U.S. Retail Sites Average 61% Machine Readability, Leaving Nearly 40% of Homepages Invisible to AI Agents
Adobe's July 2026 data puts average homepage machine readability at 61%. What the score actually measures, how it breaks down by vertical, why product pages are the real weak spot, and what e-commerce teams should fix first.
Key Takeaways
- When Adobe widened its sample in July 2026, the average machine readability of U.S. retail homepages came in at 61%, meaning roughly 40% of homepage content is not fully readable by shopping agents
- AI-sourced traffic rose 62% year over year in July and has converted better than non-AI traffic for 11 straight months. The entry point keeps widening while the sites receiving that traffic remain hard for machines to read
- The weak spot is the product page rather than the homepage. Checking whether price, stock, specs, and reviews appear in the pre-JavaScript HTML, then fixing it with server-side rendering, is the first practical move
The entry point is growing while the destination lags
That AI-sourced traffic keeps growing is no longer news. What stands out in Adobe's latest data is what happens after the agent arrives: it cannot read about 40% of what is on the page.
Artificial intelligence continues to drive higher amounts of online retail visits, but e-commerce sites are frequently not optimized for it.
chainstoreage.comStart with the traffic side. According to Adobe data reported by Chain Store Age, AI traffic to U.S. retail sites in July 2026 was up 62% year over year, and up 1,219% cumulatively since measurement began in October 2024. That traffic converts 60% better than non-AI traffic, the eleventh consecutive month in which AI has held the lead. Engagement rate runs 14% higher, time on site 59% longer, bounce 33% lower, and add-to-cart rate 28% higher. On quality of traffic, AI referrals are already one of the strongest channels a retailer has.
Alongside those figures, Adobe published a different number. Across U.S. retail homepages, the average machine readability score came in at 61%. Put the other way, roughly four in ten pieces of homepage content never reach a shopping agent. The best customers a retailer gets are arriving through a door that keeps widening, and 40% of the shelves behind it are invisible.
What the 61% score actually measures
Before leaning on that number, it is worth knowing how it was produced. The instrument is the AI Content Visibility Checker, a free Chrome extension Adobe released in June 2026. It requires no Adobe license and scores any page in one click.
The definition is plain. The score expresses, out of 100%, how much of a page's content is visible to AI. Adobe calls it a citation readability score. A page scoring 50% has half its content unreadable by machines.
The retrieval method is the important part. The extension first tries to fetch the page the way ChatGPT's web crawler would, and when that request fails it falls back to the initial HTML before JavaScript runs as a proxy for what most agents can access. The human view, by contrast, is the fully rendered page with every script and dynamic element in place. Lining the two up exposes the content people see but agents never do. So the score is not measuring design quality or SEO strength. It measures whether the information is present in the raw HTML before rendering.
That framing matters for reading the shift from April to July. In the diagnostic results Adobe published in April 2026, the homepage average was 75%. In July it was 61%. But Adobe notes that it expanded the analysis in July to a broader set of U.S. retail sites. The same cohort did not deteriorate by 14 points in three months. Widening the lens pulled in a layer of sites that had not been measured before, and that layer dragged the average down. The honest reading is not that things got worse, but that the number now looks more like the industry as it really is.
Adobe's underlying insights are based on direct online transactions covering more than 1 trillion visits to U.S. retail sites, with a companion survey of over 5,000 U.S. respondents conducted in July 2026.
A 17-point spread across verticals
The 61% average conceals a wide range.
| Vertical | Homepage machine readability | Gap vs. the 61% average |
|---|---|---|
| Apparel | 76% | +15 points |
| Electronics | 70% | +9 points |
| Cosmetics | 68% | +7 points |
| Sporting goods | 67% | +6 points |
| Furniture and home | 64% | +3 points |
| General merchandise | 63% | +2 points |
| Grocery | 59% | -2 points |
Apparel at the top and grocery at the bottom are separated by 17 points. Apparel's 76% sits well above the overall average and holds roughly the level of the earlier, narrower April sample. Grocery's 59% means more than four in ten elements of the homepage never reach an agent.
