A little over a year after OpenAI opened ChatGPT to third-party apps, roughly 30 travel brands have shipped a plugin. Online travel agencies went first, hotel chains followed in the spring of 2026, and airlines, rail operators and a cruise line arrived over the summer and fall. The first wave has done its job: travelers can now search live inventory from Expedia, Hilton, Southwest or Eurostar inside a conversation and finish on the brand's own site. What comes next is the interesting part. Almost none of these plugins know who the traveler is, which means the loyalty program, the member rate, the card in the wallet and the partner ecosystem are still waiting outside the chat. This is a summary of where the market stands as of late October 2026, what brands have learned, and how to think about the next step.
Key takeaways
- About 30 travel brands have a ChatGPT plugin as of October 2026: OTAs first (October 2025), hotels from January 2026, airlines from April 2026, rail and cruise from April to September 2026.
- Every plugin follows the same pattern: live inventory and rich cards in the chat, then a handoff to the brand's site to book. No travel plugin completes payment in ChatGPT.
- Volume is early. Booking Holdings reports LLM referrals under 1% of room nights; Eurostar reports 0.8 to 1% of tickets; Radisson reports conversion at 1.5x organic search.
- Almost no plugin knows the member. One airline authenticates loyalty members; one hotel group toggles member rates. The rest run as guests.
- The next step is Engage and Act: authenticated membership, member pricing against a real balance, the partner ecosystem, presence across assistants, and continuous operation.
What the plugins do today
Across all four groups the first-generation plugin follows one pattern, and it is a sensible one for a first release.
Search in the conversation. The traveler asks in plain language and the plugin returns live inventory: hotels, fares, cabins, routes, tours. Most render rich cards with photos, prices and a map. Southwest shows all four fare families side by side with the points each one earns. Virgin Australia shows reward seats priced in points, points plus pay, and cash.
Refine with follow-up questions. Budget, dates, neighborhood, amenities, nonstop only. The better plugins keep the context across turns.
Hand off to the brand's site to book. Payment, passenger details and reservation management happen on the brand's own website or app, usually with the itinerary carried over. This keeps the brand as merchant of record and keeps servicing, cancellations and compliance where they already work. OpenAI reached the same conclusion for the category when it folded Instant Checkout into apps in March 2026 and let brands own the transaction.
Run anonymous. With one exception, the plugins do not ask the traveler to sign in. ALL Accor can toggle between public and member rates at the program level. Turkish Airlines authenticates Miles&Smiles members on its own MCP server. Everyone else treats the traveler as a guest.
Under the hood the plugins are small. The servers that can be inspected expose between one and eight tools, nearly all search and detail lookups, and brands typically run the same server for ChatGPT and Claude with a different path per assistant. Expedia, Booking.com, Tripadvisor, Hilton, Wyndham, Accor, Turkish Airlines and Uber are all live on both. Google's AI Mode is the one surface where hotel bookings complete inside the conversation, with Google Pay and the hotel as merchant of record, and ten partners joined at its August launch.
Who has shipped a ChatGPT plugin in travel
OpenAI launched apps in ChatGPT on October 6, 2025 with Booking.com and Expedia as the travel launch partners, opened a directory in December, and renamed apps to plugins in July 2026. Travel is now one of the larger categories, with 232 plugins listed by mid-2026. The brands that have gone live break into four groups.
Online travel agencies and metasearch
Booking.com and Expedia, October 6, 2025, the two launch partners
Tripadvisor, November 6, 2025, followed by Viator in February 2026
Klook, December 2025, and TheFork, December 2025 in 13 European markets
Skyscanner, February 27, 2026, starting in the UK and US and expanding to the Gulf, Singapore and India by April
Priceline, MakeMyTrip, GetYourGuide, lastminute.com, eDreams ODIGEO, Omio, Rome2Rio, ixigo, EaseMyTrip, Despegar and 12Go, all live by the end of April 2026
Trip.com, May 2026
Hotel groups
ALL Accor, January 29, 2026, the first hotel chain, in more than 20 languages
Hyatt, disclosed February 12, 2026
Wyndham, May 6, 2026, after being the first major chain live on Claude in January
Hilton, May 2026
IHG, June 3, 2026
Radisson Hotel Group, July 28, 2026, built with Accenture in six weeks
Airlines
Virgin Atlantic, April 20, 2026, the first airline, built with OpenAI and Tomoro
Virgin Australia, June 10, 2026, the first to show reward seats priced in points
Turkish Airlines, 2026, also the first airline in the Claude directory
easyJet, October 5, 2026
Southwest Airlines, October 6, 2026, the first US airline, built with OpenAI and AWS
Rail, cruise and ground
SNCF Connect, April 23, 2026, framed as an experiment
Eurostar, August 2026, the first rail operator across Europe
Princess Cruises, September 28, 2026, the first global cruise line, built with Mobi
Uber, Turo and SIXT on the ground
Several large brands have chosen a different route for now. Marriott launched its own Ask Bonvoy assistant and joined Google's agentic booking in AI Mode. Delta expanded Delta Concierge to every SkyMiles member inside the Fly Delta app. Kayak built an AI Mode on its own site. Airbnb has said it will launch its own agent in 2027. Each of these is a reasonable call, and each leaves the brand's members unrecognized when they ask a third-party assistant for help.
