A shopper asks an AI assistant a simple, high-intent question: does this run true to size, or what is the waist measurement on a size 32. To answer, the assistant needs your sizing chart, and whether it can use it comes down to one unglamorous detail: can the model actually read the data. On a huge number of product pages, it cannot, because the sizing chart that looks perfectly clear to a human is published in a form a language model cannot parse. The chart is a picture, or a script-rendered widget, or a downloadable PDF, and to the model it is a blank space. Fixing this is one of the most concrete, winnable AEO tasks in e-commerce. Here is how extraction works and how to make your sizing data readable.
The short answer
AI answer engines extract sizing charts by reading the text of your page, so the format the data lives in determines whether they can use it. A sizing chart rendered as clean HTML text, ideally a proper labeled table, is readable and extractable. The same chart saved as an image, trapped in a script-only widget, hidden behind a click tab that is not in the page source, or buried in a PDF is effectively invisible, because it is not text the model can read. Profound’s analysis shows models lift clean, directly available data (Profound), and sizing data is only clean and available if it is real text. Make it text, and extraction follows.
How extraction actually works
Strip away the mystique: an answer engine reading your product page is working from the text and structure it can access, much like any reader of the page source. When it finds measurements written as text, in a table or clearly labeled lines, it can parse which number goes with which size and dimension and reuse that to answer a shopper. When the measurements are not present as text, there is nothing to parse. This is the same clarity-and-readability principle that governs all AI extraction, where Ahrefs found visibility rests on readable, relevant content rather than tricks (Ahrefs). Extraction is reading; if the model cannot read it, it cannot extract it.
The biggest blocker: sizing charts as images
By far the most common reason AI cannot use a sizing chart is that the chart is an image. Designers love a nicely styled size-table graphic, and it looks great to shoppers, but a JPG or PNG of a table is a picture of text, not text. A language model reading the page source sees an image reference, not the measurements inside it, so the data is unreadable. If you do one thing after reading this, it is this: stop publishing sizing charts as images and publish them as real HTML text. The pretty graphic is precisely what is hiding your data from the machine.
Blocker two: data trapped in scripts or click-tabs
The second common failure is data that only exists after interaction or client-side rendering that is not reflected in the crawlable source. A sizing chart that loads into a tab only when clicked, or is injected by a script in a way that leaves the page source empty, may be invisible to an engine that reads the delivered HTML. The safe approach is to ensure the sizing data is present in the page source that a crawler receives, not conjured only by a user action. If a simple view-source or fetch does not show the measurements as text, assume the model cannot see them either.
Blocker three: PDFs and downloads
A third pattern is putting the real sizing detail in a downloadable PDF or spec sheet linked from the page. Shoppers can open it, but the measurements are one step removed from the page content and far less reliably part of what the engine extracts to answer inline. Do not make your key sizing data a download; put it on the page itself as text. Reserve downloads for supplementary depth, and ensure the answers to the common sizing questions live in the readable page body where they can be extracted directly.
Blocker four: vague, unlabeled data
Even when the data is text, it fails if it is unlabeled or ambiguous. A wall of numbers with no clear header saying which is chest, waist, or length, or a size table where the columns are not clearly named, gives the model text it cannot confidently map. Extraction needs not just text but labeled structure, so each measurement is unambiguously tied to a size and a dimension. Clarity is as important as format, the same lesson as in does having structured tables make it easier to get an AI snippet: a clean, labeled table extracts; a messy one does not.
What makes a sizing chart extractable
Pulling it together, an extractable sizing chart has these properties. It is real HTML text, not an image. It is present in the crawlable page source, not injected only on interaction. It is structured as a table or clearly labeled lines, with every size and dimension named. Its values are concise and unambiguous. And it lives on the page itself, not in a download. Meet those and the model can read, parse, and reuse your sizing data to answer shoppers accurately, which is the goal.
Format, extractability, and the fix
This table sorts the common formats.
| Sizing chart format | Extractable by AI? | Fix |
|---|---|---|
| HTML table, labeled, in source | Yes | Keep it; this is the target |
| Image of a size table | No | Rebuild as HTML text |
| Script-only or click-tab, not in source | Usually not | Render into the page source |
| PDF or downloadable spec | Poorly | Put the data on the page as text |
| Unlabeled block of numbers | Unreliably | Add clear headers and labels |
The role of product structured data
Structured data can reinforce readable sizing content by describing product attributes in a machine-readable way, and Google documents structured data as a means to make what your content is about explicit (Google Search Central). But keep its role in proportion: evidence suggests markup alone barely moves AI citations (Ahrefs), so structured data supports, rather than replaces, having the sizing data as clear HTML text. Lead with readable content, add structured data where it applies, and treat markup as reinforcement, a balance covered in structured data for LLM SEO.
