Is AI Sending You Patients? The Intake Option and Five Referrer Hosts
Summary
Count them at the front door: add one option to the intake question "How did you find us?": an AI assistant, with the assistants named. Then count five referrer hosts in your web host's access log monthly: chatgpt.com, copilot.microsoft.com, perplexity.ai, gemini.google.com and claude.ai. Both numbers are floors: a patient who reads a name in an answer and then searches for it arrives with no trace of the assistant. A pixel would see more, and it is a method HHS has said can turn a page visit into a disclosure.
By Gale Editorial · Updated 2026-09-15. Every figure cited to a dated source. How we write.
Why the click reports miss most of it
Because an answer engine hands the patient a name more often than a visit. In Pew's March 2025 browsing data, a user who saw an AI summary clicked a traditional result on 8 percent of visits, against 15 percent without one 1Ref 1Athena Chapekis and Anna Lieb, Pew Research Center (2025).Google users are less likely to click on links when an AI summary appears in the results.Pew's March 2025 browsing data from 900 adults: 8 percent of visits with an AI summary clicked a traditional result against 15 percent without; the basis for saying an AI answer hands a patient a name more often than a visit.. The patient reads the name, closes the tab, and reaches you by phone or by a new search that carries no trace of where the name came from.
The click that does happen is filed out of sight. Google reports traffic from AI Overviews and AI Mode in the Search Console Performance report inside the Web search type, combined with ordinary search traffic 2Ref 2Google Search Central (2025).AI Features and Your Website.Google's statement that traffic from AI Overviews and AI Mode is reported in the Search Console Performance report under the Web search type, combined with other search traffic and not broken out.. A practice reading its Search Console clicks is reading a total that includes the AI clicks and cannot subtract them.
The patients are there to count. KFF's poll of 1,343 adults, taken February 24 to March 2, 2026, found that 32 percent had turned to an AI tool or chatbot for health information or advice in the past year 3Ref 3Alex Montero, Julian Montalvo III, Audrey Kearney, Isabelle Valdes, Ashley Kirzinger and Liz Hamel, KFF (2026).KFF Tracking Poll on Health Information and Trust: Use of AI For Health Information and Advice.KFF's February 2026 poll of 1,343 adults: 32 percent used an AI tool or chatbot for health information in the past year, and 58 percent of physical-health users later followed up with a provider.. Among those who asked about their physical health, 58 percent later followed up with a doctor or another provider. Some of those follow-ups land at a practice the assistant named, and nothing records it unless someone asks.
The intake option: one question, on the form and at the desk
Add one question to the new-patient form and the front-desk script, with an option that names the assistants. "How did you find us?" is on many forms already, with a list that stops at a search engine, a friend, an insurer's directory and a referring office. The list needs one more line: "An AI assistant (ChatGPT, Copilot, Perplexity, Gemini, Claude)". That line is the only instrument that catches a patient who was handed a name and clicked nothing.
Put the option on the form the patient completes before the first visit and in the desk's phone script, so the question is asked the same way however a patient arrives. Keep "Other" with a blank beside it, because a patient who says "I asked Google" may mean an AI Overview, and the blank is where the assistant's name gets written. And tally it as counts. The monthly number is how many new patients chose each option, and it carries no names.
Write the tally as a row, one per month: new patients, then the count for each option. A single month means little. But six rows show whether the AI line is growing, and set against the referrer-host count below they show how far apart the two measures sit. That gap is roughly the share of AI referrals that reach you with no trace in any log.
Five referrer hosts to count in the access logs
When a patient does click, the browser usually tells your server where the click came from, in a header called the referrer, which your web host keeps in its access log. Turn the logs on at the host, download a month, and count visits whose referrer host is one of five: chatgpt.com, copilot.microsoft.com, perplexity.ai, gemini.google.com and claude.ai. The count is a floor: a patient who reads a name and searches for it arrives from the search engine.
| Referrer host | Assistant | What the visit means |
|---|---|---|
| chatgpt.com | ChatGPT | A person clicked a link in a ChatGPT answer |
| copilot.microsoft.com | Microsoft Copilot | A person clicked a link in a Copilot answer |
| perplexity.ai | Perplexity | A person clicked a cited source in a Perplexity answer |
| gemini.google.com | Google Gemini (the app) | A person clicked a link in a Gemini answer; AI Overviews are not here |
| claude.ai | Claude | A person clicked a link in a Claude answer |
Many hosting control panels list raw access logs or a referrer report under a logs heading; a managed site builder may need the report requested from support. Filtering a month's log to those five hosts takes under an hour in a spreadsheet.
