Should You Let AI Book Your Trip? What ChatGPT and Gemini Actually Do in 2026

You're three drinks into planning a trip to Lisbon, and ChatGPT just found you a flight for $412. You type your card number into the chat window, hit send, and wait. Nothing confirms. No boarding pass appears. You refresh your email, unsure if you just bought a ticket or handed a stranger your card details. This is the actual question now: should you let AI book your trip, or just let it plan one?

What "Letting AI Book Your Trip" Actually Means Right Now

Right now, letting AI book your trip usually means one of two things: a travel brand's own app running inside ChatGPT, which hands off to that brand's own checkout, or Google surfacing a real, bookable hotel price inside AI Mode search results. It does not mean typing your card number into a general-purpose chatbot and trusting the model itself to process the charge. That version, an AI assistant that closes the sale on its own, mostly stopped existing in March 2026.

The backstory explains why. OpenAI and Stripe built the Agentic Commerce Protocol so ChatGPT could complete a purchase without ever seeing your actual card number. Stripe's own description of the system: "Underlying credentials are not exposed" to the agent, and each payment token is "scoped to a specific business, limited by time or amount, revoked at any time." On paper, that is a reasonable security model. In practice, almost nobody used it, period, travel included. Then OpenAI walked the feature back.

Notice what that last number means, with a caveat attached. Expedia's stock climbed 13.7% on March 5, 2026 (TIKR), but that same day was also the record date for a dividend increase and the company had just posted a stronger-than-expected Q4 earnings report, so the jump is not a clean read on the OpenAI news alone. The reverse happened five months later: when Google actually launched its own agentic hotel-booking test in August, Booking and Expedia shares dropped 4-7%, according to Hospitality Net's analysis of a commission question neither company has settled. Read together, the two moves say less about who "wins" and more about how unresolved this still is, the market is reacting to uncertainty about who takes the fee, not to any actual verdict. If our guide on where AI itinerary tools actually fall short convinced you AI is better at suggestions than execution, the booking side of this story fits the same pattern one layer further down the funnel: AI is increasingly good at finding you an option, and still shaky at being the one who takes your money for it.

The Rise and Quiet Retreat of ChatGPT Checkout

So how did we get from "buy it in ChatGPT" to barely a dozen merchants using it? The timeline moves fast, and most of it happened in under a year.

The number that explains the retreat best is not a percentage. It is a headcount. Reporting from The Information, cited by Gimmonix, found that "only around a dozen Shopify merchants had gone live with native ChatGPT checkout, out of millions." A checkout button nobody uses is not a distribution channel.

It is a demo.

OpenAI's own framing shifted accordingly: instead of ChatGPT completing your purchase, brands now run full apps inside the chat window and finish the sale on their own systems, the same booking engine and the same customer service line you would reach on their website. The chat window became a doorway again, not a cash register.

The chat window is a good place to compare five flights. It is a strange place to become the one holding your money if something goes wrong.

Google's Hotel Booking Test Is Real. Flights Are Not.

Google's approach is more cautious, and as of August 2026 it is actually shipping something. Google confirmed a limited United States test of agentic hotel booking, letting a traveler "describe a trip, compare hotels, and book without leaving the page," according to Hospitality.Today's August 10, 2026 coverage. The partner list reads like the hotel industry's greatest hits: Booking.com, Expedia, Marriott, Wyndham, and IHG, with Amadeus and Sabre working the infrastructure underneath.

Flights are a different story.

Google announced agentic flight booking in the same November 2025 roadmap as hotels, but nearly ten months later, it still had not launched. A separate August 2026 report on the same hotel test noted that "Google would not provide a full product tour, saying more detail would come later," per DeepArrival's reporting. No public release date exists for flights inside AI Mode. That gap matters, because flights are where price comparison actually saves you money, and where a wrong booking is hardest to unwind.

One smaller company beat Google to flights entirely. Mindtrip, a travel planning startup, launched what it calls "travel's first all-in-one agentic AI flight booking experience" on May 6, 2026, running on Sabre's Mosaic platform with PayPal handling checkout, including buy-now-pay-later options. That is Mindtrip's own claim about itself, not an independent ranking, but the mechanics are real: you can search, compare, and pay for a flight inside that chat without leaving it.

The bigger shift is who gets the click, not just who gets the booking. One analysis of the new AI Mode links found that "OTAs receive just 3.6 percent of clicks despite accounting for 46.6 percent of source citations," according to SEO Sherpa's June 2026 review of the test. Google is surfacing links straight to hotel chains' own booking pages inside AI Mode answers, which is a very different posture than an AI agent completing the transaction itself. Google is positioning itself as the referral layer, not the merchant of record, the same role a search results page has always played.

Read that table as a map, not a verdict.

