91% of hotel chains use AI. Only 28% have a strategy for it

Almost every hotel chain now uses AI somewhere. Very few have decided what for. That is the short version of h2c's AI Opportunity Study 2026, published on 1 October.

The research firm collected 122 responses from 113 hotel chains, split between 101 online surveys and 21 executive interviews, and analysed 230 AI use cases already in production. The chains in the sample run more than 8,200 properties and roughly 1.3 million rooms across Europe, the Middle East and Africa, Asia Pacific and the Americas, averaging 73 properties each.

The headline gap

91 percent of chains use AI today and another 8 percent plan to adopt within 12 to 24 months. Only 28 percent have a company-wide AI strategy led by senior leadership.

The usage split explains a lot. 60 percent of daily AI use runs through external tools such as ChatGPT. Vendor-embedded AI accounts for 27 percent and internally built tools 12 percent. So most of the AI in hotels is staff using consumer tools on their own initiative, not capability built into the systems that hold the data.

That is shifting. 31 percent now rely primarily on vendor solutions, up from 21 percent in 2025, and the share still working out their approach fell from 29 percent to 12 percent.

What it is actually delivering

69 percent report better operational efficiency and automation. 59 percent say it lets staff spend time on higher-value work. 32 percent cite a better guest experience.

13 percent report measurable return on investment. Chains rated AI's overall contribution to their business 5.6 out of 10, and their own internal AI knowledge 3.4 out of 10.

The barriers follow from that. Lack of expertise, training or skills leads at 56 percent. Integration with existing systems is second at 38 percent, then no clear strategy or roadmap at 32 percent, data security and privacy at 29 percent, data governance and access at 28 percent, and cost at 22 percent.

Agents, and who they are for

70 percent expect AI agents to be the main innovation area over the next two years, and the same share expect bookings to start arriving through AI platforms.

Where agents are already running, 70 percent are staff-facing. 55 percent of chains plan guest-facing agents, and the hesitation there is specific: unclear liability when the AI gets something wrong (56 percent), data privacy and security (54 percent), whether guests will share personal data at all (53 percent), and accuracy, reliability and bias (52 percent).

On AI-driven booking, 70 percent expect platforms like ChatGPT and Gemini to lift direct bookings, 47 percent moderately and 23 percent significantly. Only 22 percent systematically monitor how their brand appears in AI search results. A third have taken no steps at all to make their rates visible or bookable through those channels.

Worth knowing about the source

The study is sponsored by hotel technology vendors, including Apaleo, Shiji, Cendyn, dailypoint, D-EDGE, Aven Hospitality, Busy Rooms and Sirma Travel and Hospitality. The full report goes to sponsoring partners only. A public extract is on h2c's site and participating chains get an executive summary.

That is relevant context for the study's vendor-facing conclusion, though the finding itself is reasonable on its own terms.

"Hotels do not simply need more AI features. They need connected systems, accessible and well-governed data, and interoperable platforms," said Christin Haensel, director at h2c.

"The industry has moved beyond asking whether to use AI," said h2c founder Michaela Papenhoff. "The more pressing question is how hotel chains turn widespread experimentation into connected, governed, and scalable capabilities."

What this means if you run properties

Two numbers are worth sitting with. Integration being the second biggest barrier at 38 percent says the constraint is rarely the AI, it is whether your PMS will give it the data. Before buying another AI feature, it is worth asking your vendor what it can read and write, and through which API.

And 60 percent of daily use running through ChatGPT means staff are already pasting operational detail into tools nobody signed off. A policy on what may and may not go into a public model is cheaper than most AI projects, and more urgent than almost all of them.