Travelers are increasingly asking an assistant where to stay before they ever open a booking site. The shortlist gets built in that conversation, and a property that is not named in it does not get compared, priced, or considered. For independent hotels and inns, this is the rare shift that favors the small operator.
Be precise about this, because the hype and the reality point in different directions and the gap between them is where money gets wasted.
On the demand side the shift is real and fast. Phocuswright's research on generative AI in travel found around half of US travelers now using these tools for trip research, up from roughly a fifth two years earlier, while conventional search usage for travel research has fallen over the same period. Adobe Analytics measured AI-referred traffic to US travel and hospitality sites rising by orders of magnitude year over year. Similarweb has found that visitors arriving from ChatGPT convert at markedly higher rates than organic search traffic, which makes sense — they arrive pre-qualified, having already been told you fit.
On the transaction side, the picture is much quieter. Booking Holdings reported that of the 325 million room nights it processed in the second quarter of 2026, under one percent arrived through an AI referral. The infrastructure a chatbot would need to actually complete a booking — live inventory, authentication, payment settlement, cancellation handling — still belongs to the platforms.
So the correct read is narrow and useful. AI is where the shortlist gets built. The booking still happens somewhere else. Your job is to be on the shortlist, not to chase AI booking volume that does not yet exist.
In conventional search, a fifteen-room inn competes against chains and OTAs with twenty years of domain authority and budgets it cannot match. That fight is not winnable on those terms.
A recommendation engine works differently. It is trying to match a specific request to a specific property, and specificity is exactly what independents have. A traveler asking for somewhere adults-only with a wood-burning fireplace within walking distance of dinner, or a place that will hold a room for a late arrival off a delayed flight, is describing you and not describing a chain property. The constraint is only that you have to have written it down.
Language models recommend what they can state confidently and attribute. That means concrete attributes in plain text: room count, bed configurations, check-in and check-out times, whether there is an elevator, parking and what it costs, whether breakfast is included and what it consists of, pet and children policies, minimum stays, the cancellation window, what is genuinely within walking distance and how far.
Then answer the questions travelers actually ask, in the form they ask them. Is it quiet? Can I get a late check-out? Is it walkable to dinner without a car? Will the wifi hold a video call? Is there anywhere to sit outside in the evening? A page that answers those is far more citable than a page of atmospheric copy about crisp linens.
And make it agree everywhere. Your site, Google Business Profile, the OTA listings, the tourism board, the guides. Contradictory hours, room counts or amenities lower a model's confidence in every claim about you.
Here is the pattern I see most often at independent properties. The richest, most complete, most factual description of the hotel exists on Booking.com or Expedia, because those platforms force you to fill in structured fields. The property's own website carries three paragraphs of mood copy and a photo gallery.
The consequence is that the OTA becomes the authoritative source about you, and recommendations route through it rather than to you. You pay commission on business that a better-written page of your own could have captured directly. Take the structured detail you already supplied to the OTAs and put it, in your own words and with more of it, on your own site.
Start with a baseline. Ask ChatGPT, Claude, Gemini and Perplexity where to stay in your town for the kind of trip your guests actually take. Note whether you are named, what is said about you, and whether it is accurate. That takes ten minutes and tells you more than any report you could buy.
Then write the facts down, fix the contradictions across your listings, add structured data for your property, and give it a few weeks. AI engines re-ground faster than search rankings settle — this is one of the few areas where the feedback loop is measured in weeks rather than quarters.
Not to start. The baseline check costs nothing but ten minutes of asking the assistants directly, and most of the fixes are writing and consistency work. Be skeptical of anyone guaranteeing placement in AI answers — no such guarantee exists, because there is no ranking to buy.
Booking Holdings reported AI-referred bookings still under one percent of room nights in mid-2026, so the transaction layer is not there yet. The discovery layer already is, which is why the shortlist is the thing worth competing for right now.
No. The point is not to abandon them but to stop letting them be the only complete description of your property on the internet. Match and exceed that detail on your own site so recommendations have somewhere direct to point.
Trace the source. It is usually a stale directory listing, an outdated OTA field, or an old press mention. Correct it where it originates and the models generally update within a few weeks of their next crawl.
Want to know what AI currently says about your property? I run a free 30-minute diagnostic.