Adobe does not explain what drives the gap, and guessing would be unwise. What the methodology does allow us to say is this: because the score depends on what exists in the HTML before JavaScript runs, sites that assemble their front page from dynamic personalization and recommendation modules are structurally disadvantaged. Categories such as grocery and general merchandise, where stock and pricing vary by store and region and are rendered client-side, are the ones most likely to score poorly for reasons that have little to do with editorial effort.
The real damage is on product pages, not homepages
The 61% homepage figure makes the better headline, but the problem that costs retailers money sits one level deeper.
April's diagnostic also broke results down by page type. Returns and exchanges scored 82%, contact us 81%, FAQ 80%, customer service and help center 79%, loyalty and membership 78%, category-level pages 74%, and store locator 73%. Individual product pages came last at 66%.
In other words, the return policy is read cleanly while the merchandise itself is read worst. On most sites, policy pages are written as static HTML, while product pages inject price, stock, reviews, and variants after the fact through JavaScript. To a person the two look equally complete. To an agent they are not remotely the same. We covered that structural issue when Adobe Commerce made product discovery on LLM surfaces generally available.
The weakness lands harder once you look at where AI traffic actually arrives. In a Criteo analysis of 500 U.S. merchants, more than 70% of AI-referred consumers landed directly on a product page, up sharply from about 50% in mid-2025. The same analysis found AI-referred consumers converting at 1.5 times the rate of other referral channels.
The picture resolves into something uncomfortable. The page an AI agent is most likely to send a shopper to is the page AI reads least well. Polishing the homepage before fixing that is the wrong order.
The spread between sites is worth noting too. In April, the best-performing U.S. retail homepages scored 82.5% while the lowest scored 54.2%. Two companies in the same industry can differ by nearly 30 points in how much of their content an agent can see. That gap is simply the distance between firms that started work on this and firms that have not.
Reasons not to take the number at face value
Having followed Adobe's data this far, a few caveats belong on the record.
First, the score is tied to Adobe's own commercial offering. The AI Content Visibility Checker is powered by Adobe Brand Visibility, and Adobe sells LLM Optimizer as the paid product for solving the same problem. This is not a third-party verified industry standard. A diagnosis of "your content is unreadable" arriving from the vendor selling the remedy deserves a discount.
Second, machine readability is a proxy for citability, not a guarantee of recommendation. Putting everything in the HTML does not mean an AI will choose your product. Being readable is a necessary condition, not a sufficient one.
Third, there is the question of scale. According to Bain & Company, AI now accounts for up to a quarter of referral traffic at some retailers, yet still represents less than 1% of their total traffic. Growth of 62% year over year, or 1,219% cumulatively, is growth on a small base. It is not yet the kind of channel that reshapes a revenue mix this quarter, and investment decisions should start from that fact. Bain also notes that around half of consumers remain uncomfortable letting AI handle a transaction end to end.
Even so, deferring the work is hard to justify. Improving machine readability is not special-purpose construction for AI. Server-side rendering and structured data overlap heavily with conventional SEO and performance work, so the spend is not stranded on a channel that is currently under 1%.
What e-commerce teams can check today
A practical order of operations.
Begin with observation. Pick one important product page, fetch the HTML with curl, and look for price, stock, specifications, and review text in the response. What matters is the pre-script source, not what the browser inspector shows after rendering. Adobe's extension will give you the same finding as a score, but reading the raw HTML yourself tends to make the gap obvious faster.
Then set priorities. Fix product pages first. Homepage scores make for the quotable headline, but they are not where conversion happens. Product pages, then category pages, then everything else is the sequence that pays back soonest.
The fix itself usually means revisiting how pages are rendered. Adobe's own guidance for sites that load most content through JavaScript is to enable server-side rendering or use prerendering tools that generate static HTML snapshots. Adding structured data on top raises the odds that content is not merely readable but correctly understood. The full arc of that work is laid out in LLMO for e-commerce sites, and a checklist with verification steps is in how to do GEO.
Conclusion
"AI traffic is growing" has become a premise rather than a finding. The question now is the condition of what receives it. A 61% homepage score and a 66% product page score say the receiving end still has holes in it.
Worth watching is how the number moves from here. If July's wider sample becomes the new baseline, the next quarterly reading will be a real measurement of how much the industry actually moved. Ahead of the holiday season, finding out what share of your own product pages an agent can read is a cheap thing to know.
