Memory Is the New SEO: How Agentic Commerce Will Reshape Hotel Distribution

Agentic Commerce Notes for Hotels

While the rest of the world wrestles with the implications of OpenClaw, (spoiler alert: it’s not WhiteClaw-for-seafood) I thought I’d write down a few things about how I think the agentic commerce world is shaping up around hotel shopping and distribution. Over the last few months I’ve been getting lots of questions about whether or not Agentic channels are real, mostly around these three areas:

• How will Agentic channels work for hotels in particular? It’s not like buying a pair of shoes or an electric toothbrush.

• Will people really want to use them? Why? What’s wrong with the web?

• How can and should I participate? What are my options?

• Do I really have to get involved now or can I just wait a couple of years until things settle down?

Writing all this down got to be a very lengthy piece of work and I don’t want to torture anyone with reading it in one go, so I decided to break it in two: the consumer-agent side and supplier-agent side. In this post I’ll stick to the consumer-agent side and explain how it works, who the likely players are, what they will do, and how it’s different from web travel distribution. By the time we’ve looked at both sides of the diagram, we’ll be able to answer those four questions above.

Some Caveats Before We Begin

1. Even though I’m talking specifically about hotels here, something very similar is unfolding for flights, ground transportation, activities, and all the other components we think of as constituents in a well-designed itinerary.

2. I’m going to use ChatGPT as the example here because it has a unique combination of capabilities and also the largest chatbot user base among all the major models.

3. Not everything in this diagram is fully fleshed out and available today, though the majority of it is here in usable form and the rest is being developed at breakneck speed, mostly by big tech companies who directed emerging standards for the internet in the last go-round. Note: no company from travel is on this list, it falls to the tech titans like Google, OpenAI, Anthropic, etc. We are bit players on the stage at this point.

4. I really feel like a broken record on this but I have to say it again: Agentic shopping is not going to replace your web shopping channel. It will come in alongside the web the same way that the web didn’t completely replace offline travel. I fully expect offline, web, and agentic channels to persist for quite a long time. So whatever’s working for you on the web… keep on doing it! Over time there will be some share shift from web to Agentic but that’s a discussion for a different day.

Phase 1: The Consumer Agent in Travel—Establishing Traveler Intent

In the course of preparing for all this, I also put together a handy-dandy visual aid. I think it will be easiest to use this to refer to as we go forward.

The basic function of the consumer’s agent is to establish intent, to establish what it is that the customer or traveler is most likely to want in their trip.

Let’s go through each of the pieces in the process to get a feel for how each is going to work. For this article, we’ll focus on the left side, the consumer agent side, of the diagram.

Step 1: The consumer accesses an agent or LLM via a chatbot or other user interface.

The default is a chatbot (like ChatGPT) associated with a large company LLM, but there will also be custom user interfaces (e.g., Otto) that could be developed for specific types of travel. We could also see third-party hyper-personalized assistants like OpenClaw that serve as a highly effective front door for the agentic experience. And in the case of OpenClaw, it could then go up on OpenClaw’s social media sites and complain about what a lame traveler you are because you didn’t sign up for kite surfing and base jumping. The joys of AI!

Step 2: Memory and Context Building

Once the buyer agent has received your request, it may ask for further details to make sure it has a good feel for what you want, and then begins Step 2: searching through its memory system to augment that stated intent with all the relevant bits of information it’s gathered about you over all of your previous chats. This could include previous vacation searches where you told it about good and bad experiences you’ve had with destinations and suppliers, things you’ve said about members of your family and their needs during travel, or really just anything that you’ve discussed that it thinks might be relevant in planning the next trip.

In addition to its internal memory, it can now look at what I call external data for memory, which includes (with your specific permission) your emails, calendar entries, text messages, web history, etc. If you’ve had email exchanges about that terrible experience you had flying a particular carrier, that carrier is likely to be downgraded in future searches. If you had a surprisingly good experience in a particular hotel or destination, it takes note of that as well.

External memory is a combination of artifacts created on the side that may have implications for your travel, plus things that could be exclusively for travel, such as a travel profile with preferred suppliers and loyalty programs. I hope you’re getting the picture that the buyer’s agent really represents something like a smart executive assistant for travel…if your executive assistant worked 24 hours a day, only cared about getting the best for you, operated at a PhD level of intelligence, and had a vast, encyclopedic knowledge not only of your habits and desires but also of the travel world and the physical world in general.

Step 3: Matching Intent to the Outside World

Once the buyer agent has all of that personalized context loaded, it moves to Step 3 where it begins matching your intent to what might be available in the outside world. Whether you’ve chosen a specific destination or if you need help with choosing one, there are at least two collections of data to be searched. The first might be something like Google Data Commons for destination discovery. These are a series of databases made available through an agent-friendly MCP server. You can think of them as an encyclopedia of world facts that include geography, weather, and other kinds of data to help determine how well a certain destination might meet your needs for this trip. So if you’re planning this trip after your third margarita and casually ask about water skiing off the coast of the Outer Banks in November, this is where the chatbot might be able to flag that as a very bad idea.