How to make your sizing data AI-readable
The remediation is concrete. Convert every image-based size chart into a real HTML table with labeled rows and columns. Verify the table appears in the page source, not just after a click, so crawlers receive it. Move any critical sizing detail out of PDFs and onto the page as text. Label every measurement clearly so each number maps to a size and a dimension. And keep the data accurate and current. This is the same readability discipline that makes any content model-friendly, laid out in how to write RAG-friendly content, applied to the specific, high-value case of product measurements.
Why this matters
Sizing questions are high-intent, close-to-purchase questions, and getting them answered well matters twice over. When the AI can accurately state your sizing, shoppers get the confidence to buy and are less likely to order the wrong size and return it. When it cannot, the assistant may hedge, omit your product, or, worst of all, guess wrong, sending a buyer toward a return or a competitor. For merchants this connects to broader product visibility, since being readable is the precondition for being surfaced at all, as discussed in why your Shopify products are missing in Gemini. Readable sizing data is both an extraction win and a conversion-and-returns win.
How to test extractability
Do not assume; test. Ask an AI assistant your real sizing questions about a specific product, and see whether it answers correctly from your page. Check your page source, or a plain fetch of the URL, to confirm the measurements appear as text rather than only as an image or post-click content. If the assistant cannot answer or gets it wrong, trace it back to the format: it is almost always an image, a script-only render, or a PDF hiding the data. Fix the format and re-test, and the answers improve.
A worked example
Say you sell apparel and your beautifully designed size chart is a single PNG on each product page. A shopper asks ChatGPT for the medium’s measurements and it cannot answer, or invents a number, because it never read your image. You rebuild the chart as a clean HTML table, labeled by size and dimension, present in the page source, and add product structured data. You re-ask, and now the assistant states your measurements accurately and confidently. Nothing about the numbers changed; the format did, from a picture the model could not read to text it could. That single format change turned an invisible chart into an extractable one, and turned a hedge into a confident, correct answer.
Common mistakes
The biggest mistake is publishing sizing charts as images, which look great and hide the data from every model. The second is relying on script-only or click-tab rendering that leaves the measurements out of the page source. The third is stashing the real data in a downloadable PDF. The fourth is presenting unlabeled numbers the model cannot confidently map. And the fifth is leaning on structured data to compensate for content that is not readable text, when markup only reinforces and does not replace it. Make the data real, labeled, crawlable text, and the extraction problem disappears.
The bottom line
How do AI answer engines extract sizing charts from product pages? By reading the page’s text, which means the format decides everything. A sizing chart in clean, labeled HTML that lives in the crawlable source is extractable; one locked in an image, a script, a click-tab, or a PDF is invisible, no matter how clear it looks to a human. The dominant fix is to stop using image-based charts and publish measurements as real text, structured and labeled, reinforced with product structured data. Then test by asking the engines your sizing questions. Make the data readable, and AI can finally answer the high-intent questions that help shoppers buy with confidence.
Frequently asked questions
How do AI answer engines extract sizing charts from product pages?
By reading the page’s text. A sizing chart written as clean HTML, especially a labeled table, is readable and extractable, so the engine can answer measurement questions from it. A chart saved as an image, locked in a script, or hidden in a PDF is not text the model can reliably read, so it cannot extract it.
Why can’t AI read my sizing chart if it is on the page?
Because being visible to a human is not the same as being readable text to a model. If your chart is an image file, rendered only by JavaScript in a way that is not in the page source, or inside a downloadable PDF, the measurements are not extractable text, so the engine cannot use them even though shoppers can see them.
What is the best format for an AI-extractable sizing chart?
A real HTML table with clearly labeled rows and columns, present in the crawlable page source, is the most extractable format. Each measurement should be text, each size and dimension clearly labeled, and the whole table readable without a click or download. Reinforce with product structured data where it applies.
Does product structured data help AI extract sizing info?
It can help by describing product attributes in a machine-readable way, which reinforces the readable content. But it is a supporting signal, not a replacement for having the sizing data as clear HTML text on the page. Lead with readable content and add structured data as reinforcement.
