Specoria
Menu
TR Free test

InsightsProduct data

A third of a product page is invisible to AI: how to make product pages machine-readable

Adobe's checker found that US retail product pages average 66% machine-readable content, lower than homepages and category pages. Where the missing third usually hides and how to bring it into the HTML an agent actually reads.

By 6 min read

A shopper asks an assistant whether a jacket is warm enough for minus ten degrees. The answer is on your product page, in a features tab that opens with a click. The assistant reads the page, finds the name and a short description, and moves on to a competitor whose spec table sits in plain HTML. Nothing on your page was wrong. The assistant just never saw it.

In this article, we look at Adobe's April 2026 benchmark of how much of a retail page machines can read, explain where product pages usually lose content, and give a short test you can run on your own pages this week.

What Adobe measured

Adobe runs a diagnostic tool, the AI Content Visibility Checker, that scores a page out of 100% for how much of its content machines can read. A score of 50% means half the content is not readable. In April 2026, Adobe published averages for US retail sites:

  • Homepages: 75%. The best-performing retailers averaged 82.5%, the lowest 54.2%.
  • Category pages: 74%.
  • Product pages: 66%, the lowest of the main page types.
  • Other pages: store locator 73%, loyalty 78%, help center 79%, FAQ 80%, contact 81%, returns 82%.

Adobe's own reading is direct: retailers have thousands of SKUs, and much of that content is currently invisible to LLMs. The same report shows why it matters. In March 2026, AI traffic to US retail sites converted 42% better than non-AI traffic.

Two cautions. These are US averages, not a measurement of your store. And the post doesn't explain in detail what the checker counts as unreadable, so read the scores as a direction, not a diagnosis.

Where the missing third usually hides

The product page is the most dynamic page on most stores, which is exactly why it scores lowest. On most stores, the content machines miss sits in a few familiar places.

  • Content loaded after the page. Price, stock, delivery date or reviews filled in by JavaScript after the first response. An agent that reads the initial HTML sees an empty slot.
  • Tabs and accordions that fetch on click. A tab that only hides text is fine; a tab that requests the text when opened means the text isn't in the page at all.
  • Text inside images. Size charts, feature banners and "what's in the box" graphics carry facts no text reader can extract.
  • Review widgets. Third-party review boxes often load late from another domain. The star rating a shopper sees may be missing for the agent.
  • Variant data in scripts. Sizes and colors held in a JavaScript object, with no visible text or separate offer for each.
  • Specs as marketing copy. "Built for the coldest days" instead of a temperature rating. It is readable, but it doesn't answer the question.

What machine-readable looks like

What to do: Put the facts a shopper would ask about into the HTML your server sends, as text, and repeat the core ones in structured data.

  1. Name, price, currency and availability in the initial HTML. This is the first thing an agent looks for. Google also recommends Product markup in the initial HTML, because markup generated by JavaScript can make Shopping crawls less frequent and less reliable for fast-changing data like price and availability.
  2. Key specifications as a text table. Material, dimensions, weight, compatibility, care. One row per fact, in units, not adjectives.
  3. Size charts as HTML tables, not images. Keep the image if you like it, but put the numbers next to it.
  4. Reviews summarized on the page. Render the rating and review count server-side and add aggregateRating to your Product markup.
  5. One offer per variant, each with its own price and availability. We covered the structure in variant pages an agent can read.
  6. Delivery and returns in words and numbers on the product page, not only in checkout. See stock and delivery promises an agent can check.

None of this changes how the page looks to a person. A tab can still be a tab; its content just has to be in the page when it arrives.

A ten-minute test

You don't need a special tool for a first look.

  1. Open a best-selling product page and view the page source (not the inspector; the source shows what the server sent).
  2. Search the source for the price as it appears on screen. If it isn't there, an agent reading the HTML can't see it.
  3. Search for one sentence from each tab, one size from the size chart and the review count.
  4. Look for a <script type="application/ld+json"> block with "@type": "Product" and check that its price matches the screen.
  5. Repeat for a product with variants and one that is out of stock.

Write down every fact that appears on screen but not in the source. That list is your missing third.

How this connects to the rest of your data

Readable is the first layer, not the last. A price an agent can read but that disagrees with your feed is still a problem; we covered that in your feed, your page and your structured data should agree on price. And a readable page behind a bot wall is still unreadable; our bot policy by page type covers which pages to open.

Which pages to fix first

You don't need to fix every product at once. Start where a missing fact costs the most.

  • Best sellers. They are the products an assistant is most likely to be asked about.
  • Products with variants. Size and color are the most common follow-up questions, and the most common place for data to hide in scripts.
  • Products with a delivery or stock promise. "Ships today" or "only 3 left" only helps if an agent can read it.
  • Category pages. Adobe scored them at 74%. Agents use them to narrow a search, so make sure each listed product shows its name and price as text, not only inside a script that renders the grid.

Policy pages scored better in Adobe's data (returns 82%, FAQ 80%), which is not surprising: they are mostly plain text. Keep them that way, and link to them from the product page so an agent can find the terms that apply.

The bottom line

Adobe's benchmark puts a number on something product teams often sense: the product page, the page that matters most for a purchase decision, is the one machines read least. Most of the gap comes from content that arrives late, hides in images or lives in scripts. Moving the facts into the first HTML response, as text and in structured data, closes much of it without a redesign.

Specoria checks whether your product name and price are readable without JavaScript, whether the add-to-cart button is in the initial HTML, and whether your structured data is complete and consistent with the page. Start with a free readiness test.

Sources