An AI shopping agent needs the same thing a careful shopper needs: a catalog it can read without guessing. That means every product has a stable identity, a current price, a true stock status, and a description that matches the page a human sees. If any of those four are missing, wrong, or different in different places, the agent either skips your product, quotes the wrong facts, or sends the shopper to a competitor it can check.
There is no single catalog format that every agent reads, and no vendor publishes a checklist that guarantees an agent will pick your products. What the vendors do publish points in the same direction. Google says to share product data through structured data on product pages and through a feed uploaded to Merchant Center, because web crawling is not guaranteed to find all products on a site. OpenAI says ChatGPT considers structured metadata from first-party and third-party providers, and that merchants can provide a direct product feed. Perplexity surfaces sites its crawler can read, and runs a merchant program for retailers that want to share product specs directly.
Start with one source of truth
Most catalog problems start before any agent is involved. The price lives in the store platform, the stock count lives in a spreadsheet, and the description lives in a marketplace listing. An agent reads all of those places, finds the disagreement, and trusts none of them.
Pick one system as the source of truth for each fact, and make every other place draw from it. For most small brands, the store platform is the source of truth for price and stock. A feed, a marketplace listing, and a schema block should all be generated from that same data, not typed separately by hand. When a price changes, change it in the source system and let the change flow out.
Give every product a stable identifier and keep it stable. Google's product data specification uses a unique ID for each product and recommends using the SKU where possible, and keeping the ID the same when updating data. Use that ID in your feed, in your schema, and in your internal records. Do not reuse an ID for a different product.
The fields every agent needs
An agent cannot recommend what it cannot identify. These are the fields to get right first, on every product, before you add anything fancy.
Name and description. Use the real product name, spelled the same way everywhere, and a description that states what the product is, what it is for, and what it is made of. Keep the name short and specific. Keep the description factual. A name that changes spelling from place to place looks like several different products, and a description that is only a slogan says nothing an agent can match to a question.
Price and currency. State the price in the visible page text and in the schema, with the currency included. Keep the price current. When a sale price applies, state the sale price, the regular price, and when the sale ends, in the same place. An agent that finds two different prices for the same product has to pick one, and it may pick the wrong one. OpenAI says prices in ChatGPT come from third-party providers and there may be a delay when merchants update pricing, so a single correct source matters more than a fast correction in one place.
Availability. State whether the product is in stock, out of stock, or available for preorder, in the visible text and in the schema, and update it when the stock changes. Availability is one of the factors OpenAI names for ranking merchants. A product shown as available when it is not is the fastest way to lose trust, because the shopper finds out at checkout.
Identifiers. Include the brand, the SKU, and a product identifier such as a GTIN or MPN where one exists. Google's product data specification warns that incorrect, inaccurate, or missing product information can cause disapprovals or limited eligibility. Not every product has a GTIN, and you should not invent one.
Images. Provide at least one clear image of the actual product. An agent that shows a product card needs an image, and a shopper that clicks through needs the image to match what arrives. Do not use a placeholder image for a live product.
Variants. Treat each variant as its own item in the data, even when it shares a product page. A shirt in three sizes and two colors is six variants, each with its own price, stock status, and identifier. Google's product data specification requires variant attributes such as item group ID, color, and size to be submitted correctly. In your schema, use a ProductGroup with the variants nested under it where your platform supports that, and make sure each variant states its own price and availability.
The three ways agents read your catalog
Agents reach your catalog through three paths, and your preparation should cover all three, because different agents lean on different ones.
The crawl path. An agent's crawler reads your public pages. For this path, the product page is the catalog. The price, availability, and description must be in the initial HTML, not loaded only after a shopper clicks or after a script runs. Keep the crawlers that feed shopping answers allowed in robots.txt, including OAI-SearchBot for ChatGPT search and PerplexityBot for Perplexity. A page a crawler cannot reach is a product that does not exist for that agent.
The schema path. Product schema states the same facts in a machine-readable form, so an agent does not have to guess from your page design. Google says structured data can improve the accuracy of its understanding of content such as price, discount, and shipping costs on a page. Put schema on every product page, and make it say exactly what the visible page says. Check it with a validation tool before you trust it, and check it again after every theme or app change.
The feed path. A feed is a structured file listing your items, their prices, availability, and images, submitted to a service such as Google Merchant Center or directly to an agent provider. Google says that for larger sites or sites with frequently changing content, you should periodically upload new data feed files, or use an API for immediate updates. Generate the feed from the same source of truth as your product pages, so the feed and the page can never disagree. A feed typed by hand from a separate spreadsheet will disagree, and the disagreement will show up in an answer.
