practice is named or linked in that summary depends on the same pages and listings that feed the chatbots.

Google has also shown it will pull AI Overviews from some health queries when accuracy is in doubt. Our view: do not plan around any one surface. Make the facts correct and consistent everywhere an engine might read them.

Where do AI answers about clinics come from?

They come from the pages an engine can find and trust, which for a clinic usually means five kinds of source. When an answer is wrong, one of these is usually out of date.

Our summary for multi-site clinics in the United States.

The method for tracing an error to its source is the same in healthcare as anywhere. Our guide to fixing what ChatGPT gets wrong about your company sets out the capture, trace, fix and recheck steps, and the log to keep.

A worked example

The example shows the pattern we expect: a small number of stale sources causing most of the wrong answers. That is good news, because the fixes are mostly administrative.

What to do this quarter

1. Publish one page per location with hours, plans accepted, new-patient status and clinicians, in plain HTML text, and date it.
2. Reconcile every business profile, insurer directory and health directory against those pages.
3. Run a patient-style prompt set per location across the six engines, and repeat it on several days, because answers vary. Our measurement spec for AI visibility scores explains how many runs a reliable number needs.
4. Keep a dated error log, and recheck after every change in plans, hours or staff.

Our view: this is front-desk hygiene, not marketing. The cost of a wrong answer lands on the phone line and in reviews. Monitoring the answers needs no patient data, because the prompts are the ones any member of the public could type. Our healthcare providers page describes the prompt set we use for clinic groups.