Good bot, bad bot: who is really buying your tickets?

Retail will feel agentic commerce first in volume, but ticketing will feel it first in intensity, for four reasons.

• Stock is perishable. An empty seat at 8pm tonight is worth nothing at 8.01pm. Anything that fills seats closer to the event, at the right price, is valuable immediately.
• The inventory is already structured. The inventory is already structured. Venue, screen, date, time, seat, price band. This is the clean, machine-readable data that agents handle well, far more so than a catalogue of ten thousand slightly different jumpers.
• Markets are fragmented. The same film or fixture is sold through the venue, the promoter, several resellers and an app or two. People hate hopping between them. An assistant that does it for them is an easy sell.
• Friction is high. Ticketing sees cart abandonment above 70%, driven by queues, fees, forced accounts and slow checkouts. Delegating the whole thing is attractive precisely because the current experience is tedious.

That first point, scarce and perishable stock, is also the source of ticketing's biggest complication. We come back to it below.

There are three layers, each with their own emerging standards.

Discovery. Assistants need to read your listings. Anthropic's Model Context Protocol (MCP) lets an assistant connect directly to a service's live data, such as showtimes and availability, instead of guessing from a web page. A newer proposal, WebMCP, lets a site declare what actions an agent can take on it.

Checkout. OpenAI and Stripe's Agentic Commerce Protocol and Google's Universal Commerce Protocol let a purchase complete inside the chat, without sending the customer to a separate site.

Payment. Shared payment tokens, and the agent-payment schemes from Visa and Mastercard, let an agent pay with a credential the cardholder has scoped and approved, without ever handling raw card details. In the UK and EU, strong customer authentication (PSD2) adds a further step that these flows still need to handle gracefully.

The temptation is to treat this as another sales channel to plug in. That undersells it. An agent is a different kind of customer: it does not browse, it does not respond to a banner or a hero image, and it will not be swayed by brand polish. It reads, compares and decides on the data it can see.

There are two modes worth separating. In the cooperative one, the fan stays in the loop and the assistant narrows the choices. In the delegated one, the fan hands over a brief (find good seats for the whole family on Saturday, together, under £200) and the agent simply acts. Most behaviour will start cooperative and drift toward delegated as trust builds.

There is also the question of where the agent does its shopping, because that decides who ends up owning the customer. It might work entirely inside a third-party assistant such as Claude or ChatGPT, where your venue is one option among many. It might work alongside the fan inside your own app. Or it might visit your own site on the fan's behalf. The further the process moves into the assistant, the less of the relationship stays with you. The assistant sees the fan's preferences, holds the data and completes the sale, and you simply supply the seats. That is what we mean when we say a brand risks becoming a dataset.

Everything so far assumes the agent is welcome. In ticketing, that is far from a given.

Here is the part that makes ticketing different from every other sector. Limited, time-boxed supply is exactly what made it worth scalping, and exactly why the industry spent a decade building defences against software: CAPTCHAs, randomised queues, device fingerprinting, verified-fan schemes. To be clear, this is mainly a sport, music and theatre problem. Cinema has little of it, because there is rarely a shortage of seats worth reselling. But any ticketing business that spans formats has to design for it.

Now a fan's legitimate assistant turns up at the front door. To a bot detector it looks like a bot: automated, fast, not a person at a browser. Block everything and you shut out your best new customers. Wave everything through and the touts get the keys to the kingdom. So, how do ticket vendors need to start thinking to deal with this problem?

1. Verify the agent, not just the human

Visa's Trusted Agent Protocol and Mastercard's Agent Pay both aim to let a merchant recognise a legitimate agent and tie it back to a real, verified cardholder. The wider idea is sometimes called Know Your Agent: the equivalent of KYC for the software acting on someone's behalf. If every purchase traces to one accountable person, running a hundred accounts through a bot farm stops working.

2. Limit what the agent is allowed to do

Tokenised payment credentials can carry rules, such as a spend cap or a merchant restriction. Today these controls are broad, what ticketing needs is more granular control e.g. this agent may buy two tickets, for this event, in this price tier, on these dates. That granularity is not yet standard, but it will come.

3. Change when the ticket becomes real

Some of the most effective measures sit after the sale. Releasing barcodes 24 to 48 hours before the event, or restricting transfers, removes most of the profit from holding stock. That works whoever or whatever bought it.

4. Watch behaviour, not identity

Purchase velocity, repeat patterns and unusual combinations are strong signals, especially when measured against a single verified identity. One person buying two tickets is normal. One identity buying forty across fifteen events is not, however the request arrived.

No single measure solves it, and none should be treated as final. But the shift in mindset matters. For a decade the job was keeping automation out. The job now is deciding which automation to let in, and on what terms. The good bot and the bad bot will look identical at the front door for some time yet, so the businesses that win will be the ones that build ways to tell them apart, rather than shutting the door on both.

Tickets are one of the first real tests of agentic commerce. An agent is a customer, not a channel, and it decides on data, not brand. So the question is whether your business is ready to be chosen? That means keeping out the bad bots without shutting out the good ones, because the good ones are your next customers.

Five things to check now:

Findable. Can an assistant see your events, times, prices and availability in a clean, current, structured form?

Trustable. Can you let the fan's agent in and keep the tout's bot out? The old defences block both, so look for verified, scoped credentials that trace back to one real customer, not just a spend cap.

Transactable. Can a purchase complete, member pricing included, without a human stepping in at the last moment?

Answerable. When an assistant asks something unusual (accessibility, age rating, refunds), is there a machine-readable answer?

Continuous. If plans change, can the agent amend, transfer or cancel, and can you see what it did on the fan's behalf?

Fans will not announce the switch. One day a share of your bookings will simply arrive without a visit to your site. Make sure the good bots get through, and the bad ones don't.