AI Visibility in Healthcare: Stop Measuring Traffic

October 27, 2026

Last week, three studies landed with the same message: AI search sends almost no traffic. Rand Fishkin bundled them into one video, and if you run marketing for a health system, someone has probably forwarded it to you with a note asking whether you can cut the AI line from next year's budget.

Don't. The studies are right about traffic. They're measuring the wrong thing for healthcare.

Key Takeaways

AI search sends little referral traffic, and both its loudest skeptics and its loudest advocates now agree that traffic is the wrong way to measure it.
In healthcare, the AI answers that matter are comparison questions, where one provider tends to get named and everyone else is invisible.
Google has started writing its own summaries of the pages it cites, so being cited no longer guarantees your message reaches the patient.
Page structure, not brand size alone, predicts which pages AI cites, which means smaller systems and practices can compete.
The right metric is share of answer: how often you are the named recommendation on the questions that drive patient volume.

What did the skeptics actually find?

Marie Haynes argued that GEO is a distraction. The SEO Works benchmarked AI referral traffic. Kevin Indig asked whether AI drives more direct traffic. The common thread: AI platforms send a sliver of the visits that Google search still delivers.

That finding holds up. If you judge ChatGPT the way you judge organic search, by sessions in GA4, it looks like a rounding error.

Fishkin's own conclusion is the important part. He didn't say ignore AI. He said budget it as a brand channel, not a traffic channel.

Why is traffic the wrong way to measure AI search?

Mike King of iPullRank answered the studies directly: they grade a brand channel with a traffic ruler. His rebuttal points to Pew Research data showing people click a result about 8% of the time when an AI Overview appears, versus about 15% when one doesn't. The answer itself is the destination now.

King also notes that most AI-influenced visits arrive looking like "direct" traffic, and that in Indig's own data, AI share of voice tracked closely with direct traffic a few weeks later.

Think about how patients actually behave. A woman asks ChatGPT which hospital near her is best for a hip replacement. It names one. She doesn't click a link. She types the hospital's name into Google, or calls the number. Your analytics records a branded search or a direct visit, and the AI answer that made the decision gets zero credit.

Why is healthcare different?

Not every AI question is equal. A Conductor study of 14,000 AI responses across 10 industries found that up to 72% of answers to educational questions named no brand at all. On comparison questions, a single brand led 91% of responses.

Map that to the patient journey. "What causes knee pain when climbing stairs?" names nobody. "Best orthopedic surgeon in Indianapolis" or "which hospital is better for knee replacement" names a winner. Those comparison questions sit right before a patient picks a provider, which is where patient volume is won or lost.

Retail can afford to be one of five brands in an answer. In healthcare, the patient usually acts on the first name they trust.

Is Google rewriting your page?

On October 23, SEO consultant Brodie Clark spotted something new, and Search Engine Roundtable confirmed it in its own tests: the citation cards next to some Google AI Overviews now show AI-written summaries of the cited page, not the page's own text. Google hasn't commented, and it isn't on every query yet.

Here's why that matters. Say your cardiology page and a competitor's both get cited. Google summarizes yours as "offers cardiac care services." It summarizes theirs as "top-rated TAVR program with same-week appointments." You both won the citation. They won the patient.

Being cited is no longer the finish line. What the AI says about you when it cites you is now part of the scorecard.

Can smaller health systems compete with Mayo in AI answers?

Yes, on more queries than most teams assume. The usual excuse for losing in AI answers is brand size: Mayo wins because it's Mayo. A new SISTRIX study tested that directly.

SISTRIX compared 96,418 cited and non-cited pages, matching each cited page with a similar page from the same website, across ChatGPT, Google AI Mode, and AI Overviews. Because both pages share a domain, brand strength is largely taken out of the equation. What predicted citation was page structure:

A table of contents raised the odds of citation about 1.7 times
Article schema, and five or more H2 subheadings, about 1.6 times each
Genuine FAQ schema, about 1.5 times
An author box with a photo, about 1.4 times
A visible last-updated date, about 1.3 times

Pages with many of these features were cited about 2.1 times as often, yet only 10% of cited pages had them. llms.txt showed no benefit at all.

Two caveats: the data comes from German-speaking markets, and it shows correlation, not cause. It still points the same direction we see in health system scans. Condition and treatment pages rarely show an updated date, a named clinical author, or a structure AI can lift an answer from. Those are fixable gaps, and none of them requires being Mayo.

What should health systems measure instead?

Stop reporting ChatGPT referral sessions to leadership as the AI number. Both sides of this debate now agree it tells you almost nothing. Measure these instead:

Share of answer. On the comparison questions for each service line, how often are you the named recommendation, and who is named when you aren't?
Answer accuracy and framing. When AI mentions you, is the information right, and does the summary give a patient a reason to choose you?
Cross-platform consistency. ChatGPT, Gemini, Perplexity, Google AI Overviews, Grok, and Claude pull from different sources. Leading on one platform says little about the others.
Repeat measurement. Seer Interactive ran the same ChatGPT prompt twice, eight minutes apart, and the two runs shared zero search queries. One spot check proves nothing. You need a large query set, run every cycle.

That's why AHS measures 75 queries per service line across six platforms, 450 data points per cycle. A single answer is anecdote. Hundreds of answers is a trend you can manage.

FAQs

Does AI search send meaningful traffic to hospital websites?

Not much referral traffic, and multiple studies confirm it. But AI answers influence patients who then search your name or call directly, so that impact shows up as branded and direct traffic rather than AI referrals.

How should a health system measure AI visibility?

Measure share of answer: how often your organization is the named recommendation on the comparison questions patients ask for each service line. Track it across ChatGPT, Gemini, Perplexity, Google AI Overviews, Grok, and Claude, and repeat it every cycle, because single checks are unreliable.

Can a smaller health system or independent practice show up in AI answers over Mayo Clinic?

Often, yes. Research comparing pages on the same website found that page structure predicts AI citation independent of brand strength, and local signals like reviews and Google Business Profile accuracy drive "near me" answers where national brands have little advantage.

What makes a healthcare web page more likely to be cited by AI?

Clear structure. Tables of contents, descriptive subheadings, Article and FAQ schema, a named author, and a visible last-updated date are all associated with higher citation rates in a study of more than 96,000 URLs.

Is llms.txt worth implementing for a hospital website?

It isn't a priority. A large 2026 study found no citation benefit from llms.txt, while basic page structure showed clear gains.

Find out what AI is telling your patients

If you don't know whether ChatGPT names you or a competitor when patients ask about your top service lines, that's the first question to answer. Request your complimentary AI Readiness Scan.

Want this kind of analysis every week? Subscribe to The AI Visibility Brief.

Three Studies Say AI Visibility Doesn't Matter. None of Them Asked Healthcare's Question.