Airbnb Says Guests Still Want to Choose

Airbnb says third-party AI agents currently bring it very few bookings. Its explanation is that choosing a holiday often involves browsing photographs, comparing places and deciding with other people. For small businesses, the useful distinction is between tasks customers welcome help with and choices they still want to make themselves.

An agent can book a room, but should it pick the room? Airbnb's latest results raise that question at an awkward moment for companies promising to automate shopping from search to payment. The answer matters well beyond travel, especially for small businesses whose customers choose with their eyes, their preferences and their sense of who they want to buy from.

Half a million bookings

On Airbnb's 5 November earnings call, CEO Brian Chesky was asked about bookings arriving through outside AI agents, including Muse and Instinct. He said Airbnb handles nearly half a million bookings a day, while those agents contribute a very small number. Traffic from them had peaked in early October and had remained broadly steady over the following four weeks, according to Chesky. There was no claim that customers would never use agents to book travel, only that the activity was currently marginal on Airbnb.¹

That is an awkward result for an industry that often talks as though an AI assistant will soon complete every purchase. Airbnb is one of the largest and most recognisable online travel marketplaces, and its inventory is precisely the sort of material a search assistant might help people navigate. Yet early agent traffic has scarcely translated into completed reservations there. One company's early results cannot settle agent commerce, but they do challenge the idea that every buyer wants full automation.

Airbnb is certainly not rejecting the technology. Its own results show that approximately 40% of customer support issues in the third quarter were resolved without a human agent, up from roughly a third in the preceding quarter. The company reported that support cost per booking fell by more than 17% year on year, driven in large part by its AI assistant. Those are company-reported operational figures, and they describe a quite different activity from choosing accommodation.²

A guest who needs a receipt, a change to a reservation or help understanding a policy probably wants the matter sorted quickly. The task has a practical end point and relatively little pleasure attached to the work of completing it. Choosing a cottage for a family holiday is different, because the comparison can be part of the anticipation. Airbnb is automating the chores while being careful about the choice, and that distinction deserves more attention from smaller businesses considering where to spend their money.

This is particularly relevant when an owner is offered one tool for customer enquiries, another for marketing and a third to make purchasing completely autonomous. Owners should ask what customers will happily delegate and where removing their involvement would diminish the experience. There is no universal answer across a coffee order, a haircut, a handmade gift and a week away.

Browsing is part of buying

Airbnb's shareholder update offers a more detailed picture of what its customers actually do. The company says 80% of travellers enjoy researching trips, more than 90% of Airbnb guests browse multiple listing photographs, over 75% use maps when deciding where to stay and 70% read reviews before booking. These are Airbnb's figures, not independent evidence about every type of purchase. Together, though, they suggest that the apparently inefficient browsing has a job to do.²

Imagine a couple looking for a dog-friendly cottage for a long weekend. An assistant can narrow the results by location, price, parking, pet rules and available dates in seconds. It can compare cancellation terms and flag which listings appear to meet the requirements. But the couple may still want to examine the kitchen, judge whether the garden is secure, see how the bedrooms connect and decide whether the place feels right for them.

Those judgements depend on information that is difficult to reduce to one neat answer. A photograph of a sunny dining room may matter more to one guest than a slightly shorter drive; another may care about the layout or proximity to a walking route. Even the same family can disagree about what makes a good holiday. Airbnb says the average booking involves three guests, which helps explain why it is designing around shared decisions rather than assuming a single person will issue a command and accept the result.²

That explains another apparent contradiction in Airbnb's plans. The company expects to begin rolling out its own specialised travel agent in the first half of 2027. Rather than presenting everything through a conventional chat window, it says it is building an interface that can assemble maps, photographs, comparisons, calendars and itineraries around the guest's question. It is a planned product, not something already available to every guest, and its success remains to be seen.²

The design tells us what Airbnb thinks the customer still needs. A shortlist is valuable because it leaves a person with sensible options and the evidence needed to choose between them. A comparison that puts cancellation rules beside photographs is useful because it removes the tedious work without hiding the differences. In that model, the assistant earns its place by making exploration easier and more informed, not by racing to close the booking before anyone has looked at the rooms.

For an independent guesthouse, that has an immediate implication. Accurate amenities, well-lit photographs of the actual property and clear descriptions of what is included may become more valuable as discovery gets quicker. An assistant might surface a property for someone who needs step-free access or somewhere close to a station, but the listing still needs to persuade the guest once they arrive. Owners who replace genuine photographs with attractive but misleading AI images risk undermining precisely the evidence that makes a choice feel safe.

Firmly makes another bet

The same week, shopping infrastructure company Firmly announced a $15 million Series A and launched Ask Agent. The product allows shoppers on a merchant's website to describe what they want and receive answers based on that shop's catalogue, policies, live pricing and inventory. Firmly is also building connections that let merchants sell through outside AI shopping channels without constructing a separate integration for each one. The merchant keeps its systems and customer relationship while the conversation becomes another way to shop.³

At first this seems to run against Airbnb's argument. Firmly wants to make a purchase easier to complete in a conversation, while Airbnb is explaining why customers enjoy looking around. But the difference is easier to see when we consider the purchase itself. Someone replacing a particular household item might want the right specification, the best available size and a reliable delivery date, with very little browsing in between.

Even within retail, that preference changes from one purchase to another. A shopper who knows exactly which skincare product they use may appreciate an assistant that finds the correct version and places the order. Someone choosing a wedding outfit, a ceramic vase or a sofa might want to compare textures, proportions, colours and alternatives before committing. A conversational search can still help, provided it gives the shopper a better set of choices instead of treating the first plausible match as the final answer.

