How does ChatGPT’s AI Agent, Operator, feel about your booking engine?

• Synxis
• TravelClick
• Selfbook
• Siteminder
• Simple Booking
• SHR
• D Edge
• Travelanium
• Guestline
• Mews
• Webhotelier 

For simple booking queries

We used the same prompt for each booking engine: 

Please book me a room at [hotel name] from July 15th to 17th. 

Our initial findings? When given this direct, specific prompt, even without a requirement to book direct, Operator started where most humans would: a search engine. Unlike most humans though, it consistently started its journey on Microsoft Bing, despite Bing's tiny market share. (Perhaps this has something to do with the amount Microsoft has invested in OpenAI, the creator of ChatGPT...?) Based on all of our testing, if you want Operator to use any other search engine you need to explicitly prompt it to do so. 

The rest of the process was good news for direct bookings. After an initial Bing search for a hotel, Operator consistently navigated to the hotel's direct website. Once there the general flow we saw in each case (with no additional upfront instructions from the user) was as follows: 

• Searching for availability at [hotel name]
• Accessing official site, checking availability
• Accepting cookies, checking booking availability
• Accepting cookie policy, clicking book now
• Closing pop-up to continue booking
• Clicking 'BOOK NOW' for availability
• Entering dates, checking availability now
• Adjusting dates, setting guest count
• Exploring available room options listed
• Presenting room choices to user
• Selecting [room choice]
• Confirming booking for [room choice]
• Reviewing booking details, proceeding forward
• Confirming booking terms with user

Here's a video (we'd recommend watching at 2.5x speed - Operator isn't the quickest!) showing how this worked out with a hotel using Mews: 

With this flow - or something like it - Operator was able to navigate to the booking page for every booking engine we tested, which meant it completed most of the act of booking. Certainly a positive sign for the future of direct bookings in an AI-driven world. (Note that when we first tried this just a few weeks ago Operator was unable to cope with most booking engine date pickers, so things are moving fast.)

The first place we typically did see the need for user input was at the room selection step: “Presenting room choices to user”, and Operator always handed the process back to the user after the final step above: “Confirming booking terms with user”

Operator can’t hold payment details for users, so guests will still need to provide credit card information manually (as is almost always required for this final step.) 

Key successes for Operator on hotel websites: 

• The agent always navigated to the direct hotel website – even if its listing was lower than OTAs on the Search Engine Results page 
• The agent was able to make its way through verification pages, and through cookie banners.
• Operator was able to successfully navigate through to the booking page on every booking engine we tested.
• When asked to add items to a search, or change something within a search, the agent was able to do so  – for example, changing from a non-refundable rate to a refundable rate, or navigating to a room with a view.  

While generally successful on these types of queries, there were some common struggles we noted throughout the automated booking process. 

Key issues using Operator on hotel websites:

• The agent currently can’t actually book anything. 
  • While Operator is able to navigate through the website and make it to the booking page, it can’t store sensitive information like credit card details or your address. Having to input these is a critical stopping point in the agentic booking journey.  

• The agent had some difficulty selecting dates 
  • Operator was quite slow at selecting dates - something we saw pretty consistently across booking engines and across the OTA sites used. Here’s one example of the number of attempts used to select a date on a hotel using the Synxis booking engine. As you can see, Operator tried (and failed) to select the correct time frame at least 20 times, resulting in about a 3 minute delay in the booking process (again, we'd recommend speeding it up!).
    
    

        This wasn’t limited to one specific booking engine, and seemed to occur basically at random,              even on OTAs in other tests which we ran. It could easily select the date once, and struggle the            next time. This could very well be something that becomes more consistent, and quicker, as                Operator completes more of these kinds of tasks and learns to navigate sites more efficiently. 