The same log holds a second signal. OpenAI says that when a user asks ChatGPT a question, it may visit a web page with a ChatGPT-User agent, and that this agent is not used for crawling the web automatically 4Ref 4OpenAI (2026).Overview of OpenAI Crawlers.OpenAI's description of ChatGPT-User: when a user asks ChatGPT a question it may visit a web page with that agent, it is not used for automatic crawling, and robots.txt rules may not apply; the basis for reading a ChatGPT-User line in an access log as a user-initiated fetch.. A ChatGPT-User line in your log is a person, at that moment, asking ChatGPT something that made it read your page. It is a fetch on a user's behalf and no visit by a patient, so it gets its own column: it says the practice is being asked about even in a month when nobody clicked through.
What Search Console and Bing Webmaster Tools add
Both consoles report AI visibility, and neither reports a patient. Google added a generative AI performance report to Search Console in June 2026, rolled out to every site by August 31, showing how often a site's URLs appeared in AI Overviews and AI Mode, by page, country, device and date 5Ref 5Hillel Maoz and Moshe Samet, Google Search Console (2026).Introducing Search Generative AI performance reports in Search Console.Google's June 3, 2026 announcement, rolled out to all sites by August 31, 2026, of Search Console reports showing impressions of a site's URLs in AI Overviews and AI Mode by page, country, device and date, with the data still included in the overall Performance report.. It lists impressions and pages. It lists no clicks and no queries, and the data stays inside the overall Performance report, so the totals there do not change.
Bing's AI Performance report in Bing Webmaster Tools counts total citations, the times a page was shown as a source in Copilot, Bing's AI summaries and partner integrations, and lists grounding queries, the phrases the AI used when it retrieved the cited content 6Ref 6Microsoft Bing Webmaster Tools (2026).AI Performance - Bing Webmaster Tools.What Bing's AI Performance report counts (total citations, cited pages per day, grounding queries) across Copilot, Bing AI summaries and partner integrations, and Microsoft's own limits: it does not measure rankings, authority or importance, and citation-trend changes cannot be attributed to a single content change or model update.. Microsoft says the report does not measure rankings, authority or importance, and that a change in the citation trend cannot be pinned on any one content change or model update.
Read both consoles monthly and write the numbers into the tally row as what they are: impressions and citations, one step upstream of a visit and two upstream of a patient. Take the Search Console reading and the Bing reading in the same sitting, and write both down that day, so the row is never half filled.
Why a pixel is the wrong fix
A tracking pixel would see more, and it is a method HHS has said can turn a page visit into a disclosure. The Office for Civil Rights bulletin on online tracking technologies says a regulated entity may not use tracking technologies in a manner that results in impermissible disclosures of protected health information, and that a tracking vendor receiving that information needs a business associate agreement; the bulletin's own example is an appointment-scheduling page 7Ref 7U.S. Department of Health and Human Services, Office for Civil Rights (OCR) (2024).Use of Online Tracking Technologies by HIPAA Covered Entities and Business Associates.OCR's guidance that tracking technologies may not be used in a way that results in impermissible disclosures of PHI and that a tracking vendor receiving PHI needs a business associate agreement, with the appointment-scheduling page as its example; and the page's own header note recording the June 20, 2024 vacatur of the unauthenticated-public-webpage portion and that HHS is evaluating next steps..
The bulletin's reach was narrowed in court. The page's own header note records that on June 20, 2024 a federal district court in Texas vacated the guidance to the extent it said HIPAA obligations arise when a tracker connects an IP address with a visit to an unauthenticated public page about specific health conditions or health care providers, and that HHS is evaluating next steps 7Ref 7U.S. Department of Health and Human Services, Office for Civil Rights (OCR) (2024).Use of Online Tracking Technologies by HIPAA Covered Entities and Business Associates.OCR's guidance that tracking technologies may not be used in a way that results in impermissible disclosures of PHI and that a tracking vendor receiving PHI needs a business associate agreement, with the appointment-scheduling page as its example; and the page's own header note recording the June 20, 2024 vacatur of the unauthenticated-public-webpage portion and that HHS is evaluating next steps.. The rest of the bulletin still stands on the page: authenticated pages such as a patient portal, the scheduling example, and the vendor agreement. A booking page, a form and a portal sign-in are where the tracking-pixel problem is live.
