The evidence, dated, sourced, and not overstated.
Forecasts are labelled as forecasts. Every number links to its source and says what it does not mean. Last verified 25 Sep 2026.
Agents are arriving
Doesn’t say: AI-referred traffic (people who clicked from an AI answer), not agents completing purchases.
Doesn’t say: Growth from a small base.
Doesn’t say: A forecast of traffic, not orders.
Doesn’t say: Share of observed agent activity across HUMAN’s customers.
Doesn’t say: Share of observed agentic traffic, not bookings.
They often fail
Doesn’t say: One retailer, one assistant, one period. It shows the gap can be large, not its size for you.
Doesn’t say: Six sites and six unnamed models; a strong signal, not a universal rate.
Doesn’t say: Readiness as assessed in that study’s method.
Doesn’t say: Measures requests to the file, not its SEO value.
The same barriers carry legal risk
Doesn’t say: Projection for 2026 from midyear filings.
Doesn’t say: Applies to US state and local governments.
Doesn’t say: Enforcement varies by country.
“93% of merchants want AI providers to bear the loss” circulates widely but we could only find it in secondary sources, so we do not use it.
See where agents fail on your funnel.
Run one free cell on a property you own. No sales call, no card.