Another obvious place to gather huge amounts of data to help determine the specifics of the trip and where to search is in what I refer to as Experience Commons. This includes the broader web, but specifically social media, reviews, third-party databases, etc. Depending on which agent you ask, it might be looking at YouTube, Instagram, Reddit, newspaper sites, and many other online repositories of relevant travel data. The buyer’s agent then combines the traveler’s intent from Steps 1 and 2 with what’s available and how it’s perceived from Step 3 and now faces the question of “where should I go shopping?” Each agent will have a pre-screened list of places to shop filtered by relevance, capabilities, reliability, and other such criteria. No agent wants to waste its time with a source that isn’t productive and robust. It can also use an MCP lookup service to find resources that have created agent-friendly MCP servers to make the job of gathering descriptive content and offers.

The Key Insight: Memory Replaces SEO

Here’s a key takeaway for the ChatGPT agent: By the time the agent has established intent and gathered a list of places to go shopping, it has not watched any ads, there is no sponsored content, it doesn’t care how many people have looked at any option today or how many have it in their shopping cart. In short, it is completely immune to all of the persuasion signals that are ever-present in the web channel. The agent only cares about getting the best available options for this specific traveler (with two exceptions as we’ll see below) on this specific trip, and now it’s ready to go and see what kind of offers it can surface.

And here’s another key takeaway: given what we’ve seen about the way the agent builds intent, it should be very apparent how incredibly important the chatbot’s memory function is in this process. Memory builds the context that the model will use to shop. In the web world, suppliers and aggregators try to capture attention to drive transactions. But in the agentic world an agent has no attention to capture… but it does have memory, which means we need to find ways to be positively represented in our shopper’s memory to drive agentic transactions. In agentic commerce, SEO gets replaced by positive memory access. More on that later.

Shopping Pathways: How Agents Find Offers

As we leave the consumer agent side of the picture, the agent has three basic options for shopping: a web search, an MCP tool call directly to an MCP server, or an agent-to-agent conversation with the seller’s agent.

These three pathways matter because they represent increasing levels of sophistication and value for both parties. Web search is the lowest-friction option but provides the least structured data—the agent is essentially scraping what it can find. MCP tool calls offer structured, reliable data exchange but require suppliers to build and maintain MCP servers. Agent-to-agent conversations represent the holy grail: real-time negotiation, dynamic personalization, and the ability to close the commerce loop entirely within the agentic ecosystem. For hoteliers, the pathway you support determines how much value you can extract from this new channel.

So where does the agent go? It depends on the agent. The table below highlights seven travel shopping agents including one that hasn’t shown up yet…but will soon (we hope).

Choosing an Agent for Travel: The Early Bird Will Get the Worm…and Keep It!

Here’s how I think about the entries in the table below.

• The size of the user base is obviously critical, and right now it looks like a two-horse race between OpenAI and Google. Note that these are user bases for chatbots, not search.

• The existence of ads indicates that there are commercial relationships with entities outside of the LLM provider. I think of this mostly in terms of the model provider’s business model: those with ads have a large enough consumer base to monetize and are more likely to be survivors in the long run.

• Commercial relationships influencing recommendations is a key metric in my mind. Over time these chatbots will become more and more a part of our lives and consumers will want to know that they can trust the information they get from them. If we know that a model’s recommendations can be influenced by ads or other commercial relationships, that trust erodes.

• I add a separate column for the use of persistent memory because I believe it will be a fundamental driver of the entire agentic commerce flywheel. Further, I agree with many in the AI world that consumers are likely to choose either a single chatbot or separate but dedicated chatbots for specific types of tasks (e.g., one for travel and other personal domains and another for software development and other business domains) and memory will make those decisions very sticky. Once you’ve trained your executive travel assistant on all of your travel needs as well as those of your frequent travel companions, why switch to another agent? This is an area where first-mover advantage will be critical for both the chatbots and for the pathways it shops, meaning that suppliers also have a lot to lose by not stepping up early.

• Finally, how the agent shops is important because it indicates the degree to which an agent will support the full flywheel of agentic commerce in hospitality. More on that when we cover the supplier side, but an agent that only shops the web will have no more ability to close the flywheel loop than an OTA does. That in turn makes it much less attractive to suppliers.

Travel Chatbot providers as of EOY 2025

Wrapping Up

To wrap it up, the main focus of the consumer agent side is to capture as much context around the traveler’s intent and desires as possible. It becomes the supremely talented executive assistant who can now engage with the supplier’s side of the process to surface the best available recommendations and present them to the traveler. Note that for some models the best possible recommendations are relative to the traveler, while for other agents (i.e., Google and Meta) they are the options that won the bid for placement from the chatbot’s advertisers.

This sets the stage for the supplier side of the diagram, which picks up the context set by the consumer agent and uses it to create highly personalized offers.