• The process is SLOW
  
  • Issues selecting date ranges was only one of the factors slowing down Operator. In general, the process just felt quite clunky - with the average booking flow to get through to the booking confirmation page taking four to five minutes, and sometimes much longer if other issues popped up. And that’s without entering any payment details!
• Any technical difficulties led straight to OTAs 
  
  • The only times we saw the agent leave the direct website and move to an OTA were the results of onsite issues. For example, an issue loading a Simple Booking page led to the agent moving to Booking.com. Again, this doesn’t seem to be a limitation with any one booking engine. 
• Hotel-specific rules and regulations were a struggle for Operator 
  
  • Operator struggled to understand requirements that were unique to each hotel – like a 3-day minimum stay, or certain days assigned to check-in and check-out. In the case of one TravelClick hotel we tested, Operator ended up leaving the direct website to try to book on Booking.com, citing lack of availability – rather than the actual reason given by the hotel. The same issue occurred on Booking.com, so it’s not a booking-engine specific problem. But it did create some friction in the booking process and could drive a guest to an OTA website for further research. 
• Operator wasn’t consistent in the way it approached room selection 
  
  • There wasn’t really a consistent approach to room selection. In some cases, Operator chose the cheapest room rate for us. In others, it gave a list of all available rooms for that date. Most often, though, it chose two or three options from the page and asked for user input. Not the best user experience, given the risk of bookers inadvertently making decisions based on an artificially limited selection. 
• Hotel-specific rules and regulations were a struggle for Operator 
  
  • Operator struggled to understand requirements that were unique to each hotel – like a 3-day minimum stay, or certain days assigned to check-in and check-out. In the case of one TravelClick hotel we tested, Operator ended up leaving the direct website to try to book on Booking.com, citing lack of availability – rather than the actual reason given by the hotel. The same issue occurred on Booking.com, so it’s not a booking-engine specific problem. But it did create some friction in the booking process and could drive a guest to an OTA website for further research. 
• Operator wasn’t consistent in the way it approached room selection 
  
  • There wasn’t really a consistent approach to room selection. In some cases, Operator chose the cheapest room rate for us. In others, it gave a list of all available rooms for that date. Most often, though, it chose two or three options from the page and asked for user input. Not the best user experience, given the risk of bookers inadvertently making decisions based on an artificially limited selection. 
• Operator wasn’t always accessible
  
  • As a test product, Operator seemed to occasionally get overloaded. A couple times we tried to access it to make a search, we were hit with the “white screen of death”. Not a direct website-specific issue - but definitely a blocker to guests trying to make a booking. 

While it was interesting to discover how Operator fared, booking a room is obviously a very specific use case. We don’t see it becoming the norm for quite some time given the need to already know exactly where you want to stay, the relatively glacial pace of using Operator, and the fact that users will still need to input payment details in order to actually book anything. So we tested another couple of search types to see whether Operator remained a direct booking champion. 

Searches slightly higher up the funnel

Perhaps most interesting for hotels focused on direct bookings was a clear, but unsurprising pattern: when given ambiguous or higher-funnel queries that weren't asking for a specific hotel, Operator either used Bing Travel to research, or went directly to an OTA. There didn’t seem to be a clear pattern in how they determined this. Let’s take 2 different, but equivalent examples. 

We searched: 

“Can you book me a hotel room in san diego with a pool from july 15th to 17th?”.

Operator went to Bing, searched “San Diego hotels with pool booking” and proceeded to use the search filters there to find a hotel. It still did eventually end up going to Booking.com, but this seemed to be because a Booking.com listing was the first one on Bing Travel. 

This just shows how important it continues to be for hotels to optimize for existing channels like metasearch. Being active on Bing metasearch would be one of the most immediate ways to ensure that your hotel is visible to Operator in this type of situation - though similar tools may use other search engines. 

But there doesn’t seem to be a ton of rhyme or reason to Operator’s actions. Because we tried another, very similar query, and Operator took an entirely different route. 

We searched:

Can you book me a hotel with a pool in LA from July 15th to 17th?

