This insight summarises Is AI search a distraction? Not for hotels, judging by our data, by Lynn Patchett of Kollective, published in October 2026.
The debate. Three studies from October 2026 (The SEO Works, Kevin Indig and Marie Haynes) suggested AI assistants send little traffic and are hard to tie to business results. None focused on hospitality. Kollective compared their findings with GA4 data from 37 hospitality and travel websites, mostly boutique hotels: about 3.8 million sessions, around 47,000 of them from AI assistants.
What the hotel data shows:
- Organic search sent 24 visits for every AI visit, against 39 in The SEO Works benchmark.
- AI visits as a share of hotel organic traffic rose from 1.4% in September 2025 to 9.2% in August 2026.
- From January to September 2026, AI visits were about six times the same months of 2025.
- AI assistants sent about four and a half times as many visits as DuckDuckGo, and more than Bing between June and August 2026.
Booking intent. AI visitors clicked through to the booking engine at 44%, against 23% for organic search and 16.7% for all visitors. Yet AI visits converted to recorded bookings at about two thirds of the organic rate. Patchett points to attribution gaps (separate booking-engine domains, consent settings, device switching) and to guests who discover a hotel through AI, then return through brand search or direct visits.
Listing links. At hotels that tag their listing links, nearly one ChatGPT visit in four carried a Google Business Profile or Yelp tag, so ChatGPT often links through those listings. Google Analytics' new AI Assistant channel caught only 80% of unpaid AI visits to the hotels.
Practical takeaways:
- Tag Google Business Profile and Yelp links so listing visits from AI show up.
- Do not rely on GA4's default AI channel alone; check source and medium directly.
- Treat AI assistants as another discovery surface built on the same SEO foundations, not a separate GEO budget.
- Expect recorded bookings to understate AI's role when guests return later through other channels.
Patchett notes the sample is small, not random, and largely made up of Kollective clients, and invites other operators to share their data.