Agents Are Already Shopping for Your Customers. Can They Even Read Your Product Page?

Here's an uncomfortable number for a Monday: AI and agents touched 20% of all retail sales last holiday season — around $262 billion — according to the company whose CRM most retailers run on. So let me ask the question nobody raised in your last planning meeting. If a shopping agent landed on your product page right now, could it even read the price? For most retail sites, the honest answer is no. The buyers changed. The websites didn't. That's the gap, and it's wider than your dashboards are letting on.

This post pulls from open-web research done over the last few days: retail traffic reports from the analytics arm of a large design-software company, holiday sales data from a major CRM vendor, 2030 forecasts from an investment bank and a research firm, the published specifications for the new agent-commerce protocols, and an independent test of what crawlers can actually see on live product pages. Where the sources disagree, I've said so. Where a number comes from someone with a commercial reason to inflate it, I've flagged that too.

Is AI shopping traffic still too small to bother with?

The standard objection goes like this: AI referrals are a rounding error, so why re-plumb your site for them? It's not a straw man. Even the analytics firm reporting the fastest growth concedes AI traffic is still a small slice of total visits — by some estimates well under 1% of retail sessions. If you only watch volume, the skeptics have a point today.

But volume is the wrong number to watch, for two reasons. First, the trajectory. AI-driven visits to US retail sites were up 693% year over year across November and December 2025, then up 393% in the first quarter of 2026, and the same analytics team forecasts roughly 130% growth again for the 2026 holiday season. Put that next to the other channels in the same forecast — social media at 17%, paid search at 4% — and AI isn't just growing, it's growing about eight times faster than anything else you're already spending on.

Second, the quality. By March 2026 that same AI traffic was converting 42% better than non-AI traffic — a record high, and a reversal from a year earlier when it converted about 38% worse. Revenue per visit ran roughly 37% above other sources. A visitor sent by an AI assistant has usually already been pre-qualified by the assistant, so it arrives closer to the buy. Stack that on the CRM vendor's finding that AI and agents already influenced $262 billion of holiday spend, and "too small to matter" starts to sound like the thing people said about mobile traffic in 2009. Small and growing 400% a year is not small for long.

Can an agent actually complete a purchase on your site today?

Short answer: the plumbing shipped in September 2025, and your site is probably not plugged into it. Within a few weeks of each other last autumn, three agent-commerce protocols landed. A large search company published the Agent Payments Protocol (AP2), built around cryptographically signed "mandates" so an agent can prove a human authorised a purchase; it launched with 60-plus partners including card networks and processors. The Agentic Commerce Protocol (ACP), backed by a major AI-assistant maker and the payments firm Stripe, focused on checkout. And a universal commerce protocol for catalog and product data was co-developed by that same search company with Shopify and several large retailers.

Why all at once? Because, as the analyst firm Everest Group put it in its write-up of AP2, today's payment rails were built on the assumption that there's a human at the checkout — and that assumption breaks the moment an agent is holding the card. Everest's read is that AP2 has real momentum but its success "will depend on how quickly enterprises, merchants, and platforms embed it into real-world systems." Translation: the standard exists; adoption is the open question, and it's on the merchant side.

The size of that opportunity is why the analysts are paying attention. Investment-bank research estimates agent-driven US e-commerce could reach between $190 billion (base case) and $385 billion (bull case) by 2030 — 10% to 20% of all US e-commerce. The research firm Gartner, meanwhile, forecasts that AI agents will be involved in around 90% of B2B buying by 2028, routing more than $15 trillion through agent-mediated exchanges. Different scopes, wildly different numbers, same direction of travel.

Is your site even legible to the machines doing the buying?

This is the part that should sting. The analytics firm's April 2026 benchmark scored retail pages on machine readability, and product pages — the exact pages where money changes hands — came in lowest at 66%. Homepages scored 75%, category pages 74%. The best retailers hit 82.5%; the weakest sat at 54.2%. So the page an agent most needs to parse is the one your site renders worst for machines.

The cause is boringly technical and entirely fixable: most AI crawlers fetch raw HTML and don't execute JavaScript. If your price and stock status are injected client-side after the page loads — which is how a lot of modern storefronts work — a human sees them and a crawler sees a blank. An independent August 2026 test of four large retailers' product pages backs this up: between 49% and 89% of the visible text blocks were missing from the markup a crawler can read, and on one big-box retailer the product's price wasn't in the readable markup at all. Your merchandising team would never ship a page where shoppers couldn't see the price. You've shipped one where the machines doing a fifth of the buying can't.

Structured data — valid Product JSON-LD with price, availability and ratings — helps a lot here, because it hands the machine an explicit, unambiguous version of the facts. It isn't a complete fix, since no protocol guarantees a given AI system reads it. But it's the single highest-leverage thing most sites haven't done, and it costs a developer a day, not a quarter.

Can you afford to wait for the standards to settle?

The seductive counter-move is to wait. Let AP2, ACP, the catalog protocol and the various agent-to-agent standards fight it out, then back the winner. The problem is that "wait and see" is itself a bet, and the odds are bad.

Consider the honest caveats. One infrastructure analysis notes that AP2 launched with 60-plus partners but, months later, still had no shipping consumer product behind it — momentum on paper, not yet in a checkout flow. Gartner expects that even as agent volumes explode, fewer than 40% of sellers will say agents actually improved their productivity by 2028. So nobody should pretend this is a finished, obvious play. It's early, it's messy, and the standards genuinely might consolidate in a way that strands early integration work.

But here's the thing the "wait" camp misses: the expensive, protocol-specific work is not the work that matters first. Whether AP2 or ACP or something not yet named wins, every one of them needs the same foundation — a site whose prices, inventory and product facts are present in machine-readable markup. Fix that, and you're ready for all of them. Skip it, and no protocol can rescue a product page that renders blank to a crawler. The readability work is the one bet that pays off no matter which standard wins, which is exactly why it should come first and why it's indefensible that most retail product pages still fail it.

So what should you actually do on Monday?

Not a roadmap. Five concrete things, in order.

1. Open your own best-selling product page the way a crawler does. curl it, or view the raw page source with JavaScript disabled. If the price and stock status aren't in that raw HTML, you've just found your single most important fix — the one from the readability section.

2. Add and validate Product JSON-LD on product and category pages, with price, availability and ratings. It's a day of developer time and it's the cheapest leverage you have against the 66% problem.

3. Start measuring AI referral traffic as its own channel. Most analytics setups bucket it as "direct" or miss it entirely, which is precisely why it feels too small to matter. You can't manage the 400%-a-year channel you can't see.

4. Pick one protocol lane and pilot it. A clean, agent-readable product feed or endpoint is a low-regret move that most of the emerging standards can consume. You're not marrying a protocol; you're learning the shape of the problem.

5. Stop waiting for one standard to win. The foundation work — readable markup, structured data, measurable AI traffic — is identical across all of them. Do that now; decide on protocols later.

The agents are already shopping on your customers' behalf, and they're converting better than your paid search. The only real question left is whether, when one of them reaches your product page, it can read it. Right now, for most of you, it can't.