ChatGPT now sends 92% of AI referral traffic — but a quarter of it lands on the wrong page
New cross-industry data shows AI referrals consolidating on one platform, and dropping high-intent visitors on internal search instead of the page they wanted
by Markus Busch
The number. AI referral traffic has consolidated onto one platform. A new study from SEO firm Previsible, analyzing 6.77 million AI-driven sessions across 166 sites from November 2024 to May 2026, finds ChatGPT now sends 92.4% of trackable large-language-model referral traffic, up from about 84% in December. Monthly AI sessions hit 644,478 in May, nearly ten times the November 2024 figure. Claude quietly overtook Perplexity in March; both Perplexity and Copilot have collapsed from their peaks. And the traffic is volatile: a single model change halved ChatGPT referrals in one month last fall. One vendor's product tweak can cut your AI traffic overnight.
The caveat. Two, actually. The study counts standalone chatbot referrals only. It excludes Google's AI Overviews, which Previsible itself says remains the center of AI discovery and likely drives more traffic than every standalone platform combined, so "92% of AI traffic" means 92% of the smaller slice. It is also a vendor study, written by Previsible's co-founder, spanning SaaS, ecommerce, health, and legal sites with no travel or hospitality vertical. Read the numbers as directional, not as your market.
The finding that matters. Skip the horse race. The actionable number is where AI drops its visitors: roughly a quarter of AI referrals, and 28.8% of ChatGPT's, land on a site's internal search results page. The model trusts the domain but can't pick the right page, so it hands the visitor the site's search box and walks away. Previsible calls this structural to how retrieval-augmented AI works, not a passing quirk.
What it means for your hotel. It splits by who you are. Run a group, and the property finder is exactly the search box the study describes: AI sends a decided traveler to it and lets your site choose among your hotels, so its design decides the booking. Run a single property with no search box, and the problem moves to the booking engine, where AI lands the guest on a page and your date-and-rate widget and structured room data determine whether they reach the room they wanted or give up. Either way, "contact us for pricing" gives an AI nothing to quote, compare, or recommend. The visibility work that used to end at the booking engine now begins one step earlier, on the page the machine picks for you.
Enjoying this analysis? Hospitality.today delivers daily insights on hotel distribution, AI trends, and travel commerce — straight to your inbox. Subscribe for free at Hospitality.today →