What AI Search Actually Cites: 90 Days of Citation Data From 986 Pages

Most advice about getting cited by ChatGPT, Copilot, or Google AI Overviews is guesswork dressed up as strategy. So I pulled the actual numbers on our own site: 986 pages, 90 days, every citation Bing recorded. The pattern is clearer than I expected, and most of it contradicts what practices are being told to build.

Where this data comes from

Bing Webmaster Tools added an AI Performance report this year. It records every time Copilot or a Bing AI answer used one of your pages as a source, which page it was, and the "grounding query" the person actually typed. Because ChatGPT also retrieves live web results through Bing's index, this is the closest thing to a direct view of the ChatGPT channel that exists for free. Google Search Console has a separate report for AI Overview impressions, which I'll reference where it adds something.

The window is June 25 through September 23, 2026. The site is medicalmarketingfirm.com, 986 indexed pages, a mix of service pages, specialty pages, about 300 city and state pages, roughly 120 long blog guides, and a few hundred short question-and-answer pages. That mix matters, because it means the data can compare page types against each other on one domain, with one authority profile, one crawl setup, and one author.

The headline numbers

Daily volume roughly doubled across the window. July averaged about 14 citations a day. September is averaging 30, with 6 to 10 different pages cited on a typical day and a single-day high of 68. Google's AI Overview report for the same site shows the same slope, so this isn't a Bing quirk.

Finding 1: deep, single-specialty guides win, and it isn't close

Here are the eight pages that took 63% of all citations. Every one is a long guide about marketing one specific type of practice.

I had been telling clients to avoid the "complete guide" format because it tends toward generic breadth. The data says the format is fine. What matters is that the page is about one kind of practice and goes deep on that practice's specific economics, procedures, and patient behavior. A 2,500-word guide on marketing an orthodontics practice gets cited. A 2,500-word guide on "medical marketing" in general does not. The dermatology guide on our site, which is broader and older, earned zero citations in the same window. That one is getting rewritten.

For a practice, the translation is direct. Your procedure pages are your specialty guides. A rhinoplasty page that explains candidacy, technique differences, recovery week by week, revision rates, and what drives cost in your market is citable. A rhinoplasty page that says you offer rhinoplasty and lists three benefits is not.

Finding 2: citations follow questions, not pages

The fertility guide earned 136 citations over the full 90 days and only 10 in the last 28. The orthodontics guide went from 90 to zero. Nothing on either page changed. What changed is that a burst of people stopped asking the question those pages answer. In the fertility case, a single grounding query, "best marketing materials for promoting fertility benefits," accounted for 83 citations on its own, and our page held a 37% share of every citation Bing handed out for it. Then the query faded.

So no single page is a durable asset. The portfolio is. Twenty deep pages that each get cited when their question comes up is a real channel. One page that gets cited a lot for one month is a spike. This is also why I'd stop obsessing over any one guide's numbers and look at the count of distinct pages cited per day instead. Ours went from 1 or 2 in June to 6 to 10 now, and that is the number that tells me the channel is getting healthier.

Finding 3: location pages barely register

We have roughly 300 city and state pages. They exist for classic local SEO, and they do their job there. In AI search, 5 of them earned any citation at all, 19 citations combined, 1.5% of the total. The homepage, which ranks first on Google for our brand name, was cited exactly once.

The reason is simple once you see the grounding queries. People don't ask AI "medical marketing Naples FL." They ask "how do I get more Invisalign patients" or "SEO strategies for plastic surgery practices." A page whose entire reason for existing is a city name has nothing to contribute to that answer. The one geographic page that did get cited heavily, the New Jersey doctors guide with 64 citations, is a 3,000-word piece about how physician practices in that state actually acquire patients. It happens to be about a place. It isn't about ranking for a place.

For a practice, the equivalent mistake is the "botox in [suburb]" page farm. Those pages can still help you on Google Maps. They will not get you into an AI answer. Your Google Business Profile carries the local entity for AI; your content should carry the expertise.

