AI hotel bookings still need the human touch

MORE and more hotel brands and OTAs are looking to establish direct integrations with popular AI search engines. However, the realisation of end-to-end agentic commerce for hotel bookings is being delayed by a trust deficit from users and limited AI-optimised product content. Hotels, OTAs and e-commerce platforms have been working towards seeing agentic technology enact all parts of the travel booking process, from search to autonomous payments. For example, in 2023 Google’s Gemini integrated its Google Travel suite, including Google Flights and Google Hotels. In the same year, Expedia and Booking.com also initiated limited direct integration with ChatGPT. Unmatched ability Right now, AI search engines can provide live pricing and availability data and accommodation product breakdowns. Lasse Vinther, MD of Automation Architects, told Travel News that AI’s improving search engine capabilities were a valuable tool for travel agents. “AI agents’ ability to research complex purchases is unmatched, especially with the inroads of direct integration from certain e-commerce platforms,” said Vinther. “Agentic tools can automate the search-compare-enter- checkout drudgery, freeing advisers for high-value curated trips.” So, while AI searches were already driving up suppliers’ site visits in the planning phase, it currently directed users to the supplier website for them to make the final decision and booking transaction, he explained, noting that this was unlikely to change any time soon. “While AI will initially be seen as a product information source, actual conversions will be lacking due to trust and compliance issues when completing the actual transaction,” noted Vinther. The lack of trust in agentic search engines and agentic commerce stems from consumer concerns about inaccurate and biased search results, as well as those about having an AI entity to manage the payment process. Real-time data Vinther said many suppliers, OTAs and e-commerce sites were not being set up for agentic searches – their products and services were not being placed in data structures that were readable to AI. “In the hotel and accommodation industry particularly, hotels’ property management systems, central reservation systems and customer relationship management data is displayed in inconsistent formats, making it difficult for large language models (LLMs) to aggregate the data for searches.” He explained that this could result in the LLMs producing very limited search results for the user’s prompt, potentially limiting their options. Additionally, LLMs rely heavily on real-time data to ensure that they can provide accurate information about a hotel’s occupancy and room availability, but this information is not always readable to AI without in-depth integrations. “AI booking assistants need up-to-the-second accuracy on what rooms are available. If an AI relies on older, saved data, it might tell a customer a room is available when it is actually sold out, which is a mistake travel agents cannot tolerate,” he said. Biased Another big concern among users is that LLMs will provide biased search results based on paid partnerships and more extensive integrations with particular hotel brands. “Ranking bias and transparency are unresolved, but it is being increasingly regulated, particularly in the EU. AI answers can narrow choice to a handful of options, raising concerns about paid placement and hidden kickbacks. Research shows that the same OTAs that already dominate SEO will dominate LLM search results. Direct integration deals are mostly with those dominant OTAs which can offer commissions between 15% and 30% (to the LLMs directing traffic),” said Vinther. “Without transparency rules, AI answers could embed paid placement or preferred-partner bias.” However, Vinther emphasised that if the travel agent user had experience with AI they could curate prompts that instructed the AI according to their preferred partners and booking rules. “Agents can direct which brands they want to search and, if the brand is not blocking AI, then these rules created in their prompts will be adhered to. For B2B sales, this is where custom AI agents are particularly useful,” said Vinther, noting that more consortiums were developing this kind of tool. Autonomous payments LLMS have not successfully implemented systems that enable users to finalise travel product payments autonomously, partly due to user distrust and because of complex supply chains in the industry. Visa’s Stay Secure study, conducted by Wakefield Research and published in June this year, found that while more consumers were embracing AI to assist them with comparing products and prices, only 23% of South African consumers surveyed would trust an agentic AI to complete a financial transaction. Vinther said most LLMs were only now finalising some of the very first standards for AI to manage transactions, with Google in the lead. In late 2025, Google announced a massive project to let AI handle bookings, which saw major hotel groups like Marriott, IHG, Wyndham, Choice Hotels and certain OTAs join. And, as of mid- 2026, Google has started testing fully autonomous bookings in small pilots. Other LLMs have had to pull back on scaled direct AI booking programmes. In March this year, OpenAI admitted to Skift that travel booking was simply too complex for an AI to handle. This instance set a precedent in the LLM industry, seeing providers set their focus on enabling improved search-and-compare results for travel products first. “Most of the LLMs have decided to focus on supporting travel planning and inspiration. Almost all the LLMs still hand the actual booking back to the supplier’s or the OTA’s own checkout systems,” said Vinther.