What the early numbers say
Brands have been candid that volume is early. The data they have shared points in a consistent direction: the channel is small, growing fast, and converting well when it does convert.
| Brand | What they reported | When |
|---|---|---|
| Expedia Group | Answer-engine traffic is one of its two fastest-growing channels; agentic traffic shows higher purchase consideration; conversational tools yield 60%+ more traveler-intent data | Q2 2026 call; Skift Global Forum, Sep 2026 |
| Booking Holdings | Referrals from large language models remain under 1% of room nights; early signals from the ChatGPT app were faster search, better conversion and lower cancellation | Q3 2025 and Q2 2026 calls |
| Tripadvisor | AI-sourced traffic converts among the highest of any channel; fewer than 10% of travelers book with AI today | Q4 2025 and Q1 2026 calls |
| Radisson Hotel Group | Plugin visits converted to bookings at about 1.5x the rate of organic search in July and August 2026 | OpenAI case study, Oct 2026 |
| easyJet | ChatGPT referral traffic to easyJet.com up 400% year over year for flights and 125% for holidays | Launch release, Oct 2026 |
| Eurostar | AI-platform traffic up about 5x from Jan 2025 to Jun 2026, ChatGPT 85% of it; AI apps at 0.8 to 1% of ticket bookings May to Jul 2026 | Launch coverage, Aug 2026 |
| Virgin Atlantic | Its AI Concierge handled about 500,000 conversations and about £1M in bookings in five months before extending into ChatGPT | SimpliFlying podcast, Sep 2026 |
| Hilton, Wyndham, Southwest | Each described agent-driven volume as early and small while positioning for growth | Q3 2026 calls |
Figures as reported by each company, October 2026. Sources are linked at the end of the article.
Two independent data points frame the consumer side. Skift Research's October 2026 traveler study found AI present in 37% of inspiration and research touchpoints but 6% of booking touchpoints. Criteo reported that ChatGPT ad traffic converts at about 1.5x other referral channels. Both say the same thing: travelers are comfortable letting an assistant help them look, and the industry is still earning their trust for the moment of purchase.
What travelers and reviewers report
Skift, PhocusWire, Upgraded Points, The Points Guy and a number of hotel-industry practitioners have run hands-on tests over the past year. Read together, their findings describe the category rather than any single brand.
The discovery experience is good and getting better. Live prices, branded cards, maps and deep links with the itinerary carried over all work. Price accuracy against the brand site was strong for most hotel plugins tested in October 2026.
Triggering is the biggest friction. Travelers have to find, install and connect each plugin, and the assistant does not always use a plugin that is connected. OpenAI has been improving this, including suggesting plugins in the conversation, and it remains the most common complaint.
The plugin does not know the member. Reviewers used almost the same words across tests: the plugin adds the price tag, the loyalty program is missing. No OTA plugin recognizes a loyalty tier. Points prices, where shown, come without a balance. Sign-in is lost at handoff and has to be repeated on the website.
Member rates are the exception. Accor's member-rate toggle earned specific credit for giving members a reason to come through the brand's own door.
Presence across assistants is uneven. Most brands are in ChatGPT only. Testers who compared assistants found Claude's connectors surfaced brands with less friction, and a brand that is only in one assistant is absent from the others.
The analyst consensus follows from this. Forrester's mid-2026 view is that answer engines are reshaping discovery and have not yet reshaped checkout. A 20-person Hospitality Net panel in April reached the same conclusion: treat the assistant as a discovery engine today and invest in the structured data, identity and partner content that make the next step possible.
The next frontier: from Discover to Engage to Act
A useful way to read the market is as three phases of a member journey. Nearly every travel plugin today lives in the first one.