In this case, Operator searched “hotel booking site” in Bing, and chose the first site which came up in a list of organic search results. This was – you guessed it – Booking.com. It was then able to continue their search from that point forward on the OTA site.

While Operator can take various routes, it all seems to come down to one conclusion: the agent tries to navigate the web like a human might. The hotel choice and advanced filters available on OTAs (and on channels like Bing Travel or Google Hotels) suit those looking for ideas from a range of options.  

Planning rather than booking: Operator takes a different approach 

Our tests also revealed that AI agents approach higher-funnel, inspiration-stage queries much like humans do - by looking for content-rich sources like listicles and editorial recommendations. This is definitely a big AI opportunity – while guests may not be ready to hand over their full booking journey to an agent, the idea of using one to narrow down their options may feel a bit more accessible. 

We asked: 

“I'm going to San Diego from the 15th to 17th of July. It will just be me and my kids, so looking for something really family-friendly. Can you recommend 5 potential hotels and let me know why I should book each one?"

The first listing on the organic search results page was for a content-forward, optimized-for-SEO local blog called “La Jolla Mom”. 

The output for the query (after 9 minutes parsing the article) was this list, which picked five hotels from those featured. 

After we replied (picking Hotel del Coronado), Operator searched for Hotel del Coronado on the Bing search engine, and navigated through the direct booking flow. 

This behavior highlights the continued importance of traditional SEO and content marketing strategies. Hotels appearing in these high-ranking content pieces - especially with direct booking links - may still be shortlisted by AI tools, even if the agent struggles to complete the actual booking.

Another thing worth considering is whether Operator is actually the best AI tool for these kinds of searches. With Operator, this query essentially results in a summary of one web page that the agent ‘read’. A generative search engine (like Perplexity or ChatGPT) would typically scan more sources and potentially provide a wider range of information, despite its own limitations. Considering the fact that research is by its nature not completely hands-off – you’re going to want to review the results, at the very least, and likely follow up with more queries – it’s questionable how much value tools like Operator offer today.

Is your website ready for Agentic AI? 

Based on our findings, AI agents like Operator are already pretty good at navigating a range of different website designs. But there are a few key recommendations for hotels looking to make their websites more AI-friendly:

1. Focus on technical health: Many of the problems we encountered weren’t due to specific booking engines, or even formats – but issues like load speed and unpredictable onsite issues. By ensuring that you’re keeping good technical hygiene, you can avoid issues when your site interacts with AI agents. 
   
   
2. Clearly communicate booking restrictions: Make stay restrictions, minimum nights, and check-in/out limitations abundantly clear in machine-readable formats
   
   
3. Keep investing in traditional channels: In our test, we saw Paid Search, Metasearch, and SEO playing a key role in the journey that Operator took. Now is not the time to stop focusing on these elements of your marketing mix. 
   
   
4. Continue building out your SEO strategy: Invest in traditional SEO (potentially leveraging AI to do so) to appear in the content sources that AI agents consult for higher-funnel queries. 
   
   

We’re keeping an eye on these tools – and we’ll be reporting back

While AI agents represent an intriguing development in travel booking, our research suggests they're not yet poised to dramatically reshape the landscape. The challenges they face - from date selection difficulties to payment processing limitations - mean that human involvement remains essential to the booking process. And in this case, these results were based on a tool that is currently too expensive for most people interested in using more AI in their day to day. 

We'll continue to monitor this space closely and provide updates as these technologies mature, testing different tools and sharing insights that can help hotels understand where AI is headed. Meanwhile, we’ll be helping hotels to continue to succeed on their existing channels like paid search and metasearch – which, based on what we’ve seen so far, may just help them to achieve more success in this new AI world as well. 

Keep in touch, and subscribe to the Triptease blog for more updates on AI agents and how they’re developing. Next up, China’s AI Agent Manus. 

Subscribe now for these insights and more, straight to your inbox