Social → AI → Booking

New UK research from Vision One Research. In its Travel Brand Barometer survey of 3,000 consumers, 42% of Gen Z said TikTok influences their holiday decisions, versus 37% for Google and 33% for Instagram.

Then AI picks up the next bit of the journey. 51% of Gen Z have used tools such as ChatGPT to help plan a holiday. 42% would trust AI with booking advice, 42% with restaurant recommendations and 27% with flight selection.

So TikTok is beating Google for Gen Z travel inspiration and AI is getting closer to the booking. But this travel discovery is not moving neatly from Google to AI. It is fragmenting.

A traveller sees a hotel, restaurant & experience on TikTok. AI then helps to turn that inspiration into a curated shortlist and sorts out the logistics on apps like reely.travel 

An established platform may still take the booking. Uber, Jet2holidays, Premier Inn, Airbnb, Trainline all showed meaningful usage among younger travellers in the same research.

On something like Reely you’ll get a reverse auction happening across the whole basket instead of the usual manual price shopping from item to item before booking. This is the AI agent doing its work. As we’ll see in the stories below, payment is now also mostly solved.

So the job for travel brands is no longer simply “rank on Google” (and it hasn't been for a while TBH - if that is still your game you might be going the way of the farrier). The job now is to be talked about in the places where inspiration happens, described clearly enough for AI to understand and recommend you, and easy to buy when the traveller finally moves.

That is three different visibility problems, with three different bits of plumbing behind them.

I suspect plenty of travel marketing plans still treat social, AI discovery and conversion as separate channels owned by separate teams. The traveller absolutely does not care about your org chart.

If you run a hotel, DMO, tour or travel brand, can someone move from a TikTok about you to an AI recommendation to a clean booking path without losing the thread? What are you doing to keep them connected?

This content is provided by the newsletter sponsor reely.travel - the AI layer between social and booking

Your data problem is about to become a much bigger problem

Travel still has a data problem. Which is annoying, because we have all been told to “get our data in order” since the first person in a conference keynote said “big data” with a straight face.

But this time it is not about dashboards, reporting, or some theoretical future personalisation project. It is much more basic than that as Travis Pittman explains:

If an AI asks your business: “Do you have anything in Peru in September, under €3,000, with availability?” can your systems answer properly?

Live price. Live availability. Clean catalogue data. In under a second. Without guessing from a web page, waiting for a human, or stitching together three slightly different versions of the truth.

Adding a MCP server onto a mess will just lead to a bigger mess. But having your back room open and clean opens up opportunities that were never there before.

I particularly like the ability to package things on the fly from unrelated parties to build new things (that people are going to buy anyway).

Meat Proxy

Thanks to Claire Jones for introducing me to the term Meat Proxy!

“Meat proxy” is a horrible phrase. Annoyingly useful though. It describes someone who takes a question, asks AI, then forwards the answer on without really reading it, checking it, or adding anything of their own.

Using AI is not the issue. Using it well is the issue. Read the answer. Check the claims. Fix the tone. Add your judgement. Remove the bits that sound clever but are actually doing nothing.

Otherwise you are not using AI as an assistant.

You are just adding latency.

Is Jetstar getting ready for A2A?

Cory Garner has been circling something this week that I think travel people should pay attention to. It’s on the hot subject of personal agents (Instinct; Muse). And its not because every new AI assistant is brilliant. They are not. Some are genuinely impressive, some are a bit terrifying, and some are probably both.

The thread through Cory’s posts here, here and here on Instinct, Meta flight search and Muse/Duffel is that AI agents are moving from “tell me about travel” to “do travel things for me”. Search flights. Compare options. Book. Manage changes. Sit across email, calendar, payments and the messy personal context that makes travel hard.

That is where this gets interesting. Also where it gets uncomfortable because Instinct shows the trust problem. If an assistant can read your inbox, see your screen, pull codes, send emails and make bookings, then privacy and reversibility are not side issues. They are the product. One wrong action and the magic trick stops feeling magical very quickly.

There is another story of someone getting their Resy account banned because their Instinct agent hammered their API like a bot - because, well, its a bot.

The Muse/Duffel example shows the other side. If Meta can put a personal agent in front of millions of users, and Duffel can give it live flight inventory, booking and servicing infrastructure, then AI is not just another search surface. It is a potential new distribution layer. That bit is exciting.

It is much more basic: can a machine safely buy from you? Can it understand your product, get a live price, check availability, make the booking, change it later, and leave a trail everyone trusts? Because that is the line. Discovery is useful but booking is where the money and the risk turn up.

I still think we are early. The user behaviour is not settled, the trust model is not settled, and the commercial model is definitely not settled. But the direction of travel is pretty obvious now. AI agents are going to sit between travellers and suppliers. The only real question is whether your business is readable, bookable and trustworthy enough to be in that conversation.

I hear you quietly ask yourself - but what about payments…. read on to the next story!

