The store that blocks its best customer

For twenty years, e-commerce security had one rule: humans in, robots out. CAPTCHAs, bot walls, device fingerprinting, velocity checks — an entire industry exists to keep automation away from your checkout. And it was right. The bots were scalpers, scrapers and card-testers. Blocking them protected your margin.

Then the definition of “robot” changed underneath us.

The new automated buyer

When someone tells ChatGPT “order me the same running shoes in a 42,” the thing that arrives at a store is, technically, a bot. It’s also a paying customer — arguably the best kind of paying customer. It doesn’t browse forty tabs and abandon the cart. It doesn’t get distracted at the shipping form. It arrives with intent, a size, and a payment method. Conversion rate: roughly the whole thing, if you let it through.

Almost nobody lets it through.

The same wall that blocks a scalper blocks this buyer, because the wall can’t tell the difference. A CAPTCHA doesn’t ask “are you malicious?” — it asks “are you human?” Those used to be the same question. They aren’t anymore.

Who solves this, and how

The fix isn’t lowering the wall. Card-testers didn’t go anywhere; if anything, opening your checkout to all automation is the worst idea in payments this decade.

The fix is identity. The major AI operators now sign their agents’ purchase requests — cryptographic proof that a request comes from a real assistant acting for a real customer, under protocols like OpenAI and Stripe’s ACP and Google’s AP2. A store that can verify those signatures can do something new: keep the wall exactly where it is and cut a door in it. Verified agents in. Everything else stays blocked, same as today.

In the US, this already works. OpenAI’s Instant Checkout lets ChatGPT complete purchases at participating merchants. The infrastructure exists; the shift is not hypothetical.

The uncomfortable math

Here’s the part that matters if you run a store: when an agent can’t buy from you, it doesn’t try harder. It buys the second-best match from whoever can take its order. The customer never knows you existed — there was no browsing session, no retargeting pixel, no abandoned cart email. The sale doesn’t show up as lost. It shows up as nothing.

That’s what makes this shift quieter and more dangerous than the last ones. When mobile commerce arrived, your analytics screamed about bounce rates on your desktop site. When agents arrive, your analytics will show… nothing at all. The orders simply happen elsewhere.

Whatever share of shopping ends up flowing through assistants — and forecasts vary wildly, so we won’t pretend to know — the merchants who can accept those orders will take all of it before the merchants who can’t see a single data point about what they’re missing.

Being ready to sell to agents when the channel arrives isn’t a growth hack. It’s the same decision as having a website in 1999 or a mobile checkout in 2015 — made earlier this time, because now we can see it coming.