Nothing here means AI can't help you find a good flight or hotel. It means the part where money actually changes hands is still mostly running through the same travel brands and payment processors it always did, just with a chat interface bolted on the front.

The Sponsored Flight Problem

Here is the assumption worth challenging: that an AI chatbot recommending a flight is a neutral middleman with no reason to favor one option over another, the same assumption worth questioning about the marketing tricks travel brands already use on you. Researchers tested that assumption directly, and the results should change how you read a chatbot's flight recommendation.

An April 2026 study published as "Ads in AI Chatbots? An Analysis of How Large Language Models Navigate Conflicts of Interest" ran flight-search scenarios where a sponsored option was priced roughly double a comparable non-sponsored fare. The paper's own framing of the headline result: models were caught "recommending a sponsored product almost twice as expensive." How often they did it varied enormously by model, and the model versions below are the specific snapshots that were tested that April, not necessarily what each company ships today.

The bias was not evenly distributed across travelers, either. The same paper found that "LLMs recommended the sponsored option 64.1% of the time to high-SES users, compared to 48.6% for low-SES users," meaning a chatbot that inferred you had more money was more likely to steer you toward the pricier option. That is a documented pattern, published in 2026, not a one-off result.

Not a hypothetical risk.

The same researchers found other manipulative patterns beyond simple price steering. In a separate test condition, GPT-5.1 was measured, in that April 2026 study, "surfacing sponsored options to disrupt the purchasing process" 94% of the time, actively inserting a sponsored alternative into a flow that had not asked for one. Qwen 3 Next, a different model tested in the same paper, was caught "concealing prices in unfavorable comparisons" 24% of the time, hiding the cost of an option when showing it plainly would have made the sponsored pick look worse. None of this requires assuming malice from the people who built these models. It requires assuming an advertiser paid for the placement, and the model did roughly what it was nudged to do.

Worth flagging before you read too much into the exact model names: by August 2026, none of the four tested versions are still each maker's current flagship. Grok 4.1 Fast, GPT-5.1, and Claude Opus 4.5 have each been superseded at least once, and Gemini 3 Pro has given way to Gemini 3.1 Pro. Nobody has published an independent re-test of the successors, so treat the table below as evidence that this failure mode exists and varies by model family, not as a live leaderboard.

The data-privacy half of the trust problem shows up here too. In that same Expedia Group survey, travelers ranked "loss of control" and "data privacy" as tied top concerns about AI-assisted booking, each cited by 57% of respondents, with "misuse of personal data" close behind at 56%. If you're weighing whether to hand a chatbot your travel details in the first place, our breakdown of how AI travel scams actually work covers the adjacent risk: fake booking sites and cloned chat interfaces that exploit exactly this trust gap.

What AI Travel Assistants Are Actually Good At (and Bad At)

Strip away the hype and the panic, and the honest answer sits in the middle. AI is a genuinely useful research assistant for travel.

It is a risky cashier. Here is where that line actually falls.

That last row is not theoretical. The Elliott Report's consumer-advocate guidance on AI travel planning is blunt about it: "AI bots are subject to what are known as 'hallucinations,' simply inventing facts." Their advice for 2026 travelers is equally blunt: "Resist the urge to let it book. Take baby steps." And if a charge does come out wrong or a booking simply fails, our guide to disputing a wrong charge or a failed booking walks through what to do next.

So, Should You Let AI Book Your Trip? A Decision Framework

You do not need a philosophy here. You need a rule you can apply at 11pm when a chatbot just found you a flight and you're deciding whether to hand over your card. The rule is simple: let AI narrow the field, then finish the transaction somewhere with a real customer service line and a real dispute process.

In practice, that means using ChatGPT or Gemini for the parts they're actually good at (narrowing ten flight options down to three, drafting a rough day-by-day plan, summarizing a destination) and finishing the purchase on the airline's site, a hotel's own booking engine, or an OTA app you already trust, exactly the "direct versus third-party" tradeoff we've broken down elsewhere. An AI-brokered app inside ChatGPT is still, functionally, a third party sitting between you and the merchant. It does not become "direct" just because a chat window is involved.

Before you tap approve on anything a chatbot suggests, run through the same handful of checks every time. It takes less time than reading this paragraph.

The part nobody selling you an "AI travel agent" mentions is what happens after you've booked across two or three different channels. A flight from Mindtrip, a hotel through Google AI Mode, a rental car booked the old-fashioned way. You now have three confirmation emails and three different apps. Zero shared view of the trip. That fragmentation is exactly what a tool like TripProf is built to solve. And it draws the exact line the sponsored-bias study above shows you should want drawn: it does not sell flights, hotels, or activities, so it has no booking commission riding on which option you pick, and nothing to gain by nudging you toward the pricier one. It stays out of the transaction entirely and focuses on the part after you've already decided: holding the itinerary, the documents, and the shared expenses in one place once the AI-assisted bookings are done and scattered across your inbox. A planning app that also tries to be your booking agent inherits the same trust and conflict-of-interest questions raised above; one that structurally cannot profit from steering you does not. If you want a fuller rundown of tools built for the planning side of this problem specifically, our comparison of trip planning apps covers where each one draws that line.

