Buying AI is easy. Killing manual work is not. Hotels keep proving it.
According to PhocusWire, new H2c and NYU/RateGain/HEDNA data from early October 2026 shows a sector that has bought the tools and kept the grind. More than half of hotels are using or procuring generative AI. Fewer than one in ten report cuts of more than 30% in manual work. That gap is the story. Not the model. Not the vendor deck. The operating system of the hotel.
“We’ve adopted AI” is not a productivity claim
H2c found that 69% of hotels still have staff manually entering guest preference data. Only 19% use AI-driven extractions from guest interactions. The State of Distribution 2026 report adds that more than 80% of commercial teams still burn one to two days a week producing and analysing reports by hand.
Read that again. Four days of a commercial month can vanish into reporting theatre while the brand tells the board it is “AI-led”.
Amanda Moore, VP of digital experience at Preferred Travel Group, put the problem cleanly: layering AI on top of disconnected data does not remove the complexity underneath. Ksenia Tarasova at Penta Hotels went further — AI is only as useful as the data and processes behind it. Duplicate guest profiles, messy consent, preferences scattered across systems: put a model on that and you amplify the mess.
So the debate is not “is hospitality late to AI?” It is already buying. The debate is whether hotels are buying relief or buying another silo.
Midsize chains look awkward for a reason
Both Moore and Tarasova point at midsize groups — under 100 properties, a few countries — as the commercial “sweet spot”. They adopt preset automation more than large chains or independents. About 14% of them report labour cuts above 30%. Their AI search share sits around 6% of reservations, the highest of the three size bands, against a hotel-wide AI search share of about 4%.
That is not magic spend. It is organisational shape. Big enough to hire specialists. Small enough to avoid the legacy maze that slows global brands. Tarasova’s line is the one CMOs should put on a slide: a mature system with good APIs and clean data beats a shiny new AI platform sitting in another silo.
If your group is mega-brand scale, the uncomfortable implication is obvious. Your AI budget may be funding theatre until someone owns data quality across the guest journey.
Distribution still prefers the OTA, not the chatbot
The same distribution report is blunt on where bookings land in 2026: OTAs 34%, hotel websites and apps 18%, offline 25%, GDS 8%, direct metasearch 5%, direct voice 6%, AI search 4%. Core direct-booking kit is nearly universal. Demand capture is not.
Moore’s point for independents is one we have argued before when OTAs and Tripadvisor picked up share as Google Hotels wobbled: you cannot outspend the OTA at discovery. You win by giving travellers a reason to book direct — storytelling, member value, metasearch visibility, frictionless book, and CRM that recognises the guest after they arrive in your ecosystem.
AI search at 4% is not nothing. It is also not a strategy if 55% of hotels report little or no change to distribution because of it. Hotels should treat AI answers the way they learned to treat organic search and metasearch — monitor presence, fix structured facts, stop hoping a chat widget rewrites the funnel. That sits alongside how AI is already reshaping holiday planning and the budget reality we covered for how a travel CMO should prioritise SEO in H1 2026: fragmented data kills every fancy retrieval layer you buy on top.
Fix the work, then buy the model
Moore’s forward look is the right brief for Q4: stop asking “where can we use AI?” and start asking “which workflows should run differently because AI exists?” Tarasova is even more direct — clean and connect the existing stack before you shop for another platform.
That is the take worth arguing in a hotel marketing meeting:
- Adoption metrics are vanity if manual reporting still eats two days a week.
- Fragmentation is the bottleneck, not GPT vs Claude.
- Midsize agility is a clue, not a consolation prize for big brands.
- AI search at 4% will not save direct share while OTA discovery still wins.
If your 2026 AI programme started with a tool shortlist and no data-ownership map, you did it backwards. Process and org first. Model second. Otherwise you will still be pasting guest preferences into a spreadsheet in 2027 — with a nicer chatbot watching you do it.
Follow the newsletter: Dan Taylor on LinkedIn