Finding 4: AI trusts us to explain, not yet to recommend

This is the finding that changed our own plan. When our pages are cited, they take a large share of the answer: 20% to 37% of all citations for the top queries. But look at what those queries are. "SEO strategies for plastic surgery practices." "Cosmetic surgery SEO." "Scaling cosmetic dentistry with digital marketing." "Med spa SEO strategies." Nearly all of it is research and how-to. The queries where someone is deciding who to hire are almost absent. "How to choose a medical marketing firm in Los Angeles" produced 5 citations at a 36% share, and that's about the only buyer-intent query in the file.

In other words, AI has decided we are a credible source on how this work is done. It has not yet decided we are the answer to "who should I hire." Those are different questions and they get answered by different pages. The first is answered by guides. The second is answered by comparison content, "how to choose" content, pricing transparency, and third-party signals like reviews and mentions on other sites.

Practices have the exact same gap, and most don't know it. Your procedure pages can earn you a spot in "what is a deep plane facelift" answers all day long. The query that books a consult is "best facelift surgeon in Westchester" or "how do I choose a facial plastic surgeon," and the pages that win that one are about credentials, patient selection, before-and-after depth, and what a good consultation looks like. If you only have the first kind of page, AI will quote you and send the patient to someone else.

Finding 5: two blackouts, and what caused them

There are two windows in the data with zero citations and zero cited pages: July 28 to August 5, and August 21 to 23. Twelve days total. That isn't normal variance, a site that's getting cited 6 to 10 pages a day doesn't drop to zero pages by chance. It looks like the crawler could not fetch the site. In our case, the Cloudflare account had a managed rule blocking AI crawlers that we didn't know was on, and it wasn't fully corrected until September 9. The doubling in daily citations came after that fix, though I can't cleanly separate it from the guide updates we shipped in the same weeks.

If you host on Cloudflare, the "Block AI bots" setting was defaulted on for many accounts this year. Here is how to check in 60 seconds. A blocked site can rank perfectly on Google and be invisible to every AI assistant at the same time, and nothing in Search Console will tell you.

What to do with this

1. Check crawler access first. Cloudflare, robots.txt, and any security plugin. Everything below is wasted if GPTBot, OAI-SearchBot, Bingbot, and ClaudeBot get a 403.
2. Pull your own report. Verify your site in Bing Webmaster Tools (free, ten minutes) and open the AI Performance section. Export the queries. That list is the only real view of what people are asking AI about your specialty, and it will surprise you.
3. Build one deep page per procedure, not one thin page per city. Candidacy, technique, recovery timeline, risks, cost drivers, and how you decide who is a good fit. Specific numbers from your own practice are what get quoted.
4. Add the "how to choose" layer. One page per major procedure on how to evaluate a surgeon or provider for it, written honestly enough that it would be useful even if the reader picks someone else. This is the page that converts an AI citation into a consult.
5. Keep the top pages fresh. Whatever your five most-cited pages are, refresh them quarterly with a visible updated date. Citations decay when the page looks stale next to a newer competitor.
6. Track distinct pages cited per day, not total citations. Total citations swing with whatever question is trending. The count of different pages being cited is the health metric.

The honest caveats

This is one site, one niche, and one 90-day window. The Bing report only covers Copilot and Bing AI answers directly, with ChatGPT inferred through Bing's index. Google's AI Overviews cite a somewhat different set of our pages, with more weight on the homepage and state hubs, so the two systems don't fully agree. And the page-level report totals 1,248 citations while the daily overview totals 1,577 for the same window, which suggests the page report drops some low-volume pages. None of that changes the shape of the findings. Deep single-topic pages dominate, location pages don't, and the buyer-intent gap is real.

I'll republish this with the next 90 days of data. If the guide refreshes and the "how to choose" pages we're adding move the buyer-intent number, that will show up here.