Agent intent

See what AI agents ask your store

Serving agents well is half the work. Knowing what they wanted is the other half: the questions behind your traffic, captured at your own gateway, shown in your Console.

What agents searched for

The search terms buying agents send your store — through your site search and through protocol endpoints (MCP, UCP, ACP). Not sessions reconstructed by guesswork: the literal queries, counted.

The searches you could not answer

Every search that came back with zero results is demand that walked away. The Console lists the exact queries so you can decide whether the gap is your catalog, your titles, or your synonyms.

Visitors sent by AI conversations

When an assistant cites your store and a human clicks through, the visit carries the platform it came from. You see which AI platforms send people — and which of those visits are followed by orders, labeled as the inference it is.

Answer rate, per platform

Of everything agents asked your store, how much got a real answer instead of a wall of HTML — as one headline number, tracked week over week, and broken down by agent platform.

Where it comes from

Captured at your gateway, not bought from a panel

Intent signals are read from traffic your store already receives: search parameters on requests, the queries agents send your protocol endpoints, and the referrer a human arrives with. Nothing is scraped, sampled, or estimated — if your store was not asked, there is nothing to show, and the Console says so.

Fail-closed screening

Free-text queries pass a PII screen at the edge before they may be recorded. A string that could identify a person — an email, a phone-shaped number, a token, a pasted link — is dropped entirely. Never partially redacted, never stored, never counted.

URLs never carry free text

Recorded URLs are filtered against a short allowlist of parameters that identify which page was asked for. Search text lives only in its own screened field, so reports group cleanly and secrets cannot ride along.

Real counts, labeled inference

Every number is a count over a stated window — nothing modeled, nothing projected. Where a metric is an inference (orders following an AI-referred visit), it says so where it renders.

Where it lands

In your Console, and in your inbox

Intent panels live on the Console insights page alongside coverage findings, each with real counts and one concrete fix. The optional weekly digest mails your stores' numbers — answer rate, unanswered searches, AI-referred visitors — so a week you don't log in is not a week you fly blind. Both are part of Pro; every new workspace starts with a Pro trial.