In April, our map of travel apps inside ChatGPT had a conspicuous hole where the airlines should have been. Every flight result came from an OTA or a metasearch engine, and the carriers' own fares, cabins and loyalty programs were virtually invisible. Five months later, Virgin Atlantic, Iberia, Turkish Airlines, Ryanair, Pegasus, LATAM, Virgin Australia and Akasa Air all sit in ChatGPT's flight search category, and Turkish Airlines has started buying ads inside ChatGPT conversations.

The speed of that catch-up offers insight about how carriers now see AI assistants as a sales channel to be managed alongside the website, the mobile app, GDS and NDC. It also raises a harder question. Today a traveler asks an assistant for options and clicks through to book. Tomorrow a personal agent will shop, compare and buy on their behalf, and it will not care about the homepage the airline spent years optimizing.

The assistant is becoming a storefront

Travel intent is migrating to the chat window

Semrush projects that visits from AI search will overtake traditional organic search by early 2028, and that the average visitor arriving from an LLM converts 4.4 times better than one arriving from a search results page. The explanation is intuitive for anyone who has booked a long-haul trip: by the time a traveler clicks through from ChatGPT, the comparison work is already done. For airlines, whose booking funnels are notorious for abandonment between search and payment, a channel that delivers pre-qualified intent deserves more than an innovation budget.

Being mentioned is not the same as being bookable

It helps to think of AI visibility in three layers, each answering a different question for the brand.

• Model knowledge: "Does the model know we exist?" What an LLM learned about a brand during training.

• Retrieval: "Does the model mention us?" The pages and feeds an assistant fetches and cites in real time, where SEO and GEO live.

• Action: "Can the model access our data directly?" Whether an agent can plug into the airline's booking system itself, pulling live fares and seat availability, holding an itinerary, adding a bag or changing a flight, with the same business logic the website applies.

Most airlines have spent the past two years on the first two layers. The third one, built on MCP servers and structured inventory, is where tickets are sold.

ChatGPT plugins: airline counters move inside the chat

Virgin Atlantic went live in mid-April and claimed the first airline plugin in ChatGPT. A traveler types "Premium flights to Los Angeles next month" and gets a shortlist from the carrier's live inventory, then continues to virginatlantic.com to pay. Iberia followed with an app launched at South Summit Madrid that pairs flight search with a visual fare calendar for travelers whose dates are still open. Turkish Airlines went broader, exposing 23 tools that cover fares by cabin, real-time flight status, PNR lookup and baggage allowances.

The pattern is consistent. Search, compare and select happen in the conversation, but payment remains on the airline's website (as long as in-app payment won’t be accepted). That redirect sends the user on familiar payment rails, but it also reintroduces the friction the assistant had just removed.

This is what separates an MCP app from a mention in an AI answer. The airline decides which fares appear, how cabins are presented, what the interactive view looks like and where the traveler lands afterwards. As one industry analyst put it about Virgin Atlantic's app, it works as a closed distribution environment rather than a neutral comparator. For carriers that have spent a decade pushing direct booking, that is the point.

Listed is not the same as surfaced

Until recently, apps only appeared when the user named them. That is changing. Claude introduced dynamic discovery in April 2026, followed by ChatGPT in September, surfacing relevant apps from the conversation's context, and apps we track saw traffic climb steeply in the months that followed. Paid placement is arriving too, with Turkish Airlines testing ChatGPT Ads through Criteo this month.

Personal agents: the next customer is software

From assistants that answer to agents that act

ChatGPT plugins are the visible part of a larger shift. AI has already started replacing search, and recently agents started taking action at consumer scale. Meta launched Muse on 8 September in the US, on iOS, Android, the web and WhatsApp. Within two weeks it was the top free app on theApp Store, with more than 900,000 downloads by third-party estimates. Instinct is still an invite-only beta, but it has taken over Silicon Valley feeds by doing the same jobs from an iMessage or WhatsApp thread.

Travel is already one of its main uses. Users report it booking family flights to Cancún, linking separate United reservations, monitoring sold-out tickets and checking honeymoon itineraries across Delta, Air France and Air Tahiti, where it found three errors in the traveler's own notes. As travelers hand more of the trip to an agent, the airline website stops being the place where decisions are made.

