GLP-1 Clinics: Measuring ChatGPT and AI Referrals

If you run a GLP-1 clinic, AI assistants are almost certainly sending you visitors right now, and your reports are almost certainly hiding them. GLP-1 medication is exactly the kind of topic people take to ChatGPT, Perplexity and the rest, which makes a weight loss clinic likelier than nearly any other practice to be receiving assistant referrals it cannot see. Curve keeps the referring hostname as the traffic source, so chatgpt.com and perplexity.ai show up as their own rows in your sources report, and Curve Analyst lets you ask what those visitors did once they arrived.

TL;DR

• People ask assistants about GLP-1 eligibility, side effects, cost, and where to get it, and the assistant links to the clinic pages it cited. Those are usually your cost, eligibility, comparison, and FAQ pages, not your homepage.
• The clicks arrive with a referrer hostname such as chatgpt.com or perplexity.ai. Some arrive with no referrer and look like direct traffic, so whatever you count is a floor.
• These visitors arrive later in their decision and with a narrower question than a search visitor. Judge them on booking rate, not on session count.
• The standard fix, a custom channel group in GA4, requires GA4, and Google does not sign a BAA for it. A clinic site should not be running it at all.
• Curve records the referrer hostname server-side under a BAA, each assistant appears as its own source, and Analyst answers landing page and conversion rate questions about them in plain language.

Why does a GLP-1 clinic get more AI assistant referrals than it can see?

Because the questions people have about GLP-1 medication are exactly the kind they now put to an assistant instead of a search engine. Am I eligible. What are the side effects. How much does it cost per month if insurance will not cover it. How does one medication compare to another. Where can I get it near me, and how fast. An assistant answers all of that in one conversation where a search engine would have demanded a string of queries.

The assistant cites sources, frequently clinic pages, and a visitor who wants more than the summary clicks through. That click lands in your analytics like any other referral, which is the whole problem. Default channel rules were written before assistants sent traffic, so chatgpt.com drops into the catch-all referral bucket next to a link from a supplier. Referral looks small, and the Monday meeting moves on.

How the clicks arrive

Most of them carry a referrer hostname. Click a link inside a ChatGPT answer in a browser and the request to your site names chatgpt.com as the referrer. Perplexity sends perplexity.ai. Any tool that stores the raw hostname has these visits somewhere; the question is whether it surfaces them.

Some clicks arrive with no referrer at all. Native apps, redirects that drop the header, and privacy settings on the visitor's device all produce a visit that looks like someone typed your URL. Treat any AI referral count as a floor, and treat an unexplained rise in direct traffic to your cost page as a hint rather than noise.

Which pages do AI assistant visitors land on?

Rarely the homepage. The assistant links to the page it pulled the answer from, and the answer to "how much does GLP-1 treatment cost at a clinic" comes from a cost page, not from a hero banner. So the landing pages for this traffic cluster around your informational content:

• Cost and pricing pages, including anything that explains what is and is not included in a monthly fee.
• Eligibility pages: who qualifies, what intake looks like, what rules someone out.
• Comparison pages, whether that is one medication against another or your program against a subscription telehealth service.
• "Is it safe" and side effect FAQs, the page a nervous person reads before they decide to talk to anyone.
• Logistics pages: how appointments work, how medication is supplied, how soon someone can start.

Those pages were probably written to inform rather than to convert. Plenty of clinics leave the FAQ with a footer link and nothing else. If the assistant is sending people to the FAQ, that is where the booking path has to be visible. The guide to GLP-1 clinic landing pages that convert without collecting PHI covers what such a page should look like. You cannot fix a page you do not know is receiving the traffic.

Why do AI assistant visitors behave differently from search visitors?

They arrive later in the decision. Someone who searched "GLP-1 clinic near me" is at the start, scanning a list of options. Someone who spent a long conversation asking an assistant about eligibility and cost, then clicked the clinic the assistant cited, has already done the comparison. They arrive with a narrower question: do you take patients like me, and how soon can I start.

That changes what a good page looks like for them, and how you should judge the traffic. A visitor who already knows the rough cost and the eligibility rules does not need your explainer. They need the next step to be obvious. Bounce rate tells you less than usual here: someone who got their answer and booked from one pageview looks like a bounce.

The raw count will also underwhelm you. Search will send more sessions than assistants for a long time, so compare on count and assistants lose. Compare on booking rate and the picture can look quite different, and only your own numbers will tell you which way it goes.

What should a GLP-1 clinic measure?

Three views, and volume is the least useful of them.

Landing pages by assistant

Which pages do visitors from chatgpt.com land on, and are they the same pages visitors from perplexity.ai land on? This is your closest available proxy for what the assistants are citing. If both send people to your cost page and neither to your eligibility page, the eligibility page is either not being found or not worth citing, which is a content problem you can act on. If one sends people to a blog post you forgot existed, that post needs a booking path this week.

Booking rate against organic search

Take the goal you care about, a booked consult or a submitted intake form, and compare the rate at which AI assistant visitors reach it against the rate for organic search and paid search visitors. Per goal, per source. Then run it per landing page, because a low overall rate is often one page with no next step dragging down the rest. If the rate trails search, look at the pages before you conclude the traffic is weak.

