How AI Travel Assistants Could Cost Airlines Sales

The Booking an Airline Loses to an AI Assistant Leaves No Trace: No Bounce, No Abandoned Cart, No Dip in the Dashboard

A survey of 1,000 US adults commissioned by travel concierge company Greetwell in August 2026 found that 53% had tried AI for travel. Only 14% used it to research and book, while 34% used it to plan a trip without booking through it. Read quickly, that looks like reassuring news for airlines: most people still book the old way. Read closely, it describes a different problem. If a third of AI users are doing their planning inside an assistant, the decision about which airlines make the shortlist is increasingly made before the traveller ever reaches an airline website, in a place the airline’s own booking analytics cannot see.

What the Survey Actually Shows, and What It Does Not

The Greetwell figures describe a trust gap rather than a rejection of AI. People are comfortable asking an assistant for ideas, and far less comfortable letting it transact. The same research found that 55% of AI users hit at least one bad recommendation of any kind, and one in six was pointed to a place, tour or activity that was not real. Users who ran into problems were more than four times as likely to report declining trust. A separate Expedia Group survey of more than 5,700 adults in the US, UK and India found 53% comfortable receiving AI travel suggestions but 68% preferring to book with established travel brands rather than AI agents.

That combination matters for airlines. The booking itself, for now, still mostly happens on a familiar site. The shortlist, though, is increasingly being formed upstream by an assistant. An airline can be priced competitively and still never appear in that shortlist if the assistant cannot work out what the airline is actually offering.

Why the Lost Customer Never Shows Up in Analytics

The Breaking Travel News analysis of this problem captures it in its headline: “No Bounce. No Dip. Just a Booking You Never Knew You Lost.” Traditional web analytics measure what happens once someone arrives: how many visited, how many abandoned a search, where they dropped out of the funnel. A traveller who asks an assistant for the best flight options and receives an answer that never mentions a given airline generates no visit, no search and no abandonment. There is nothing to measure because the airline never entered the conversation. The missed sale does not register as a loss; it simply fails to occur.

The same analysis draws a useful distinction between two tests that airlines tend to conflate. One is findability: whether an AI system understands the airline and its products correctly. The other is bookability: whether an agent can actually complete a transaction. An airline can pass one and fail the other, and each fails in a different place.

Why Inconsistent Product Descriptions Are the Likely Culprit

An airline’s offer is described in many places at once: its own site, its distribution partners, comparison sites, travel blogs and forums. Fare families, baggage allowances, lounge access and loyalty benefits are often described differently, or out of date, across those sources. A traditional search engine returns links and lets a human reconcile the differences. An AI assistant has to synthesise a single answer, and when sources conflict about what a fare includes, the assistant has a real choice: state something that may be wrong, hedge, or leave that product out in favour of one whose details are consistent and easy to state. That is the mechanism the Greetwell-linked coverage points to, and it is reasoned inference as much as measured finding: ambiguity makes a product harder to recommend confidently.

That is also precisely the problem Bonafide, one of the four World Aviation Festival finalists this feed covered this week, says it is built to solve, by creating context layers that help AI systems “understand and accurately represent” a brand’s business.

Who the Assistants Actually Send Travellers To

The same analysis reports data from a study of 101 airlines that makes the stakes concrete. For 85 of the 101 airlines, AI assistants linked more often to travel agents and comparison sites than to the airline’s own website. The most frequently cited intermediaries were Kayak (in 27% of answers), Skyscanner (23%) and Trip.com (18%). The assistants also behave very differently from each other: ChatGPT linked the airline’s own site in 50% of answers and an intermediary in 5%, while Claude’s figures were 11% and 50%, and Perplexity’s were 19% and 93%.

That spread is an important caution against treating “AI” as one channel. An airline’s visibility can look healthy in one assistant and be nearly absent in another, and in the assistants that favour intermediaries, the airline may be reaching customers only through someone else’s listing and someone else’s margin. It also puts a new light on moves this feed has already covered: China Eastern listing on KAYAK’s AI search, and Eastar Jet partnering with Fliggy’s Qwen-powered assistant. Both are airlines securing a place inside an AI-mediated intermediary, rather than waiting to be found on their own sites.

How This Differs From Search-Engine Optimisation

Traditional search optimisation is largely about ranking: getting a page near the top of a list a human then scans. Making products legible to AI is a different task. The output is not a list but a single synthesised answer, so what matters is whether the underlying facts are consistent, machine-readable and trustworthy enough to be quoted accurately. A well-ranked page with contradictory fare details can lose out to a plainer one whose details agree everywhere. The work moves from persuading a human reader to being unambiguous to a system that cannot afford to guess.

How Much This Might Be Costing, and Why to Treat the Number Carefully

The analysis estimates that, for the median airline, the modelled revenue opportunity from AI findability alone is about $600,000 a year, calculated from the airline’s estimated digital revenue, the share influenced by AI assistants and the airline’s measured gap, capped at 1.2% of digital revenue. It is worth being clear about what this is. It is a modelled estimate with an explicit cap, not a measured loss, and it comes from an analysis that exists to make the case for taking this problem seriously. It is more useful as an order-of-magnitude indicator than as a figure to put in a budget.

The Counterweight: Booking Is Still Mostly Human, for Now

The strongest argument against alarm is the survey’s own booking number. Fourteen percent using AI to both research and book is real but small, and the trust data suggests many travellers will keep a human hand on the final purchase for some time. The risk is not that AI takes over booking overnight. It is that it quietly reshapes which options travellers consider, while the transaction stays where it always was. Airlines that treat AI visibility as a future problem may find the shortlist has already narrowed.

What Airlines Can Actually Do

The practical implication is unglamorous. Keep product descriptions consistent across the airline’s own site, distribution partners and any third-party data feeds. Test individual assistants directly and read server logs, as the analysis suggests airlines can. Treat findability and bookability as separate problems, and track both. None of this requires abandoning the existing booking funnel, only recognising that a growing share of the competition for a traveller now happens before the funnel begins, in a conversation the airline is not part of unless its product information is clear enough for an AI to repeat accurately.