Shipping, returns, and the facts buyers ask about
An agent's job does not end at naming a product. A shopper asks about shipping cost, delivery time, and returns, and an agent that cannot find those facts either guesses or leaves them out. Publish your shipping and return policies on pages an agent can read, and state the key facts on the product page itself: the shipping cost or the threshold for free shipping, the delivery window, and the return window.
You do not need to mark up every policy detail on every product page. You do need the facts to be findable, consistent, and current, on pages a crawler can reach.
The same applies to discontinued products. When a product is discontinued, say so on the product page, mark it out of stock, and remove it from the feed. An agent that keeps recommending a product you no longer sell is a sign that your catalog is stale.
What about the new commerce protocols
You will hear about new protocols for agentic commerce. Google's guide to generative AI features notes that protocols like the Universal Commerce Protocol are emerging that will allow search agents to do more. Neither vendor describes these as a replacement for the crawl, schema, and feed paths above. Treat them as a next step to watch, not as the first step to take.
The transaction layer is a separate question from the catalog. A catalog tells an agent what you sell and whether it is available. A transaction layer lets an agent complete a purchase. Any agent shops an Aizii-optimized site via MCP/API. Do not claim that a specific agent transacts on your site unless that is true and verified.
The catalog cleanup routine
A catalog is not a project you finish once. It is a list you keep clean. Build a short routine and run it on a schedule you can keep.
- Check the top sellers first. Open your best-selling product pages as a shopper would, and confirm the price, stock status, and description match your store admin. Fix the top sellers before you touch the long tail.
- Compare the feed to the pages. Pick a sample of products and compare the feed entry to the live product page. If they disagree, fix the source of truth and regenerate the feed.
- Check the schema output. Run a sample of product pages through a structured data validation tool and confirm the price, availability, and identifiers are being read.
- Remove or mark what you no longer sell. Discontinued products should be marked out of stock, removed from the feed, and given a clear page that says so, with a link to the replacement where one exists. A stale product in the feed is a wrong answer waiting to happen.
- Watch for drift after every change. A new app, a theme update, or a bulk import can change prices or schema output without anyone noticing. Recheck a sample of pages and the feed after any of those changes.
The fundamentals
These mechanics are shared across every engine and covered in full in their own guides:
- •The llms.txt format and a complete sample, see our guide to writing a good llms.txt file
- •The markup AI systems read to verify facts, see our product schema guide
- •The complete crawler rule reference, see our robots.txt guide for AI search
Say you run a camping cookware brand
Say you run a small brand that sells camping cookware: a nesting pot set, a folding stove, and a set of enamel mugs. Your store admin is the source of truth for price and stock. Your product pages state the weight, the capacity, the materials, and the price in plain text, and your schema states the same facts. Your feed is generated from the same admin data, so the feed and the pages agree. One size of the pot set sells out, and you update the admin the same day, so the page, the schema, and the feed all show that size as out of stock together. A shopper asks an assistant for a lightweight pot set for two people under a set budget. The assistant can check your weight, your capacity, your price, and your stock, and it names your set. A competitor has a similar set, but its feed still shows a size it sold out of weeks ago, and its page shows a different price. The assistant cannot trust the competitor's facts, so it does not name them.
What this guide does not cover
This guide covers preparing your catalog so AI shopping agents can read it and trust it. It does not cover the per-engine playbooks for getting each assistant to find you, the full technical audit that checks every crawler and every page, or the transaction layer that lets an agent complete a purchase once your brand is recommended. Those are separate jobs, and this guide is the map, not the whole trip.
Check your own brand now
Run the free scan. One click, no signup, and it shows whether AI crawlers can read your brand today: aizii.ai/aeo-scan
Run the free scanSources
- Google Search Central, Share your product data with Google (specific): https://developers.google.com/search/docs/specialty/ecommerce/share-your-product-data-with-google
- Google Search Central, Product structured data (specific): https://developers.google.com/search/docs/appearance/structured-data/product
- Google Search Central, Optimizing your website for generative AI features on Google Search (specific): https://developers.google.com/search/docs/fundamentals/ai-optimization-guide
- OpenAI Help Center, Shopping with ChatGPT Search (specific): https://help.openai.com/en/articles/11128490-shopping-with-chatgpt-search
- OpenAI, Overview of OpenAI crawlers (specific): https://platform.openai.com/docs/bots
- Perplexity, Perplexity crawlers (specific): https://docs.perplexity.ai/guides/bots
- Schema.org, Product type (specific): https://schema.org/Product
Note: links marked specific point to the page that supports the claim in this guide. Links marked general docs point to the vendor's general docs section for that topic, not to a page that proves every step. Check the vendor docs for the current path before relying on a menu name or click path in this guide.
