AI-referred traffic to hotel and short-term rental websites quadrupled between July 2025 and July 2026, yet it still accounts for less than one visit in a hundred. Those two figures, from Similarweb data published by Bain & Company on September 30, frame the operator's question this winter: the AI guest is real, but she is rare.

Bain tracked three to five large sites per category and found every travel segment — OTAs and cruise lines included — below 1% of total traffic. A forty-room independent may look quite different. Booking Holdings, the largest seller of hotel rooms, counts the same way. On its August earnings call the company said traffic from AI assistants, paid and unpaid combined, stayed significantly below 1% of room nights, and that the share had not moved in quarters.

How big is AI-assisted travel planning really?

The planning behavior is far larger than the traffic suggests. Phocuswright's March report found 56% of US leisure travelers used AI for at least one trip in the previous twelve months, up from 43% in the second half of 2025 and 33% in the first half. Phocuswright researchers called it the fastest change in traveler behavior they had measured in more than a decade.

Retention is high. Bain's consumer research found that 49% of US travelers who have used AI for travel now rely on it most or all of the time; across the UK, France and Germany the figure is 41%. Bain does not disclose when the survey ran or its sample size — worth knowing before quoting it in a budget meeting.

Most of the planning happens inside the conversation, where standard analytics cannot see it. What reaches a hotel website is the last step.

Does the AI guest actually book?

Adobe's data on US travel websites says yes, increasingly. In July 2026, traffic arriving from AI sources was up 119% year over year. The sharper number is conversion. In July 2025, visitors from AI converted 47% less often than everyone else. By May 2026 the gap was 28%. By July it was 1%.

A year ago the chatbot visitor looked far more than she bought. Across travel sites she now arrives, compares and books at nearly the rate of a guest coming from Google or an email. Adobe sells tools for tracking AI traffic, so it has an interest in the trend — but these are counted visits and purchases, not survey intentions.

The booking still runs through the property's own site. Phocuswright found that half of travelers who use AI answers in search engines still click through to the source website afterward. Earlier reporting found that after an AI recommendation, well over half go on to the property's site, and travelers who trust AI check almost as often as those who don't. She arrives having read about the hotel elsewhere — perhaps fresh from an OTA listing showing a total for the whole stay — and gives the site one look to confirm it.

What does the direct-booking share show?

Bain's report includes a chart built on Euromonitor data: travel spending booked in advance grew from €1.1 trillion in 2011 to €2.4 trillion by 2024 and 2025, while the share booked direct held at roughly 45% in every year shown. The chart covers all travel, flights and cruises included, so it cannot be read as a hotel figure. Bain's reading: the flat line hides a long fight. Travelers moved to OTAs in large numbers, and hotel brands spent heavily on guest-relationship software just to hold share steady.

What should operators do this winter?

Do not rebuild around a sub-1% channel, but do measure it. The practical move: filter twelve months of website analytics for referrals from chatgpt.com, perplexity.ai, gemini.google.com and copilot.microsoft.com, then ask how many visits came that way, how they converted against Google organic, and which page they landed on. Run the same report quarterly — the one AI number no vendor is selling.