Task gain

People are starting to hand assistants the whole errand. Book the flight, fill in the form, buy the thing. When they do, the customer never reads an answer or sees a citation, and the brand that wins is the one the agent can actually get the job done with.

I think task completion rate becomes the number that matters for AI search, the way rankings and then clicks did before it. It’s the first visibility measure I know of that ends in a transaction, and almost no brand can see its own. So we built a way to measure it, and ran the first round.

Sydney to Melbourne and back, one adult, one checked bag, the cheapest total fare, taken to the payment screen. The same prompt, pasted word for word into every assistant and agent we could sign into, up to three times each, on 3 and 4 November 2026.

No agent was given a card, a password or a one-time code, so nothing was booked. Reaching the payment screen with the right flights in the cart counts as done.

Jetstar had the cheapest fare, A$355 with a bag, and 12 runs got there. Its site also errored, blocked or challenged agents in at least 19 runs across nine products, and nine of Virgin’s ten wins came in a run where Jetstar’s site had just failed that agent.

None of that would show up in Jetstar’s analytics. The agent never becomes a session.

This is the finding we didn’t go looking for. In pass 1 every completing agent stopped at the passenger form. By pass 3 two of them didn’t, because they’d met the tester before, and they scored the only two 5s in the study.

The caveat is a big one: the memory came from earlier runs of this same study. A customer’s first attempt won’t look like that. I suspect their tenth will.

We went in expecting logins and CAPTCHAs. Logins never happened, and 30 of the 40 interventions were the passenger form: title, name as on ID, date of birth, mobile, postcode.

In four runs Jetstar’s session timed out while the agent waited for a person to answer. The gap between an agent asking and a human answering is now a failure mode with a brand attached.

Measure it first. The rubric is published, and three runs per product is enough to start. I think the first result will tell you more than a year of citation tracking.

Then treat agent experience as a discipline beside user experience: price visible without a login, forms an accessibility tree can operate, sessions that outlast a human answering a question, and a bot policy that can tell a customer’s agent from a scraper. A stall that survives two runs on two days, in the same place, is yours to fix.