Source split over time

Assistants are not interchangeable, and they change what they cite. Track chatgpt.com, perplexity.ai, and the others as separate sources and watch the split month over month. A source that goes quiet usually means the assistant stopped citing the page it was linking to, sometimes because you edited it. A source that appears from nowhere means something you published got picked up. Neither shows up in a channel-level report.

Why is the GA4 custom channel fix off the table?

Every guide to measuring AI traffic tells you to open GA4, build a custom channel group, and add a rule matching assistant hostnames. The mechanics work. The platform does not, for a clinic. Google does not sign a BAA for Google Analytics 4, and under HIPAA a vendor that handles protected health information on your behalf is a business associate that needs one. On a GLP-1 clinic site, the pages a visitor reads (eligibility, side effects, the cost of a specific medication), the form they submit, and the identifiers GA4 collects alongside them can say something about that person's health. The FTC has also pursued health data disclosures to ad platforms on unfairness grounds, as in the Hims & Hers complaint, so the exposure does not end at HIPAA.

A custom channel group puts a new label on data that should never have been collected that way. The fuller argument, and the alternative, is in how to track ChatGPT and Perplexity referrals without GA4.

How do Curve and Analyst answer the question?

Curve records the referrer hostname on its own servers when the pageview arrives. Events go to Curve's US-hosted infrastructure first, where anything that should not leave is stripped before conversions are forwarded to ad platforms. Curve signs a BAA with every customer on every plan. No GA4 tag is involved anywhere in the path.

Because the hostname is kept as the source, chatgpt.com and perplexity.ai appear as their own rows in the sources report, alongside google and facebook. At the channel level they count as referral today; there is no separate AI assistant channel yet. The sources report and a source filter are how you see them as a group, and the same source is carried through to goals and funnels, which is what lets you compare booking rate for assistant visitors against search visitors.

Where Analyst fits

Curve Analyst is the chat inside the Curve dashboard. Ask how much traffic came from chatgpt.com this month against last month and you get a stat card and a trend line; ask which pages those visitors landed on and how many booked, and you get a table. It can filter by source, so "visitors from AI assistants" is a question it can answer now without waiting for a dedicated channel.

Every number it gives is the result of a query against your own account. It never estimates. If a metric is empty it says so, if data is still syncing it says so before the number, and it treats "unknown" and "zero" as different things, which matters when you are asking whether a source went quiet or was never measured. It runs on Claude through Amazon Bedrock, inside infrastructure covered by Curve's BAA with AWS. Your data is not exported to a separate AI product and is not used to train models. We built it read-only: it cannot change tracking or destinations, and it does not look up individuals. It keeps the date range and filters of the screen you opened it from. More in what Curve Analyst is and how it works.

What should you write so assistants cite your clinic?

Pages that answer the exact question a person types into the assistant. Assistants cite content they can pull a direct answer from, so a page titled with the question and opening with the answer does better than a services page that mentions the topic in passing. For a GLP-1 clinic that means a plain cost page stating what the monthly fee includes, an eligibility page that says who qualifies, an honest side effect FAQ, and a comparison page for whichever question your intake staff hear most.

Being cited and measuring the click are separate jobs. The citation side has its own playbook in how healthcare practices get cited by ChatGPT, Perplexity, and Claude, and the local visibility side, which still feeds many assistant answers to "near me" questions, is in weight loss clinic local SEO for GLP-1 searches. Measurement tells you which pages are earning citations and bookings, which is how you decide what to write next.

What to ask Curve Analyst

Open Analyst from the sources report so it inherits your date range, then start with these three. The first is your headline number, the second tells you which pages to fix, and the third tells you whether the traffic is worth chasing.

• "How much traffic came from chatgpt.com and perplexity.ai this month vs last month?"
• "Which pages do visitors from AI assistants land on, and how many of them book a consult?"
• "What is the conversion rate of AI assistant visitors compared to organic search this quarter?"

Analyst answers with a chart and says which goal or source it is quoting. The announcement, Introducing Curve Analyst, lists what it can and cannot answer.

Frequently asked questions

Does ChatGPT send a referrer when someone clicks a link to my clinic?

Usually, yes. A click from the browser version of ChatGPT arrives with chatgpt.com as the referrer hostname, and Perplexity sends perplexity.ai. Clicks from native apps or through redirects can arrive with no referrer and get counted as direct, so the figure you see is a minimum.

Why does my current analytics tool show no AI traffic at all?

Because it is filing those visits under the generic referral channel. Open that channel, break it down by source, and you will probably find chatgpt.com in the list. The traffic was there; the report never named it.

Can I build an AI channel in GA4 instead of switching tools?

You can build it, and it will label the traffic correctly. The problem is that GA4 itself is not covered by a BAA from Google, so a clinic site should not be sending patient browsing data to it regardless of how the channels are named.

Can Analyst tell me what question the visitor asked the assistant?

No. The referrer carries the hostname, not the conversation, and Analyst does not look up individuals or answer person-level questions. Your best proxy is the landing page: the page the assistant cited tells you roughly what it was asked.

Will AI assistants show up as their own channel in Curve?

Today they show up as separate sources and roll up into referral at the channel level. Whether a dedicated AI assistant channel gets added is an open product decision. For now, filtering by source in the sources report, or asking Analyst to, gives you the same answer.

If your GLP-1 clinic is being cited by assistants and your reports cannot show it, book a demo of Curve and ask Analyst the three questions above.