What is an AI-ready product page?

A product page is designed for a person. They look at the photo and work out the color, zoom into the size chart image, open the shipping tab and, if they are unsure, message the store's chat. An assistant answering a shopper in ChatGPT, in Perplexity or in Google's AI Mode does none of that: it reads what it can read, and what is not written down, it does not know. "AI-ready product page" is not a standard or a feature of any platform. It is the name we use in this glossary for the question that gap opens: what does an AI assistant need from a page that a person does not? The answer is not a different page, but the same page with what the person does not miss and the machine lacks.

What an assistant needs that a person doesn't

Microsoft explains it from the search engine's side: AI assistants do not read a page top to bottom the way a person does. They split it into smaller pieces, and those pieces are what gets ranked and assembled into the answer.⁠^{[1]Microsoft Advertising (Krishna Madhavan, October 8, 2025). Optimizing Your Content for Inclusion in AI Search Answers} Five conditions follow.

It can get in and read the HTML. The first requirement is access. OpenAI, for example, separates the agent that surfaces websites in ChatGPT search, OAI-SearchBot, and warns that sites that block it are not shown in ChatGPT search answers.⁠^{[2]OpenAI. Overview of OpenAI Crawlers} And getting in is not enough: Google reminds you that server-side rendering is still a good idea because not all bots can run JavaScript.⁠^{[3]Google Search Central. Understand JavaScript SEO basics} A price, a description or reviews that only appear after a script runs may not exist for the crawler that visits your page.

Every fact in a sentence that makes sense on its own. If the assistant keeps a fragment, that fragment has to say something complete. Microsoft asks for sentences that make sense out of context, one- or two-sentence answers to a question, and lists and tables that can be copied cleanly. And it warns of the opposite: an adjective with no context leaves the AI unsure how to classify the product.⁠^{[1]Microsoft Advertising (Krishna Madhavan, October 8, 2025). Optimizing Your Content for Inclusion in AI Search Answers} "Lightweight" next to the photo works for a person; for the assistant, "weighs 540 grams per boot" works. That is why product attributes, with their name, value and unit, weigh more on this page than on any other.

Identifiers that say which product it is. A person recognizes a product by the photo and the name. A machine comparing the same item across several stores, feeds and reviews needs an identifier. Google recommends including every global identifier that applies in product markup, the GTIN among them, and describes the MPN as the number that uniquely identifies a product for a manufacturer.⁠^{[4]Google Search Central. Merchant listing (Product, Offer) structured data} OpenAI's product feed specification goes further: brand is required, and a valid GTIN or MPN has to be given unless the product has no identifier.⁠^{[5]OpenAI, Agentic Commerce. Products (product feed specification)}

The same thing everywhere. An assistant can cross-check the page text, its JSON-LD and the feed. Google requires that structured data faithfully represent the page content and that you not mark up what the visitor cannot see,⁠^{[6]Google Search Central. General structured data guidelines} and Microsoft points out that consistency between the title, the H1 and the description improves AI systems' confidence in what they read.⁠^{[1]Microsoft Advertising (Krishna Madhavan, October 8, 2025). Optimizing Your Content for Inclusion in AI Search Answers} One price in the text and another in the markup is not a detail: it is a contradiction the machine has to resolve without you.

The attributes AI commerce asks for. OpenAI's specification says what a system that sells from a conversation wants to know about a product: a plain-text description of up to 5,000 characters, the main material, a color consistent with the image, at least two of the three dimensions, the net weight without packaging, the number of reviews and the average rating.⁠^{[5]OpenAI, Agentic Commerce. Products (product feed specification)} It is a good list of what your page should say in text. And Google, for its part, says that Google Merchant Center feeds help your products show up in its AI answers.⁠^{[7]Google Search Central. Optimizing your website for generative AI features on Google Search}

In a store

A fictional example: a hiking boot's product page, what a person sees on it and what an assistant needs written down.

No row changes what the shopper sees. Every one changes what an assistant can repeat when someone asks which boot weighs under 600 grams and takes crampons.

How to check it

1. Open your robots.txt and confirm you let through the assistants' search agents, such as OAI-SearchBot.⁠^{[2]OpenAI. Overview of OpenAI Crawlers}
2. Open the source code of a product page (Ctrl+U, not the inspector) and look for the price, the description, the main attributes and the rating. What is not there may not reach a bot that does not run JavaScript.⁠^{[3]Google Search Central. Understand JavaScript SEO basics}
3. Read each attribute and each key sentence out of context. If "lightweight" or "durable" do not say how much, rewrite them with a number and a unit.⁠^{[1]Microsoft Advertising (Krishna Madhavan, October 8, 2025). Optimizing Your Content for Inclusion in AI Search Answers}
4. Compare the page, the JSON-LD and the feed on name, price, availability, brand and GTIN. They have to say the same thing.⁠^{[6]Google Search Central. General structured data guidelines}
5. Ask an assistant three facts about your product that are on the page, and compare its answer with what the page says. One test is a sample of one, but it shows which facts it is not finding.

Related concept: product page

The product page is the page; the AI-ready product page is that same page judged by a different reader. Everything a good product page needs (name, images, price, variants, description) this one needs too, and nothing added for the machine gets in the person's way. The difference is in what the person solves alone and the assistant does not: inferring from a photo, reading an image, opening a PDF, waiting for a script, recognizing a product without its code. When those conditions are met across the whole catalog and not on a single page, the term is a machine-readable catalog.

To go further, the catalog and entities reading path orders the rest of the terms, and the guide to auditing a product page's structured data turns it into steps.