How AI Shopping Agents Read Your Product Pages (and What They Need)

An AI shopping agent reads a product page as data. It takes the text in the page, the structured data in the code, and any feed or API your store offers. It needs a clear name, attributes, price, currency, stock status and policies. It does not need your layout, and it often cannot use what is hidden in images or scripts.

What an agent actually sees

Different agents work in different ways. Some load the page as plain HTML. Some run a browser and execute JavaScript. Some read images with a vision model. Some skip the page and use a feed or an API.

You cannot know which one visits you. So the safest rule is simple: put every important fact in plain text in the HTML, and repeat it in structured data.

An agent can draw on these sources:

• Visible page text. Title, description, bullet lists, size guides, shipping notes.
• Structured data. JSON-LD blocks that follow schema.org, such as Product and Offer.
• Meta tags and page titles. Open Graph tags and the title tag.
• Product feeds. For example, a feed you send to Google Merchant Center.
• APIs and tools. Public store APIs, or an MCP server if the store offers one.
• Images. Read by vision models, but with less precision than text.

What an agent needs from a product page

An agent has to answer a buyer's question. To do that, it needs facts it can trust. Check that each of these is present and correct.

Identity

• A specific product name. "Merino beanie, grey, one size" beats "Winter hat".
• Brand and model.
• Identifiers such as GTIN, MPN and SKU, where you have them.

Offer

• Current price as a plain number.
• Currency.
• Stock status, with the real state.
• Sale price and the end date, if there is one.

Attributes

• Size, color, material, dimensions, weight.
• Compatibility, for example "fits iPhone 15".
• Variants, with each option clear and linked to its own price and stock.

Terms

• Shipping cost and delivery time for the target country.
• Return window and return conditions.
• Warranty terms.
• Whether the price includes VAT.

Trust

• Real reviews with real ratings.
• Seller name and contact details.

What blocks agents

Many problems come from design choices that work for humans but not for software.

• Facts only in images. Size charts, spec sheets or "free shipping" banners that exist only as pictures.
• Content that loads late. Descriptions, prices or variants that appear only after a script runs or a button is clicked.
• Hidden tabs. Important details inside tabs or accordions that are not in the HTML.
• Vague titles. Names that use only a brand style or a nickname.
• Marketing text without facts. "Amazing quality" says nothing an agent can compare.
• Conflicting data. The page shows 29.90, the structured data says 24.90, the feed says 27.90.
• Pop-ups and consent walls. They can cover the content or block a simple crawler.
• Duplicate product pages. Several URLs for the same item, with different data.

How to test your own pages

You can run a simple check on a few top products. It takes about 20 minutes.

1. Open the product page and choose "View page source". Search for the price and the product name. They should appear in the HTML.
2. Turn off JavaScript in your browser and reload the page. Check which facts are still there.
3. Paste your structured data into the Schema Markup Validator or Google's Rich Results Test. Fix errors and warnings.
4. Compare the price and stock in the page, the structured data and your feed. They must match.
5. Copy the visible text of the page into an AI assistant. Ask: "What are the size, material, price and return terms of this product?" If the answer is wrong or empty, your page is unclear.
6. Check your robots.txt file. Make sure you did not block crawlers that you want to allow.

Run these checks on your own pages and your own test data. Do not repeat them at a large scale on a live system.

A page that works well for both people and agents

A strong product page has this order:

1. A clear H1 with the product name.
2. A short summary of two to three sentences.
3. Price, currency and stock status near the top.
4. A list of key attributes as bullet points.
5. A short, plain shipping and returns block.
6. Structured data that repeats the same facts.
7. Images with useful alt text.

This helps normal search engines and human visitors too. You do not need to build a separate version for agents.

Frequently asked questions

Do AI agents read JSON-LD structured data?

Many systems do read it, because it is a clean and consistent source of facts. Support differs by agent, so do not rely on it alone. Keep the same facts in the visible text too.

Should I make a separate page for AI agents?

No. Use one page with clear text and matching structured data. A separate version can drift out of date and may cause conflicting data.

Should I block AI crawlers in robots.txt?

That is your business choice. Blocking can keep your content out of AI answers, and allowing can bring in visits or sales. Decide per crawler, and review the official crawler names of each provider.

Do images still matter?

Yes, for human buyers and for vision-based agents. But every key fact should also exist as text, because text is more exact and cheaper to read.