Firmly's technical approach matters because a confident recommendation is worthless when the product is unavailable. Its integrations use merchant-approved connections to the systems behind the storefront, so prices, stock and checkout details come from the business's own records. That is substantially more dependable in principle than having an agent guess availability from an old product page. The announcement does not show that Ask Agent will improve conversions for every shop.³

There is also an affordability question for the smallest operators. Axios reported Firmly's merchant pricing at $2,000 to $15,000 a month, depending on merchandise volume, although its published report does not break out separate pricing for Ask Agent. That is an important distinction when discussing AI for small business: a promising technology can show where commerce is going without being the appropriate purchase for a six-person boutique today. It is better to examine what the customer wants to do than to assume each new shopping channel deserves a subscription.⁴

The common thread between Airbnb and Firmly is the amount of decision-making the customer wants to keep. An agent could be excellent at retrieving product facts, narrowing an overwhelming selection and processing an order after the person has made up their mind. That does not require every transaction to disappear into a single prompt. For independent retailers, the practical ambition may be simpler: let customers ask better questions of the catalogue, show them real options and make the chosen item easy to buy.

Bluehost builds the next step

Bluehost's 6 November launch moves the discussion from shopping agents to a task many businesses have wanted to fix for years. Its AI Builder lets users describe a website or working web application in ordinary language, then revise the result through conversation. The company lists booking systems, customer portals, quote requests and internal tools among the possible applications, with hosting included and a free starting plan. This is a launch announcement describing available capabilities, not evidence that any generated application is automatically ready for a busy business.⁵

For a small accommodation provider, the appeal is obvious. A host might want a simple enquiry form for longer stays, a way to capture arrival preferences or a page that explains optional extras before a guest books. Building that first version without commissioning a bespoke project could save significant effort, although the amount saved would depend on the specific business. The guest would still be able to inspect the property, ask questions and decide whether to stay.

Think about a beauty studio with several treatment lengths and employees who work different days. The owner may be able to describe a booking system that reflects those details rather than forcing everything into a generic appointment template. That makes the software more closely resemble the way the business already works. The owner would still need to test simultaneous bookings, cancellations, incorrect inputs and the handling of customer information before relying on it.

Making a first version easier to build also changes how a business can experiment. Instead of committing months to a full customer portal, it could test whether customers want to submit a request online, which questions they ask and which steps still require a phone call. A working application can be measured against completed enquiries or fewer missed bookings, rather than the fact that someone generated a polished web page. A useful booking flow begins with the customer's actual task, including the awkward cases that appear after launch.

Bluehost, Firmly and Airbnb operate at very different scales, but they address consecutive parts of a customer journey. Someone discovers a business, works out whether the offer suits them and then needs a dependable way to act. An AI assistant may help with any of those steps, while the business remains responsible for the promises made in its photographs, product information and booking terms. Better software will make the handover between those moments less frustrating, without forcing the entire experience into a chat box.

Give people something to inspect

The figures in Airbnb's update are a reminder that visual material is not merely decoration around the transaction. More than 90% of its guests browse multiple photographs before booking, according to the company, and reviews and maps also help people form a view. For a smaller business, photographs of the actual place, product and people can perform a similar job. They give a prospective customer evidence that a claim is credible and an experience is worth paying for.²

A salon can show the finish of a particular cut and explain how it was achieved. A restaurant can photograph the dish the kitchen is serving this week, alongside the setting a guest will walk into. A boutique can show how an item fits, moves or looks in natural light. None of those examples requires an invented lifestyle scene, and each tells a potential buyer something that a polished generic caption cannot.

This is where Instagram for small business deserves to be treated as more than a publishing schedule. A good small business Instagram strategy gives strangers enough real detail to consider a visit, a booking or a purchase. Instagram AI content can help owners sort existing media, draft captions in their normal voice and plan a manageable posting rhythm. The value of those tools depends heavily on the material they start with and whether the post accurately reflects what customers will find.

The same lesson applies to an assistant reading a website. A shop with vague descriptions and inconsistent product photographs leaves both the customer and the software guessing. A host that explains parking, access, room layout and pet rules clearly makes the choice easier for someone browsing directly or arriving through an AI recommendation. Better content is therefore useful in two directions: people can judge the offer, and tools can retrieve facts without inventing them.

An owner deciding what to improve this month could begin by watching where customers hesitate. Do they ask repeatedly whether a room has a separate entrance, whether a jacket runs small or whether a salon takes evening appointments? Those questions suggest missing information or a frustrating step in the process. Fixing them may be more valuable than generating another collection of posts designed only to keep the feed busy.

Let the guest pick

Airbnb's comments should not be read as proof that autonomous shopping has reached a dead end. Agents are young, customer habits change and certain purchases will become much easier to delegate as the technology improves. What Airbnb has identified is a reason to preserve browsing where browsing is part of the value. Its own plans combine automated support with an experience that still expects guests to compare, discuss and decide.

For smaller businesses, a sensible test is to separate the tasks customers find tiresome from the moments they enjoy. Faster answers about availability, fewer form fields and more reliable checkout are worth pursuing. Genuine photographs, meaningful comparisons and the pleasure of finding exactly the right place or product deserve equal care. That gives owners a much clearer brief for AI than a general instruction to automate every interaction.

A guest who has asked an assistant to find a cottage may still want to open ten photographs before choosing one. That is not a failed automation. The software did its job when it made those ten photographs worth looking at, and the business did its job when the reality matched the promise.

Sources