The Front Door

Most sites aren't blocking agents on purpose.

18 of the 23 sites we watch publish nothing at all about AI agents. 2 say agents are welcome. None say keep out.

So whatever an agent runs into on your site, there's a good chance nobody chose it. It's a bot rule someone wrote years ago to keep scrapers out.

What to do. Two questions worth putting to your team. Are we turning agents away? Did anyone mean to?

More sites turned an agent away than let one in.

9 of 23 are open to every kind of agent. 6 are closed to all of them.

5 let a real browser in but turned away anything that fetched the page without running it. That's the crawler behind AI search answers getting the door, while agent mode walks in.

3 returned a bot challenge that lets some sessions through. We haven't called those yet.

What to do. If you're in that group, decide whether you want AI search to see your catalog. Right now it can't.

Resale marketplaces are the hardest door to get through. Real-estate marketplaces are the easiest.

Agents let in, by vertical. Each bar is scaled to that vertical's own count.

Real-estate marketplaces let 4 of 6 in. Resale marketplaces let 1 of 6 in.

12 of these 23 readings are the first we've taken. We'll check them again next week before we call them settled.

What to do. Look at your own vertical. If you're in resale marketplaces and never made a call on agents, you're probably blocking them. Worth measuring before the September 15 default change.

Agents rarely finish the job, and they quit in different places.

This one covers home-services sites only.

Of 44 agent attempts across 4 home-services sites over 3 cycles this week, 2 reached the point of submitting a lead.

a national home-services marketplace: 9 of 11 stopped at picking a service (1 at the lead form).

a caregiver marketplace: 5 of 11 stopped at the lead form (3 at picking a service).

a task and handyman marketplace: 7 of 11 stopped at the multi-step quote form (2 at the lead form).

a national handyman marketplace: 2 submitted a lead; 8 of 11 stopped at the lead form (1 at the address step).

3 cycles, 11 agents per site. We can see where a run stopped. We can't yet tell whether the site stopped it or the agent gave up on its own.

What to do. If your funnel has a multi-step form, that's where agents quit. Might be worth checking whether people quit there too.

We got prices back. We can't prove where any of them came from.

This one covers grocery sites only.

Across 10 grocery sites, agents came back with a price and stock status 40 times out of 100. The best site answered 10 of 10. 3 gave us nothing at all.

Then we went back through those 40 answers to see which store each price had been read from. 0 of 40 could say.

That's a limit on what we can prove, not evidence the prices were wrong. We're flagging it because it changes what the 40 is worth.

What to do. If your prices move by location, look at what an agent sees before it's picked a store. That's the number a shopper ends up with.