Your next customer is a bot with a budget. Is your website ready?

For two years, "AI agents will shop for people" was a conference-keynote prediction. Then the traffic data arrived. Adobe Analytics — covering over a trillion visits to US retail sites — reported AI-referred traffic up 393% year over year in Q1 2026, on top of holiday-season growth of nearly 700%. Since Adobe started tracking in late 2024, AI-referred retail traffic has grown more than fourteen-fold.

The volume story is striking. The quality story is the one that should reorganise your roadmap:

Visitors arriving from AI sources also engage more — longer sessions, more pages, higher revenue per visit. Which makes sense: by the time an assistant sends a human (or itself) to your product page, the comparison shopping already happened inside the model.

The catch: most websites are hostile to their best new traffic source

Adobe's follow-up finding is the uncomfortable one: retail sites are lagging badly on machine-readability — the structured data, coherent page semantics and unambiguous flows that AI agents need to parse a site reliably. Every team has war stories of the human version: the variant picker nobody understands, the shipping cost that appears at checkout step three, the "add to cart" button that's actually a mailing-list trap. Humans muddle through those. An agent acting on a customer's instruction — "find me a coffee machine under $150 and buy it" — hits the same friction and does something worse than complain: it leaves silently and buys from a site it can parse.

Agentic traffic doesn't file complaints. It just completes the task somewhere else.

How do you test a website against a customer who's an AI?

You can't recruit AI agents to a usability lab. But you can simulate them — with the demographic and psychological grounding that makes the simulation informative rather than generic. That's what Lanice AI's UX Simulator does: demographically-grounded AI personas drive a real Chromium browser on your live site, with a task written in the customer's terms and a research objective the persona never sees. Every step is screenshotted and explained twice — the mechanical trigger ("the search box was the only visible entry point") and the psychological driver ("she distrusts sponsored results and scrolled past them"), tagged against a controlled vocabulary: price_sensitivity, social_proof, impatience, decision_fatigue.

Session analytics then aggregate the journeys: task-completion funnels, the exact page where journeys died, the drivers behind each abandonment, and persona-vs-persona comparisons — the budget-conscious parent versus the impatient mobile-first shopper versus the low-digital-confidence senior. The output doubles as your agent-readiness audit: where a psychologically-grounded persona stalls is exactly where a shopping agent will stall, because both are language models parsing your DOM and your copy.

An agentic-readiness checklist to start with

• Reveal costs early. Surprise shipping at the last step is the single most common journey-killer for price-sensitive personas — and an instant abandonment for agents comparing total price.
• Make state changes legible. If adding to cart doesn't produce an unambiguous confirmation in the page, an agent can't tell whether it worked.
• Kill ambiguous controls. Variant pickers, hover-only menus and modal traps stall agents the way they confuse low-confidence human users — audit them first.
• Ship structured data. Product schema, availability and pricing markup are how agents confirm what a page is — Adobe's data shows most retail sites still don't.
• Rehearse, then re-rehearse. Run persona journeys before and after each fix, the way you'd A/B test creative. Treat agent-readiness as a metric, not a project.

The brands that treated mobile as a curiosity in 2010 spent 2015 doing emergency responsive redesigns. Agentic commerce is compressing the same arc into a couple of years — and this time, the visitors that can't use your site are the highest-converting traffic you have.