Agents compare on facts that can be queried

A human shopper can be swayed by a hero image and a well-placed promo banner. An agent reads structured data: the fare, the fare rules, the baggage allowance, the change fee, the seat map, the connection time. Whatever an airline does not expose through a machine-readable interface simply does not enter the comparison. The distribution moat therefore moves from brand recall to the completeness and freshness of the data an agent can reach.

Experience still decides the purchase

Facts alone do not explain why travelers book. Most frequent flyers carry a mental ranking of carriers (one they avoid whatever the price, one they tolerate, one they choose even at a premium). This they build from experience: legroom in the cabin, the friendliness of the crew, on-time departures and how a disruption was handled. A good personal agent will learn that ranking and weigh it as heavily as the fare.

In this way, service quality becomes a distribution asset, provided the airline makes it legible: on-time performance, cabin perks, lounge access, how rebooking works. The app itself is part of the experience too, and a carrier whose assistant answers clearly and resolves a problem in two messages earns the next booking as surely as one with a good crew.

Loyalty becomes a permission the traveler grants

The traveler's agent will know they hold Gold status, prefer an aisle and never check a bag, but only if the airline lets it connect to the loyalty account. Carriers that offer authenticated access (points balance, upgrade eligibility, member fares) give the agent a concrete reason to route the booking to them rather than to an OTA quoting the same flight. MCP, now supported by OpenAI, Anthropic, Google, Microsoft, Muse & Grok, is the open standard that makes those connections portable across assistants instead of rebuilt for each one.

What changes for selling airline tickets

Direct distribution gets a second chance

In April, OTAs were attributed in ChatGPT answers roughly five times more often than airlines. That gap came from speed of integration rather than from any structural advantage. Airlines already run NDC APIs, real-time seat inventory and dynamic pricing at scale, and an app or MCP server lets them surface their own offer, with their own fare families and loyalty benefits, before an intermediary takes the attribution.

Ancillaries move from the upsell page into the dialogue

On a website, seats, bags and upgrades are pushed in a sequence of screens after the fare is chosen. In a conversation they can be offered at the moment they matter: the traveler mentions a ski trip, and the agent prices the sports equipment allowance alongside the fare. Retailing becomes contextual, which is exactly what NDC was designed to enable.

Post-booking turns into a selling moment

"Show me all flights" is where every OTA already competes. The questions only an airline can answer come after the purchase: Which terminal? Can I check in yet? Is my flight delayed? Can I upgrade? Has the aircraft been swapped? A post-booking assistant built on airline data owns that relationship, and each answer is an opening for a paid upgrade, a lounge pass or a rebooking.

Checkout will unlock a seamless experience

Today apps hand the traveler back to the website to pay. As agentic payments mature on the major platforms, the expectation is that search, selection and purchase will happen without leaving the chat. Airlines will need order management, payment and servicing flows that an agent can call safely, with the same compliance guarantees as their web checkout.

What's next: build for the agent, not only for the chat

Start where OTAs cannot follow

A flight search app is table stakes, and the carriers that launched this year prove it can ship in months or even weeks. Differentiation will come from what only the airline knows: live operational data, the traveler's booking and loyalty profile, fare rules and servicing. The strongest first move pairs a better booking assistant with a post-booking experience.

One integration, every assistant

Building separately for ChatGPT, Claude, Gemini and Copilot repeat the channel-by-channel mistakes of the GDS era. An MCP server with an interactive app layer runs across hosts from a single codebase, and gives the airline one place to measure how agents and travelers use it.

This is the work Alpic does for airlines and across travel. We are a technology partner to Lufthansa, working with its teams on how the airline will sell to and serve travelers inside AI assistants. Our team also built Kiwi.com's MCP app early on, one of the first 50 connectors in Claude's directory and its first consumer app; we watched its traffic surge x20 once dynamic discovery was released. Everything we ship runs on Skybridge, our open-source framework, and Alpic Cloud, which hosts, tests and monitors each app in production.

The airlines that spent a decade winning travelers back from the OTAs now face a new negotiation, and this time the other side of the table is an agent.

We'll be at the World Aviation Festival from 13 to 15 October. If your team is weighing its first move into AI distribution, come and find us there.