Here is the week that gave it away. From around October 8 to October 15, 2026, during the rollout of GPT 6 and its new interface, the links inside ChatGPT answers stopped carrying their usual tag. Normally a click from ChatGPT arrives with a small label, utm_source=chatgpt.com, that tells your analytics where it came from. For about a week those tags went missing, and so did the referrer data behind them.
The result was clean enough to see across the industry. One analysis of roughly 500 pages found ChatGPT attributed visits fell by about 31,000 a day while direct visits rose by about the same amount, with total traffic basically flat, as reported by Marketing4eCommerce. Nothing had actually changed about where people came from. The only thing that changed was whether the tracking could see it. OpenAI called it a bug and began fixing it on October 15, bringing tagged links back to between 95 and 100 percent of normal.
So the crisis passed. But it accidentally proved something that never passes: a lot of the patients AI sends you have always landed in your reports as direct, and always will.
Why AI visits hide in the direct bucket
Direct traffic was always supposed to mean the visitor had no trackable source. You type the address, you use a bookmark, you tap a link in a text message. Google Analytics shrugs and calls it direct. The trouble is that a lot of AI traffic now behaves exactly like that, for a few plain reasons.
- People copy and paste. A patient reads your name in a ChatGPT answer, copies the web address, and pastes it into a fresh browser tab. That opens a brand new session with no referrer at all. Analytics has no choice but to call it direct.
- Apps strip the referrer. When the tap happens inside the ChatGPT phone app, the browser that opens often receives no referrer header. Same story with AI features baked into other apps and browsers. The trail is gone before it reaches you.
- Privacy settings cut the thread. No referrer policies and secure to insecure page jumps can both wipe the source, so even a real click arrives looking anonymous.
Add it up and the undercount is large. Independent write ups, including this breakdown of dark traffic, estimate Google Analytics misses somewhere between 30 and 50 percent of AI driven visits. In one 2026 client dataset, up to 30 percent of leads said they found the brand through an AI tool, while only a sliver of that showed up as AI referrals in the analytics. The machine is doing more for you than your charts will admit.
Why this matters more for a medical practice than for a shoe store
Patients research doctors harder than almost anything else they buy, and they have moved to AI fast. The 2026 Patient Choice Report from rater8 found 47 percent of patients have now used an AI tool to research a provider, up from 31 percent only nine months earlier, and AI has become the single most named influence on which provider they pick, ahead of both Google search and a referral from another doctor. That is a huge share of your new patient flow riding on a channel your dashboard keeps mislabeling.
Picture the real sequence. A woman asks ChatGPT for a good dermatologist in her city who takes her plan. It names yours. She copies your web address, pastes it, reads your page, and books. In your analytics she is a direct visit, indistinguishable from a repeat patient checking your hours. If you judge your marketing by that report, you will conclude AI did nothing and your website just magically grew its direct traffic. Both conclusions are wrong, and both are expensive.
The budget mistake this causes
Here is where it stops being a trivia fact and starts costing money. When the patients from search and AI pile up in the direct column, the work that earned them looks like it produced nothing. We have watched owners stare at a report, see a thin line next to organic and almost nothing next to AI, and decide to cut the exact investment that was filling their schedule. They are not being careless. They are trusting a number that is lying to them.
The fix is not to distrust all your data. It is to stop treating the analytics dashboard as the whole truth and start cross checking it against what patients tell you directly. We dug into why the quick answer at the front desk can be misleading too in whether you should trust your how did you hear about us data, and the honest version is that no single source is complete. You triangulate.
How to actually see the patients AI sends you
You will never get this to a perfect number, and you do not need to. You need enough signal to make good decisions. Three layers get you most of the way there.
One, label the AI traffic you can see. In Google Analytics you can build a custom channel group that pulls known AI sources like chatgpt.com, perplexity.ai, and gemini.google.com out of the generic pile and names them plainly. It will not catch the copy and paste crowd, but it turns the visible portion from a guess into a number you can watch grow. Guides like this one on tracking AI referral traffic walk through the setup, and we cover the practice specific version in how to track AI traffic to your medical practice.
