A store usually already runs several tools that record some part of AI agent activity. The useful move is to know which question each one answers, and where the gaps are for your store.
| Question | Where to look | What it measures | Read it knowing |
|---|---|---|---|
| How many people clicked through from AI assistants? | Web analytics | Sessions with an AI assistant as the referrer; in GA4, the AI Assistant channel | Visits without a referrer are filed as Direct, whatever sent them; clicks from Google's AI Overviews and AI Mode count as Organic Search |
| How often do my links appear in Google's AI answers? | Search Console, Generative AI performance report | Impressions in AI Overviews and AI Mode | Impressions, not clicks; low-volume sites may see no report |
| Which AI channel sold what? | Your commerce platform | Orders by sales channel, for example Shopify's Agentic Storefronts | Covers orders placed through the platform's own AI channel integrations; without one, look at orders by landing-page referrer or UTM source |
| Which automated clients reached my site, and what did my rules do? | CDN or bot management | Requests by client, verified status, allow or block decisions | Check retention, and how verified bots and agents are grouped. On a hosted platform such as Shopify you don't run the CDN; ask your platform or bot app what it reports |
| Is this agent order legitimate? | Fraud and risk provider | Agent-placed orders and the decision on each | Coverage before checkout varies by provider; ask whether it labels agent-placed orders and how many it declined |
| How do AI answers describe my brand? | AI visibility tools | Answers to a set of selected prompts | A sample of answers to selected prompts; pair it with your traffic and orders |
| How far did each agent get toward cart and checkout, with every figure labeled by how it's known? | Merchant-side evidence layer | Planned: how far each agent got, traffic that couldn't be identified, and where agents dropped off | Being built, not live. Designed to label every figure observed, verified, claimed, modeled or unknown |
How to use the map
Pick the question you need answered first, then open the tool in that row. If two rows disagree, that's information: a client that doesn't run your analytics tag can still appear as a request in your CDN logs.
Where Cartograph.ai™ fits
The last row is what we're building. Cartograph.ai is being built to show merchants how far AI shopping agents get toward cart and checkout on their own storefront, with every figure labeled by how it's known. It is designed to check identity claims against what each operator publishes and to show anything it can't confirm as a claim. It doesn't block bots, sell ads or tune content for AI search. Register interest.
That's the first set of Navigations: site, access, discovery, monitoring. All five are on the Navigations page.