Discover. Browse the brand's inventory and content as a guest. This is where the first wave landed, and it was the right place to start.
Engage. The assistant knows the member: status, balance, bookings, enrollments, the card in the wallet. It shows member rates, points prices against a real balance, and the partner offers that fit this trip. Accor and Turkish Airlines have taken the first steps here.
Act. The member does something with that context without leaving the conversation: activates an offer, enrolls with a partner, redeems, or completes a purchase with the card already stored in the program wallet. Google's AI Mode shows what Act looks like for a hotel booking. For the loyalty side of the business it is still open ground.
Five capabilities separate Discover from the other two, and as of October 2026 almost no travel plugin has them.
Authenticated membership inside the assistant. One airline authenticates members today, on its own server. The rest run as guests.
Member pricing and benefits. Member rates, status perks, and points priced against the member's actual balance, with the same accuracy the brand site has.
The partner ecosystem. Every plugin exposes first-party inventory: rooms, fares, cabins. None surfaces the shopping, dining, gift card, rideshare or cobrand card partners that make up a large share of program value and that a trip naturally triggers.
Presence across assistants. ChatGPT, Claude, Google and Meta's Muse are all taking travel connections now. A member who asks in the second assistant should get the same brand, the same identity and the same rules as in the first.
Continuous operation. A plugin is a product, with guardrails, caching, cost tracking, audit, and a way to evolve the tools and widgets as the program's partners and campaigns change. Most first-wave plugins were built as a project with a fixed scope.
None of this is a criticism of the first wave. It is the natural sequence: get discovered, then get recognized, then let the member act.
A framework for evaluating a travel plugin
Whether you are scoring your own plugin or a competitor's, these eight questions cover what matters. Score each 0 (absent), 1 (partial) or 2 (complete).
| Dimension | Question | Discover 0 to 1 | Engage and Act 2 |
|---|---|---|---|
| Inventory | Does it search live availability and pricing with rich cards and maps? | Yes, with a handoff to book | Yes, with the member's rate and points shown against a real balance |
| Accuracy | Do prices and points match the brand site? | Usually | Governed, cached and audited so they always do |
| Identity | Does the assistant know who the member is? | Guest only | Authenticated member with status, wallet and bookings |
| Benefits | Are member rates, perks and status recognized? | Program-level toggle at best | Applied automatically for the signed-in member |
| Partners | Are the program's partners and the cobrand card part of the conversation? | First-party inventory only | Partner offers matched to the trip and the member, with the card in wallet |
| Actions | What can the member do without leaving the chat? | Browse and click out | Activate, enroll, redeem, reserve, purchase with the stored card |
| Reach | Which assistants carry it? | One | ChatGPT, Claude and others from one server |
| Operation | How is it run after launch? | A project that shipped | A product with guardrails, cost tracking, audit and a build-and-evolve loop |
Score each dimension 0 (absent), 1 (partial) or 2 (complete). A first-wave plugin typically scores 6 to 8 of 16.
A first-wave plugin typically scores 6 to 8 of 16, almost entirely in the Inventory, Accuracy and Reach rows. The gap is identity and everything that follows from it.
How to choose your next step
The right move depends on where your brand is today.
You have no plugin yet. Start with Discover, and build it so Engage is a configuration change rather than a rebuild. That means designing the server around member identity from day one, even if the first release runs as a guest, and publishing to more than one assistant from the same server.
You have a Discover plugin and it is working. Measure what the first wave has measured: referral growth, conversion relative to organic, and intent data. Then add the member. Sign-in, member rates and points against a real balance are the three changes reviewers and travelers ask for first.
You have a plugin and you run ads in ChatGPT. You already pay for discovery. The return on that spend goes up when the member who clicks is recognized, sees their rate, and can act. Treat the plugin as the landing experience for the ad.
You have your own assistant and no third-party plugin. Keep the assistant. Your members also ask ChatGPT, Claude and other assistants today and they get answered as strangers. A member-aware presence in third-party assistants complements the owned assistant and keeps the booking and the relationship with you.
You run a loyalty program with a partner ecosystem. This is where the largest gap is. A trip is a sequence of partner moments: shopping before departure, dining and tours at the destination, rideshare at landing, the cobrand card throughout. No travel plugin surfaces any of it yet.
Three questions to ask any vendor or internal team:
Can a member authenticate inside the assistant, and does the session carry through the handoff?
Does one server publish to every assistant, with the same rules and the same audit trail?
Can the business team evolve the plugin, its tools and its widgets after launch without a new engineering project?
