AI sends you higher-converting traffic. Can it actually read your product pages?
By Lindsey J
AI traffic already converts better than everything else
AI-referred traffic to US retail sites grew fast in early 2026. It was up 393 percent year over year in the first quarter, according to Adobe's own analysis of US retail sites. It was up 693 percent year over year during the November and December 2025 holiday season, the same analysis found.
The traffic is not just bigger. In March 2026, AI-referred visitors to those same US retail sites converted 42 percent better than non-AI traffic, Adobe reported. A year earlier, in March 2025, AI traffic converted 38 percent worse than non-AI traffic. Adobe called the swing a reversal. The same March 2026 data found AI-referred visitors spent 48 percent longer on site and viewed 13 percent more pages than non-AI visitors.
Retail is what Adobe measured. Any team that watches conversion by channel has reason to check whether the same pattern is starting to show up in its own numbers, and whether the traffic can find anything to convert on once it arrives.
The pages the traffic lands on are not built to be read by machines
Adobe scored how much of a page's content a machine can actually parse, using its own AI Content Visibility Checker, in the same analysis. Homepages averaged 75 percent readable. Category pages averaged 74 percent. Individual product pages, the page an AI shopping answer most needs to cite for price, availability, and specs, averaged only 66 percent. Even the best-performing homepages in the sample averaged 82.5 percent, and the lowest-performing averaged 54.2 percent.
Adobe's own description of the gap:
Retailers have thousands of SKUs, and our data shows that much of the content is currently invisible to LLMs.
Picture a kitchen appliance brand whose product page lists capacity, wattage, and dimensions only inside a comparison image. The numbers never appear as text or markup anywhere on the page. A shopper asking an AI assistant to compare three blenders never sees that product in the answer, because nothing on the page tells the assistant what the product is. The traffic the brand is counting on never arrives for that page. That is the same traffic Adobe's data says converts better than any other channel, and it is not missing because the content is wrong.
The fix is structured data, and the content already exists
Google's own structured data guidelines describe what a page needs for its content to be read as data rather than decoration. Markup needs to be accurate and match what a visitor actually sees on the page. It needs to use the most specific schema.org type and property names that apply, with enough recommended properties filled in to produce a high-quality result. It needs to stay current as those values change. A spec sheet rendered only as an image gives a crawler nothing to parse. A spec rendered only after a script runs does the same.
What most of these product pages need is rarely new content. The content already exists. What's missing is making it legible to something that is not a person.
Before the next content sprint, check
- Does the product page's markup include price, availability, and key specs in a defined format like schema.org Product, not only in the page's visible copy?
- If you ask an AI assistant to describe the product using only that page's URL, does it get the specs right?
- Is any of that spec data only inside an image, a PDF, or a widget that needs a script to render?
- Does FAQ or spec content load with the page, or does it sit behind a tab or accordion an automated reader will not open?
- When did the structured data last get checked against a validator, instead of assumed to still be correct?
A decision rule for which pages to fix first
If a product page's machine-readability score is lower than the homepage's, on Adobe's method or on any comparable page-level audit, fix the product page first. It is both the page AI-referred shoppers are most likely to need and, in Adobe's data, the page scoring worst across the study.
A test to run this week
Pick the three highest-traffic product pages. Ask an AI assistant, naming only each page's URL, to state the product's price, key specs, and whether it is in stock. Compare the answer to the page. A wrong or missing answer is not the assistant failing. It means the page is not in a format the assistant could read, the same reason it would be left out of an AI-generated answer a shopper sees instead of it. An earlier piece covers what happens once answers replace clicks.
Aero audits product and category pages for exactly this gap, and sequences the fix by where AI-referred traffic already converts best. Talk to us about an audit.
FAQ
Does this only matter for retail?
Adobe's figures above are retail-specific, and the exact multipliers may not hold in another category. The mechanism behind them, that an AI system can only cite what it can parse, is not retail-specific. Measure a sample of your own pages before assuming the same numbers apply.
Is adding schema markup the whole fix?
No. It is the floor. Markup that does not match what is actually on the page, or that goes stale once a price or spec changes, fails the same test a missing tag does. Treat structured data as information that needs the same accuracy standard as the price itself.
How does this fit with an existing SEO or crawlability audit?
It belongs in the same audit pass. An agent-readiness audit already checks whether a page can be operated by something that is not a person. Add a check for whether an AI system can parse the page's content to that same pass, rather than running it as a separate project.