See also: TechCrunch on Instinct and Duffel on Muse.

Payments are solved so let’s move on to the next AI problem

Agentic travel booking has had one very obvious missing piece for a while. Payments.

Buried into the longform version of the Resy story I mentioned above from Mobi’s Andrew Boch is something I think I might have missed up until now. And maybe you had also.

The most recent awkward bit has been: the agent found the thing, now who actually pays? That is starting to look much more solved than I expected. As Boch explains:

“Now the good news: half of this functionality already works well: the money. Muse and Instinct both settle through Stripe's Link, so the user approves what the agent spends and the agent never holds the card.”

This short video is worth watching, and Stripe’s agentic commerce explanation is the useful plain-English version. The key idea is that the agent does not need your raw card details. It can use a scoped payment token: limited to a merchant, amount, transaction, and time window.

That is a big deal. You, the human, are kept in the loop with push notification which you have to action to authorise payments.

It does not solve every travel problem. Far from it. Travel still has changes, cancellations, chargebacks, refunds, schedule shifts, supplier rules, support, fraud, duty of care and all the other fun things that make our industry such a relaxing place to work. But payments no longer look like the blocker.

Agentic booking is not waiting for one magic AI breakthrough anymore. The payment rails are forming. The distribution surfaces are forming. The user behaviour is early, but not imaginary.

So the question gets a bit sharper. When an AI agent is ready to buy, can it buy from you?

This week there might have also been an answer to that one too………. read on!

If you think someone (or everyone) you know or work with could grow from being more informed on the topic of ai + travel (or could use the training above) then please forward this email to them and they can click the button below:

Claude takes you to the transaction’s front door - wherever that may be.

Linas Beliūnas has a useful breakdown of Anthropic’s new Claude Commerce Agents here: Claude Commerce Agents, and the shorter LinkedIn version is here: his post.

The bit I think matters is what Claude has done, and what it very deliberately has not done.

This is not Anthropic trying to become the new shopping mall. It is not Claude saying “buy everything through me and I will own the customer relationship”. That is much closer to the OpenAI / Google / Perplexity / Amazon direction. Remember OpenAI shut down Instant Book after just 5 months. That is probably instructive (or what it just the payments trust issue above??).

Anthropic seems to be making a different bet: put the agent inside the merchant.

That is interesting for travel because it keeps the catalogue, checkout, payment, fulfilment, customer data and responsibility with the actual business. The AI helps the customer or staff member navigate the messy commercial intent layer: search the catalogue, compare options, answer questions, maybe prepare the basket, maybe support the booking flow. (For reely.travel we are super excited by not having to custom build this!!!)

But, and this is the important bit, it does not complete the transaction.

Linas points out that the reference implementation does not place the order or charge the card. The model cannot just go rogue and buy something. There are caps, approval gates, staged actions, server-issued IDs, and all the boring-but-essential controls that make commerce less terrifying. So this is a transaction story, but not quite a payments story.

Claude is getting closer to the moment of purchase. It is helping structure the path to purchase. It is showing merchants how an AI commerce agent could sit inside their own property rather than handing the customer to someone else’s AI interface.

Why this is probably more exciting than it first looks on the surface is that this approach by Anthropic caters for every possible instance. Got an MCP - great! A2A - huzzah! Still working through WhatsApp, you’re still included too! For our industry that is pivotal.

As a step to get the customer through to the best individual supplier of a customer’s desires and not just to those who can afford to build the rails - it’s super exciting.

Slack Group!

Podcasts and Sponsors

Podcasts now on Spotify and Apple Podcasts:

All podcasts are now showing up on Spotify and Apple Podcasts for your easy listening pleasure! Now is a great time to binge those you may have missed.

The highly anticipate episode with Vadym Klimov from WinWin Travel is now out. Hear the story of losing everything due to a war to restarting on the AI path.

You can hear the episode here

This could be your brand’s newsletter (and podcast).

We are currently seeking submissions from leading AI companies in travel who understand that all businesses should be media businesses.

Following the example of dojo.live and Encora we offer the opportunities to communicate your brand’s USP’s to a B2B travel audience from the leading Travel + AI resource on a bi-weekly, done-for-you basis for all of 2027 (and beyond).

Contact Tony if interested to learn more and see if we are a good fit for one another.

Otherwise:

Partner with Us - sponsorship info for the newsletter and podcast can be found here.

Most clicked last week was the link to the the Phocuswire story from WENRIX on why humans shouldn’t be in every loop!

That’s it - you’ve made it to the end of this edition. If you’re thankful for this newsletter - you can always buy me a coffee.

I’ll be putting the result of the most clicked post in next week’s edition so you can see where others are focusing. If I’ve missed something, you’ve got a tip or any feedback at all - you can simply reply to this email and it will come straight to me. I’m doing this for You so please don’t be shy to tell me what you think