Two, just ask. A how did you hear about us field on your booking form and intake, with a real AI assistant option on the list, catches exactly the visits your analytics cannot. Self reported data is messy, but messy and visible beats precise and invisible. This is the single cheapest upgrade most practices can make this week.
Three, train the front desk. One question to every new caller, how did you find us, written down, not guessed. Over a month that handwritten tally will tell you more about AI and search than any export, because it captures the patient who never left a trail at all.
The bigger lesson underneath the bug
The October tagging glitch got fixed in a week. The thing it revealed did not get fixed, because it is not a glitch. As patients keep moving to AI to find doctors, a rising share of your most valuable traffic will keep arriving unlabeled. That cuts two ways. It means you are probably underestimating your results. It also means your competitors are too, and the ones who figure out how to measure it, even roughly, will make smarter moves than the ones flying on a dashboard that says direct.
The deeper point is that being the name AI gives out is now worth real money, whether or not you can perfectly count it. If you want the companion piece on how those tools decide who to name in the first place, we laid it out in how AI decides which doctor to recommend.
How EtherealMinds handles this
We work only with healthcare practices in the United States, so this comes up constantly now. An owner tells us their direct traffic is climbing and they have no idea why. Nine times out of ten the why is AI and search doing their job without getting credit. Our part is twofold. We set up the tracking so the visible slice of AI and organic traffic is labeled and watched instead of dumped into one gray column, and we build the thing that earns those visits in the first place.
That means a website wired with real analytics and clean conversion tracking so you can tell a booked patient from a bounce, the local and AI search work that gets your practice named in the answers patients are already reading, and the full patient acquisition system that turns one of those clicks into a booked appointment. You do not get to control how AI tags its links. You do get to control whether you are measuring the right things and earning the visit in the first place.
Not sure where your patients really come from?
Book a free strategy call and we will look at your analytics with you, show you where AI and search traffic is likely hiding in your direct numbers, and set up tracking that gives you an honest picture of what is actually bringing patients through the door.
Book a free strategy call →Frequently asked questions
Why does ChatGPT traffic show up as direct in Google Analytics?
When a patient taps a link inside the ChatGPT app, or copies a web address out of an answer and pastes it into their browser, the visit often arrives with no referrer and no campaign tag. Google Analytics has nothing to attribute it to, so it files the session under direct traffic, the same bucket as someone typing your address from memory. The patient really came from ChatGPT, but your dashboard cannot see it.
What was the October 2026 ChatGPT tracking bug?
Roughly from October 8 to October 15, 2026, during the rollout of GPT 6 and its new interface, many outbound links in ChatGPT answers stopped carrying the utm_source=chatgpt.com tag and passed no referrer. Reported ChatGPT referral traffic fell by about 31,000 visits per day while direct traffic rose by about the same amount. OpenAI called it a bug and rolled out a fix starting October 15, restoring tagged links to between 95 and 100 percent of normal. The leak it exposed, however, was never fully fixed, because a lot of AI traffic has always landed as direct.
How much AI traffic does Google Analytics miss?
Independent analyses put it at roughly 30 to 50 percent of AI driven visits going uncounted, mostly because referrers get stripped inside mobile apps, people copy and paste links instead of clicking, and privacy headers block the data. In one 2026 client dataset, up to 30 percent of leads said they discovered the brand through an AI tool, while only a small fraction showed up as AI referrals in the analytics. The gap means AI is almost always doing more for you than your reports admit.
How can a medical practice track patients who come from ChatGPT?
Use three layers. First, build a custom channel in Google Analytics that groups known AI sources like chatgpt.com, perplexity.ai, and gemini.google.com so the visits you can see are labeled clearly. Second, add a how did you hear about us field on your booking form and intake, with an AI assistant option, since self reported data catches what analytics cannot. Third, train your front desk to ask new callers how they found you and write it down. No single method is complete, so the combination is what gives you a real picture.
Should I stop investing in SEO if my analytics show mostly direct traffic?
No, and this is the costly mistake the direct traffic problem causes. When AI and search visits hide inside the direct bucket, the work that earned them looks like it did nothing, so owners cut the budget that is actually bringing patients in. Before you judge any channel, line up your analytics with what new patients tell you at the desk. In most practices the honest picture shows search and AI pulling far more weight than the dashboard alone suggests.