Where SmartCXP AI™ fits
Dfinitiv built SmartCXP AI™ for the second and third phases. It brings the loyalty program into the conversation so the brand shows up as itself rather than as a guest: the member authenticates, the assistant engages based on status and profile, and the member can act using their information and the card stored in the program wallet. One server publishes to ChatGPT, Claude and other assistants. SmartCXP AI Forge lets product and digital teams build the plugin, its tools and its widgets and keep evolving them; SmartCXP AI Engine runs them with guardrails, caching, cost tracking and audit; SmartCXP AI Center operates them. The platform underneath, SmartCXP Cloud, has run in production for a major US carrier for a year with 100% availability.
If you already have a first-wave plugin, we can show you what the Engage and Act phases look like in your brand. AI Forge offers a free preview at forge.smartcxp.com.
Frequently asked questions
What is a ChatGPT plugin for travel?
A ChatGPT plugin (called an app until July 2026) is a brand-built connection, based on the Model Context Protocol, that lets ChatGPT search the brand's live inventory and show results inside the conversation. For travel brands that means hotels, flights, trains, cruises or tours, usually with photos, prices and a map, followed by a link to book on the brand's own site.
Which travel companies have a ChatGPT plugin in 2026?
About 30. Online travel agencies include Expedia, Booking.com, Tripadvisor, Viator, Priceline, Skyscanner, Trip.com, MakeMyTrip, eDreams, Klook and GetYourGuide. Hotel groups include ALL Accor, Hyatt, Wyndham, Hilton, IHG and Radisson. Airlines include Virgin Atlantic, Virgin Australia, Turkish Airlines, easyJet and Southwest. Eurostar, SNCF Connect, Princess Cruises, Uber, Turo and SIXT round out rail, cruise and ground.
Can you book a hotel or flight inside ChatGPT?
As of October 2026, no travel plugin completes payment inside ChatGPT. Plugins search and compare, then hand the traveler to the brand's website or app with the itinerary carried over. OpenAI folded its Instant Checkout feature into apps in March 2026, leaving the transaction with the brand. Google's AI Mode is the one assistant where hotel bookings complete in the conversation, using Google Pay with the hotel as merchant of record.
Do ChatGPT travel plugins recognize loyalty programs?
Mostly not yet. Turkish Airlines authenticates Miles&Smiles members on its own server, ALL Accor can toggle member rates at the program level, and a few plugins show points prices without a balance. No OTA plugin recognizes a loyalty tier, and sign-in is typically lost at the handoff to the website.
How much business do ChatGPT plugins drive for travel brands?
Brands describe the channel as small and growing. Booking Holdings reports LLM referrals under 1% of room nights. Eurostar reports AI apps at 0.8 to 1% of ticket bookings. Radisson reports plugin visits converting to bookings at about 1.5x the rate of organic search. easyJet reports ChatGPT referral traffic up 400% year over year.
How do you evaluate a travel ChatGPT plugin?
Score eight dimensions from 0 to 2: inventory, accuracy, identity, benefits, partners, actions, reach across assistants, and ongoing operation. A first-generation plugin typically scores 6 to 8 of 16, with the gap concentrated in identity and the capabilities that depend on it.
What comes after discovery for travel plugins?
Two phases: Engage, where the assistant knows the member's status, balance, bookings and card in wallet and shows member rates and relevant partner offers; and Act, where the member activates an offer, enrolls with a partner, redeems or purchases with the stored card without leaving the conversation.
Sources
Unite.AI: OpenAI reports early results for Radisson's ChatGPT plugin (Oct 2026)
easyJet: flights and package holidays in ChatGPT (Oct 5, 2026)
Rail Business Daily: Eurostar first rail operator in ChatGPT (Aug 2026)
Southwest Airlines: first US airline plugin in ChatGPT (Oct 6, 2026)
Virgin Australia flight search lands in ChatGPT (Jun 10, 2026)
SimpliFlying: Virgin Atlantic CCO on going first on ChatGPT (Sep 8, 2026)
Princess Cruises: first global cruise line with an AI planning app (Sep 28, 2026)
Skift: Testing ChatGPT apps and Claude connectors (Jul 6, 2026)
Upgraded Points: 12 AI travel booking tools tested (Oct 6, 2026)
Hotelemarketer: I tested every ChatGPT travel app (Mar 30, 2026)
Forrester: The state of agentic commerce in mid-2026 (May 28, 2026)
Hospitality Net panel: Is "Book on ChatGPT" hitting the same wall as "Book on Google"? (Apr 7, 2026)