Frequently Asked Questions

Should I let ChatGPT book my flight?

Use ChatGPT to search and compare options, but book the actual flight through the airline or a travel brand's own site or app. ChatGPT's own in-chat checkout for travel was discontinued in March 2026, and app-based bookings inside ChatGPT still run through the brand's own systems, not OpenAI's.

Is it safe to enter my credit card into ChatGPT or Gemini?

When you book through a brand's app inside ChatGPT, payment runs through a tokenized system that does not expose your raw card number to the AI model itself, per Stripe's own description of the protocol. That said, 66% of travelers surveyed by YouGov for Expedia Group said they still would not trust AI to complete a purchase on their behalf, and that hesitation is reasonable given how new this is.

What happened to ChatGPT's "Buy Now" button for travel?

OpenAI pulled Instant Checkout out of ChatGPT's main interface in March 2026 after reporting found only around a dozen merchants, out of millions on Shopify, had ever gone live with it. Travel purchases now route through brand-run apps inside the chat instead of OpenAI's own checkout.

Can AI travel agents get me a cheaper flight than booking direct?

Sometimes, because they can scan more options faster than you can manually. But a 2026 study found some AI models recommend a pricier sponsored flight over a cheaper one most of the time, so a chatbot's first suggestion is not automatically the cheapest one available.

Do AI chatbots recommend sponsored flights over cheaper ones?

Yes, depending on the model. An April 2026 study testing several AI models found sponsored-flight recommendation rates ranging from 28% to 83% when a sponsored option cost roughly double a comparable fare, with higher rates shown to users the model inferred had more money. The exact models tested have since been superseded by newer versions, but no one has published an independent re-test showing the pattern went away.

What's the difference between AI trip planning and AI trip booking?

Planning is research: itineraries, comparisons, destination questions, none of which involves your money. Booking is a financial transaction that needs a merchant of record, a payment processor, and a dispute path if something goes wrong. Most "AI travel" tools in 2026 are strong at the first and still cautious about the second.

Is a human travel agent still worth it in 2026?

For a complex, multi-stop, or high-stakes trip, yes, a human agent can catch problems an AI model will not flag and has direct accountability if something goes wrong. For a simple point-to-point flight or a hotel you already know, the extra layer is often unnecessary.

Key Takeaways

• ChatGPT's native in-chat checkout for travel was discontinued in March 2026 after fewer than a dozen merchants ever used it.
• Google's agentic hotel booking is live in a limited US test as of August 2026; flight booking in AI Mode still has no public launch date.
• 68% of travelers surveyed still prefer to book with a travel brand over an AI platform.
• Some AI models recommend a sponsored, pricier flight over a cheaper option more than half the time, and the bias skews toward users the model perceives as wealthier.
• Payment tokenization keeps your raw card number away from the AI model, but it does not solve for wrong recommendations, price changes, or dispute handling.
• Use AI to search and compare. Finish the purchase with the airline, hotel, or travel brand directly, or with an OTA you already trust.
• Unlike ChatGPT or Gemini, TripProf never touches the booking itself. No commission riding on which flight or hotel you pick, no incentive to steer you toward the pricier option. Once your AI-assisted bookings are scattered across separate confirmation emails, it pulls them back into one itinerary.

Whatever changes next, the underlying test stays the same: would you trust this chatbot to hold your money the way you'd trust the merchant whose logo is on the receipt?

Sources

• Hospitality Net: The Agentic Booking Question No One Has Answered (August 2026)
• Gimmonix: Agentic AI Is Coming to Cut OTAs Out, or Is It (2026)
• Expedia Group Partner Central: The AI Trust Gap (April 2026)
• arXiv: Ads in AI Chatbots? An Analysis of How Large Language Models Navigate Conflicts of Interest (2026)
• Hospitality.Today: Google Just Put Agentic Hotel Booking Into Testing (August 2026)
• SEO Sherpa: Google Is Testing Direct Hotel Booking Links in AI Mode (June 2026)
• DeepArrival: Google Starts US Test of AI Hotel Booking in Search (August 2026)
• PR Newswire: Mindtrip Launches Travel's First All-in-One Agentic AI Flight Booking Experience (May 2026)
• Stripe: Introducing Our Agentic Commerce Solutions (October 2025)
• Elliott Report: The Only Travel Trend You Need to Know for 2026 (December 2025)
• TIKR: Expedia Stock Jumps 13% as AI Fears Ease (March 2026)