The hotel title is not the outbound link
Most hoteliers I talk to assume the bolded hotel name in a ChatGPT answer is the link to their website. It is not. Clicking the hotel title opens an in-ChatGPT side panel with a description, an address, and a row of buttons (Reserve / Directions / Website). To actually reach the hotel website the user must click again — either on Website inside the panel, or on Website inside the hover card on the map. So the “direct” link is two clicks deep, behind a UI that ChatGPT controls.
Meanwhile, the small grey citation chips next to bullet points (oyster.com, tripadvisor.com, etc.) do leave ChatGPT in one click. So the question “does ChatGPT send traffic to my hotel?” has the wrong subject. The single-click outbound traffic in a hotel answer goes mostly to aggregators and editorial sources. The hotel itself sits two clicks deep behind a modal.
The shelf, surface by surface
Map markers (and hover)
In-app, then outboundThe embedded map at the top of the answer pins each recommended hotel. Hover a pin and a small card appears: thumbnail, name, rating, price, and three buttons — Reserve, Directions, Website. Reserve and Directions stay inside ChatGPT (or hand off to Google Maps). Website is the only true outbound jump from this surface, and only on hover.
Hotel title → modal panel
In-appClick the bolded hotel name in the bullet list and a side panel slides in: photography, address, phone, a longer description, and the same Reserve / Directions / Website tabs. This panel is the densest in-ChatGPT placement for a hotel and the one most users see — but it doesn’t leave ChatGPT until the user clicks Website inside it. The title link is an in-app expander, not an outbound link.
Citation chips
Outbound, one clickInline grey chips (e.g. oyster.com) attached to specific claims. Hover reveals the page title; click leaves ChatGPT directly to the source. These are the genuinely outbound clicks inside the answer body — and they almost never go to the hotel itself. They go to whichever site ChatGPT used to ground the claim: Oyster, TripAdvisor, Condé Nast, regional press, niche editorial.
links_attached block
Outbound, top-of-answerSome answers carry a small “Sources” or pinned-link strip near the top (and a longer list at the bottom). These are the model’s top-ranked references surfaced as clickable units. Highest visibility per pixel of any outbound surface in the answer — and again, almost always non-hotel sources.
What this means for “ChatGPT traffic”
If you instrument referrers with that surface map in mind, the picture splits into three buckets that get conflated in dashboards:
The attribution trap
A hotelier checking GA4 for “ChatGPT referrals” sees only bucket 1 — the smallest. The interest signal is much larger; it just doesn’t leave the chat. Conversely, an aggregator that shows up in citations sees a stream of single-click referrals from chatgpt.com and reasonably calls it “AI traffic.” Both numbers are real. They are not measuring the same thing.
What the full piece will cover
- Annotated screenshots: every clickable surface in a hotel answer, labelled with what it does on click vs. on hover.
- Captured network traces: which surface fires which request — in-app modal vs. outbound navigation.
- Citation distribution: across N hotel queries, what share of citation clicks would go to OTAs, editorial sites, niche aggregators, and the hotel itself if click-through were uniform.
- Why this matters for ChatGPT’s hotel pipeline economics: the model has every incentive to keep the user inside the panel, because that is what differentiates the answer from a SERP.
- Practical: where to optimise if you want hotel-domain referrals (the Website button target) vs. citation visibility (the chip text and source ranking).
Summarize with AI
