AI Overviews Are Eating Your Practice's Website Traffic. Here's What to Measure Instead.

A practice owner called me three weeks ago, genuinely worried. Her Google Analytics showed website sessions down 31 percent year over year. She had not changed anything. Same SEO vendor, same ad budget, same content. By every number she knew how to read, her marketing was failing.

Then I asked her one question. Was the phone ringing less?

It was ringing more. New patient appointments were up. The front desk was busier than it had been in two years. Her traffic was down and her practice was growing at the same time, and nobody had told her those two things could both be true.

They can. They are, in a lot of practices right now. The reason is AI Overviews, and the deeper problem is that the number she was watching stopped measuring what she thought it measured.

What Actually Changed

For fifteen years, website sessions were a reasonable proxy for interest. Someone wanted to find a dentist, they searched, they clicked your site, that click logged as a session. More interest, more sessions. The proxy held.

Google’s AI Overview broke the proxy. Now a patient searches “how long does a dental implant take to heal” and Google answers the question at the top of the page, in text, pulling from sites like yours without sending the click. The patient gets the answer. You get nothing in your analytics. The session that used to fire never fires.

This is not a small effect. For informational health queries, the kind patients run constantly while researching care, a large share of searches now end without a click to any website. Your content may be the source the AI quoted. You will not see it as traffic.

So the practice that is doing well at answering patient questions can watch its sessions fall precisely because it is doing well. The better your content answers the question, the more likely the AI uses it to satisfy the patient on the results page. You earned the trust and lost the click.

Raw sessions cannot tell the difference between “fewer people care about us” and “AI is answering on our behalf.” Those require opposite responses. One means fix your marketing. The other means your marketing is working and your dashboard is lying. You cannot run a practice on a number that conflates them.

The Metrics That Still Tell the Truth

I spent most of my career in analytics before I did marketing, including federal data work where being wrong about a number had consequences. The discipline that carries over is simple. When a metric stops correlating with the outcome you care about, you stop reporting it and you find one that still does. Here is what I have practices track now instead of, or alongside, raw sessions.

Booked appointments and qualified calls, measured at the source. This is the only number that was ever real. Patients in the chair. If you are not already capturing booking-form completions and tracking inbound calls with a system that logs them, that is the first build, ahead of everything else on this list. Call tracking that ties a phone call back to how the patient found you turns “the phone is ringing more” from a feeling into a number you can defend.

Branded versus non-branded search demand. Pull this from Google Search Console, not Analytics. When people search your practice name directly, that is demand AI Overviews cannot intercept, because there is no generic question to answer. If branded impressions and clicks are rising while total sessions fall, that is a healthy practice with a shrinking informational-traffic tail. That is the pattern you want. If branded demand is also falling, then you have a real problem, and now you know it is real.

High-intent page engagement, not homepage volume. Stop counting visits to your blog and start counting actions on the pages that sit next to a decision. Views of your “new patient” page, your insurance and financing page, your specific procedure pages, your contact page. A practice can lose 30 percent of its top-of-funnel informational sessions and gain bottom-of-funnel intent at the same time. Segment your analytics so you can see the bottom of the funnel on its own, because that is where the revenue lives and it is the part AI Overviews touches least.

AI-assisted and AI-referred traffic. This is new and most practices are not looking. Patients increasingly research in ChatGPT, Perplexity, and Gemini, then come to your site directly or through a referral link the AI provided. In GA4 you can see the front edge of this by isolating referral traffic from chatgpt.com, perplexity.ai, and gemini.google.com, and by watching for a rise in direct traffic that lands on deep interior pages rather than your homepage. A patient who lands straight on your dental-implant page with no referrer often came from an AI answer that named you. Build a segment for it and watch it grow.

Conversion rate of the traffic you do get. When the informational tire-kickers stop showing up because the AI handled their early questions, the visitors who remain are further along and more ready to act. Your session count drops and your conversion rate should climb. If sessions are down 30 percent and conversions are flat, you are converting better per visit, which means the lost traffic was low value. Watching conversion rate move inversely to sessions is how you confirm that the traffic you lost was the traffic you did not need.

How To Set This Up Without a Six-Month Project

None of this requires a new platform. It requires pointing your existing tools at the right things.

In GA4, build three audiences or comparisons and save them so you stop looking at the blended number. One for high-intent interior pages. One for AI-referral and deep-direct traffic. One for converters. In Search Console, save a filtered view for branded queries so you can separate demand for you from demand for answers. Add call tracking if you do not have it, because a practice that cannot count its phone calls is flying with half its instruments covered. Then change your monthly review so the first number you read is booked appointments and the last number you read is raw sessions, instead of the other way around.

That reordering is most of the fix. The data was never the problem. The hierarchy of attention was.

What Good Looks Like Now

A healthy practice in 2026 can show falling total sessions, rising branded search, flat or rising interior-page engagement, growing AI-referral and deep-direct traffic, a climbing conversion rate, and more appointments on the books. Read the old way, top line down, that looks like decline. Read the right way, bottom line up, it is exactly what growth looks like when the search layer changed underneath you.

The practices that get hurt over the next year will not be the ones losing traffic. Every practice is losing that kind of traffic. The ones that get hurt are the ones that panic at the wrong number, cut the content and the spend that were actually working, and dismantle a healthy marketing engine because the dashboard they inherited was built for a search engine that no longer exists.

Measure the patients, not the pageviews. The pageviews were always just a stand-in for the patients, and the stand-in stopped standing in.

---

William Hunt is a healthcare technology consultant who runs HuntGrowth, a marketing practice for independent medical and aesthetic practices. His background combines computer science and federal data and analytics work (Department of Defense, the White House, AARP) with fifteen years of marketing execution. He works with one to three practices at a time on measurement, AI search visibility, and patient acquisition.