The booking that came from ChatGPT and never showed up in your analytics

A guest tells you at check-in that they found you through ChatGPT. That evening you open Google Analytics, filter for chatgpt.com as a source, and find two sessions for the whole month. Neither booked. The guest is not lying and your analytics is not broken. They are measuring different things.

A booking from ChatGPT fails to show up for one of three reasons. The click arrived with no referrer and was filed as Direct. No click happened, because the answer gave the guest everything they needed to phone you or book through a third-party listing. Or the click came days later through a branded Google search, and Organic took the credit. Each is a different problem. Only one is partly fixable.

Why a booking from ChatGPT never shows up in your analytics

The referrer is missing

Analytics tools work out where a visitor came from by reading the referrer header the browser sends when someone follows a link. Many AI referrals arrive without one. Links opened from a mobile app, a desktop app, an in-app browser, or pasted into a new tab tend to lose it. A session with no referrer, no UTM parameters and no click identifier is Direct by definition.

So that visit sits in the same bucket as someone who typed your address from memory. The tool was never given the information.

No click happened

An assistant's answer often contains the whole decision. Your name, your neighbourhood, a line about what you do well, sometimes your phone number. There is nothing left to click.

This is not a rare edge case. The Semrush ghost citations study found 61.7 percent of citations produced no brand mention at all, and that ChatGPT cited a source 87 percent of the time while naming a brand 20.7 percent of the time. Citation and mention are separate events. When you are named in prose with no link beside your name, the guest has your name and nothing to click. Analytics only counts clicks.

For local businesses there is a further layer. In an independent test of 2,880 prompts (Ben Fisher, August 2026), Yelp data was grounding ChatGPT's local results in about 96 percent of runs, following OpenAI's non-exclusive licensing deal with Yelp in July 2026. The details a guest acts on may be Yelp's copy of yours, and the tap that follows may go to Yelp, to Maps, or to the phone dialler. None of those routes touch your site.

The click came later, under a different name

The third path is the oldest problem in marketing wearing a new coat. Someone reads the answer on Tuesday, searches your name on Thursday, clicks the top result and books. Organic Search gets the credit, or Google Business Profile does if they went through Maps. ChatGPT appears nowhere in the chain. Last-click attribution was never designed to look further back.

What your Direct traffic is hiding

Direct is where a tool puts everything it cannot classify. AI referrals share the bucket with bookmarks, typed URLs, links from email clients that strip referrers, and links from messaging apps. So when Direct rises, the honest reading is that something unclassified rose. You cannot subtract your way to an AI number.

There is one tell worth watching. Direct sessions that land on a deep page, a specific room, a menu, a single tour itinerary, are almost never typed. Those are followed links whose origin was lost somewhere between the answer and your server. That is a hint about the trend, not evidence about the count, and we would not present it to a client as more.

Keep the scale in mind. Conductor's benchmark puts AI referrals at roughly 1 percent of traffic. Ahrefs reported AI referrals at 0.5 percent of its own site's traffic and 12.1 percent of its signups. That is one company's figure, but the shape is worth noting: a small share of sessions, a larger share of outcomes, and the sessions already undercounted.

Pew Research tracked 900 US adults across 68,879 Google queries and found users clicked a traditional result on 8 percent of searches with an AI summary against 15 percent without. Only 1 percent clicked a link inside the summary. Even on Google, where the referrer usually survives, the click is the minority outcome. Being answered without a visit is the normal case. Your analytics was built for the other one.

What you can know

The referrals that carry a source

Some AI clicks do arrive with a referrer. Build a channel group in GA4, or whatever you run, that matches chatgpt.com, perplexity.ai, copilot.microsoft.com, claude.ai and gemini.google.com as referring domains. This is the first thing we set up, because it costs nothing and gives you a floor. The true number is above it and you do not know by how much.

The first-party reports

Bing Webmaster Tools added an AI Performance report in February 2026 showing Copilot citations and the grounding queries the model generated. Google Search Console's Generative AI report, global since August 2026, shows impressions only. Both describe visibility, which is a different question from bookings. How to track AI search visibility covers the setup.

The "how did you hear about us" field

This is the only instrument that can see the no-click path, and most businesses waste it.

Make it a fixed list rather than a free-text box, and put "ChatGPT or another AI assistant" on it as a named option. Ask it on the booking form, in the reservation confirmation, in the phone script and at check-in. A free-text box gets "online" and "a friend". A named option gets a tick.

Then respect its limits. People misremember. They report the last step, not the first, so the guest who saw you in ChatGPT and then Googled you will say Google. The field measures memory and the channel group measures clicks. Two floors for two different things, and they do not add up to a total. Report them side by side and say so.

Prompt sampling

If the question is whether assistants name you at all, ask them, properly. LLM outputs are non-deterministic. The sampling guidance is 12 or more repetitions per prompt; five runs give a margin of error of roughly plus or minus 27 points. A single screenshot of ChatGPT recommending your hotel proves it happened once. How AI assistants choose which businesses to recommend covers what moves that result.

What you cannot know

A plain list, because this category is full of people who will not say it.

• The number of bookings that came from an AI answer.
• Which prompt produced the mention.
• Whether any particular Direct session started in an assistant.
• Whether the guest who arrived through Yelp or Maps was sent there by ChatGPT.

There is no AI equivalent of Search Console click data. If a report shows you AI-attributed revenue to two decimal places, it was built on an assumption. Ask which one.

What to do this week

1. Build the AI referral channel group and label it a floor wherever it appears.
2. Replace the free-text source question with a fixed list that names AI assistants, on the form and in the scripts.
3. Pull Direct sessions by landing page and note whether deep-page landings are trending up. Do not convert it to a number.
4. Register Bing Webmaster Tools and read the AI Performance report.
5. Pick ten prompts a real guest would type, run each twelve times a month, and record who gets named.
6. Stop reporting AI bookings as one figure. Report both floors and say they cannot be summed.

The guest at the desk who said "ChatGPT" is the best data point you will get this month. Write it down before they reach the lift.