So the measurement is built without one. Remove third-party analytics and ad pixels from any page a patient uses to book, message, complete a form or sign in, and read the referrer from your own server log instead; the log sends nothing to anyone. That is pixel-free measurement: fewer numbers, every one of them first-party. A self-scheduling page is the very page the bulletin uses as its example, so check it first: load it with the browser's network panel open, filter to third-party domains, and look for anything that fires before the patient acts. Whether a particular script on a particular page amounts to a disclosure is a question for your compliance counsel. Whether you need the pixel to count AI referrals is the question here, and you do not.
The monthly tally row
One row a month, kept as counts. Write the month, the new patients who chose the AI option at intake, the visits from each of the five referrer hosts, the ChatGPT-User fetches, the Search Console AI impressions and the Bing citations. That row is the AI referral tally, and after six months it answers the question this page opens with for your practice, in your own numbers, with no pixel.
Keep the row with the practice's other operating numbers, and fold it into the quarterly audit of the five sources, so the tally sits beside the check that the address, the hours and the new-patient status still agree everywhere an assistant reads them. Gale holds the intake wording and the monthly tally, as counts and nothing else, alongside the practice's gale.care profile page, which carries no third-party pixel.
What the row cannot tell you is what the assistants said. For that, the monthly prompt panel is the instrument: ask each assistant the questions your patients ask and record whether the practice is named and what facts it is given. The tally says whether anyone came. The panel says whether you were offered. Read together, month by month, they are as close as a small practice can get to an answer engine's referral report.
Common questions
Run your practice on Gale
The software is free. Gale earns one flat 3.5% all-in per paid transaction — only on transactions that actually pay. No subscription, no setup fee, no network cut.
Start or manage a practice →References
- 1.Athena Chapekis and Anna Lieb, Pew Research Center (2025). Google users are less likely to click on links when an AI summary appears in the results. Pew Research Center. link ✓Pew's March 2025 browsing data from 900 adults: 8 percent of visits with an AI summary clicked a traditional result against 15 percent without; the basis for saying an AI answer hands a patient a name more often than a visit.
- 2.Google Search Central (2025). AI Features and Your Website. Google Search Central documentation (developers.google.com). link ✓Google's statement that traffic from AI Overviews and AI Mode is reported in the Search Console Performance report under the Web search type, combined with other search traffic and not broken out.
- 3.Alex Montero, Julian Montalvo III, Audrey Kearney, Isabelle Valdes, Ashley Kirzinger and Liz Hamel, KFF (2026). KFF Tracking Poll on Health Information and Trust: Use of AI For Health Information and Advice. KFF. link ✓KFF's February 2026 poll of 1,343 adults: 32 percent used an AI tool or chatbot for health information in the past year, and 58 percent of physical-health users later followed up with a provider.
- 4.OpenAI (2026). Overview of OpenAI Crawlers. OpenAI Developer Platform documentation. link ✓OpenAI's description of ChatGPT-User: when a user asks ChatGPT a question it may visit a web page with that agent, it is not used for automatic crawling, and robots.txt rules may not apply; the basis for reading a ChatGPT-User line in an access log as a user-initiated fetch.
- 5.Hillel Maoz and Moshe Samet, Google Search Console (2026). Introducing Search Generative AI performance reports in Search Console. Google Search Central Blog. link ✓Google's June 3, 2026 announcement, rolled out to all sites by August 31, 2026, of Search Console reports showing impressions of a site's URLs in AI Overviews and AI Mode by page, country, device and date, with the data still included in the overall Performance report.
- 6.Microsoft Bing Webmaster Tools (2026). AI Performance - Bing Webmaster Tools. Bing Webmaster Tools Help Center. link ✓What Bing's AI Performance report counts (total citations, cited pages per day, grounding queries) across Copilot, Bing AI summaries and partner integrations, and Microsoft's own limits: it does not measure rankings, authority or importance, and citation-trend changes cannot be attributed to a single content change or model update.
- 7.U.S. Department of Health and Human Services, Office for Civil Rights (OCR) (2024). Use of Online Tracking Technologies by HIPAA Covered Entities and Business Associates. HHS.gov, HIPAA for Professionals, Privacy guidance (bulletin issued Dec. 1, 2022; revised Mar. 18, 2024; content last reviewed June 26, 2024). linkOCR's guidance that tracking technologies may not be used in a way that results in impermissible disclosures of PHI and that a tracking vendor receiving PHI needs a business associate agreement, with the appointment-scheduling page as its example; and the page's own header note recording the June 20, 2024 vacatur of the unauthenticated-public-webpage portion and that HHS is evaluating next steps.
https://www.gale.care/for-providers/aeo-count-ai-referrals · 7 sources. Competitor details are cited to dated public sources and maintained as they change; figures are estimates, not commitments. Synthetic demonstration.