(AI) Agents don’t browse
The question of whether AI agents would become a meaningful purchase channel stopped being theoretical sometime last year. Adobe’s data shows AI traffic to retail sites climbed 269% year-over-year this March. Microsoft Copilot Checkout is live. Google’s AI Mode is rolling out to select retailers. The agent as front door isn’t a prediction anymore.
But most retail sites were built for humans. Agents work very differently.
Agentic shopping is merit-based. Not search-engine merit, where ad spend and SEO optimisation dominate the top positions. Agent merit – where the brand with the clearest, most complete, most accessible product data is the one the agent recommends. For retailers with good products and thin marketing budgets, that’s a different kind of playing field.
What an agent actually sees
When someone visits your store, they see the photography, the hero banner, the carefully crafted product copy, the social proof. That whole visual layer is designed to convert a browsing person.
An agent queries your product data. It’s looking for a title, a price, material, dimensions, availability, a return policy. Your photography, your UX, your hover animations – irrelevant. What matters is whether your schema markup confirms what you’re selling.
If your product title is aspirational (“The Adventure Pack”) rather than specific (“40L waterproof hiking pack, laptop compartment, navy”), the agent isn’t sure what it’s looking at. If your return policy lives in a JavaScript accordion that crawlers can’t open, the agent doesn’t know your returns are free. If your product information is embedded in a display template rather than structured data fields, the agent moves on.
Not to a worse brand. To a clearer one.
The data quality dividend
The brands winning agent-driven discovery right now aren’t necessarily the biggest. They’re the ones whose product data is complete, specific, and accessible. Clean product fields. Honest specifications. Policies in plain text. Customer reviews and Q&As on product pages that agents use to assess whether something genuinely fits what a shopper asked for.
Cloudflare released a free tool this month – isitagentready.com – that scans any site and scores it across five categories: discoverability, content accessibility, bot access control, protocol discovery, and commerce. Thirty seconds. Most retail sites score poorly, not because retailers are fundamentally behind, but because most of the infrastructure was designed before agents existed.
If you’re already in the Adobe Commerce world, it’s worth knowing that Adobe LLM Optimizer exists – it’s part of the ecosystem and included in most existing licences. Worth a look before you go hunting for third-party tools.
A third of ecommerce businesses haven’t started addressing any of this. Forty percent are mid-way through it.
Where to start
The foundation is structural. A valid robots.txt with AI bot rules. A sitemap. An llms.txt file – think of it as robots.txt for the AI era, a lightweight signal that tells language models what content exists on your site and how to find it. Schema markup on product pages so agents can confirm title, price, availability, and specifications without having to interpret your layout.
Then the information layer. Return policies, shipping terms, FAQs – in plain text pages agents can actually reach, not buried in dynamic templates or accordion menus. Specific, factual product descriptions rather than marketing language agents can’t parse.
For retailers ready to go further, the Universal Commerce Protocol (UCP) – backed by Google and major global retailers – is the emerging open standard for agent-driven transactions.
Building foundations that serve every surface you build next is exactly what this requires. Retailers who sort their data now benefit from agent discovery, from search, from better internal analytics, from cleaner integrations. Good data compounds.
Scan your site at isitagentready.com. Fix the structural things first.
The timing question
Morgan Stanley estimates nearly half of online shoppers will use AI agents by 2030, with agents accounting for roughly a quarter of their spending. Whether those projections land precisely or not matters less than the direction: this is becoming a real channel, fast, and most roadmaps haven’t accounted for it.
There’s a pattern in how retailers tend to respond to shifts like this. Some wait until the channel is undeniable, then scramble to catch up. It’s the same instinct that produces last-minute EU AI Act compliance – we covered that here last week – where the urgency of a deadline does the work that curiosity should have done months earlier.
The brands that ride waves like this aren’t usually the ones who moved fastest once it was obvious. They’re the ones who moved early enough to learn properly – to make mistakes when the stakes were low, to refine their approach before competitors had even started thinking about it. That’s how you generate revenue nobody else is looking at yet.