Skip to content
Chalk Labs / AI SEO
How-to guide

How to Optimize Product Pages for ChatGPT and Google AI Mode Shopping

By Rahil Jain, founder of Chalk LabsUpdated How we research

The short answer

To optimize product pages for AI shopping, make price, stock, specs, shipping and returns visible in plain HTML, add matching Product schema, and keep your product feeds in step with the page. Google uses Merchant Center and structured data, and OpenAI accepts product feeds. Neither publishes a ranking formula, so accuracy and consistency are your levers 14.

  • Price, availability, specs, shipping and returns should sit in plain visible HTML, not inside images, tabs or PDFs.
  • Google says providing both structured data and a Merchant Center feed maximizes eligibility and helps it verify your data 1.
  • OpenAI accepts a product feed so ChatGPT can accurately surface products. Eligibility does not guarantee display 5.
  • Neither company publishes how it ranks products. Do not trust anyone who claims to know the formula.

How do AI shopping answers choose products?

The public record is thin. Google said on 20 May 2025 that its Shopping Graph held more than 50 billion product listings, with more than 2 billion refreshed every hour 6. OpenAI's documentation says merchants can share structured product data so ChatGPT can accurately surface their products in search and shopping 4.

What neither company publishes is the ranking rule that picks one product over another. We found no official statement that ranks price, reviews or brand strength. So the practical goal is narrower: make sure every system that reads your data finds the same, complete and current facts. A wrong price or an out-of-stock item that still shows as available is the easiest way to lose a recommendation.

Do it step by step

0 of 8 done0%

  1. Make a sheet with one row per product: price, currency, stock status, GTIN, brand, size or weight, shipping time and return window. This sheet is the truth that every other copy must match.

  2. Open with a one-sentence summary, then a plain-text spec list, then who it is for and who should skip it. Move specs out of images, tabs and PDFs into HTML text.

    Tip: Read the page with styles off, or with JavaScript disabled, to see what a simple crawler sees.

  3. Use the Product template below. Print the price, currency and availability in the markup from the same database field as the visible price, so they cannot drift apart.

  4. Put the return window and shipping terms in visible text first. Then add hasMerchantReturnPolicy and shippingDetails markup that states the same terms.

  5. Check that each product has an ID, title, description, link, image link, price, availability and brand, and that the title and stock match the landing page. Add GTINs where you have them.

  6. Use the feed record template below as a guide to the required fields. Check OpenAI's current onboarding steps, then send a sample feed and refresh it whenever price or stock changes.

  7. Show genuine reviews on the page with the reviewer's name where you have consent. Keep ratings in markup and feeds tied to the exact product variant they describe.

  8. Compare page, markup and both feeds for the ten products. Then ask ChatGPT and Google AI Mode five real buying questions each and log which products appear, with a date.

    Tip: Repeat monthly. Answers vary between runs, so one test proves little.

Your ticks are saved in this browser only.

Copy-ready templates

Product and Offer JSON-LD

json
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Trailhead 28L Daypack",
  "image": [
    "https://www.example-shop.com/img/trailhead-28l-front.jpg"
  ],
  "description": "A 28 litre daypack with a rain cover, laptop sleeve and a 650 gram body weight.",
  "sku": "TH-28-GRN",
  "brand": {
    "@type": "Brand",
    "name": "Trailhead"
  },
  "gtin13": "4006381333931",
  "offers": {
    "@type": "Offer",
    "url": "https://www.example-shop.com/packs/trailhead-28l",
    "priceCurrency": "USD",
    "price": "79.99",
    "availability": "https://schema.org/InStock",
    "itemCondition": "https://schema.org/NewCondition",
    "hasMerchantReturnPolicy": {
      "@type": "MerchantReturnPolicy",
      "applicableCountry": "US",
      "returnPolicyCategory": "https://schema.org/MerchantReturnFiniteReturnWindow",
      "merchantReturnDays": 30
    }
  },
  "aggregateRating": {
    "@type": "AggregateRating",
    "ratingValue": "4.6",
    "reviewCount": "212"
  }
}

Placeholder product and shop. Every value must match what the page shows. Remove aggregateRating unless real reviews appear on the same page.

One product record for a ChatGPT-style product feed (JSONL, one line per product)

json
{"item_id": "TH-28-GRN", "title": "Trailhead 28L Daypack, Green", "description": "A 28 litre daypack with a rain cover, laptop sleeve and a 650 gram body weight.", "url": "https://www.example-shop.com/packs/trailhead-28l", "brand": "Trailhead", "seller_name": "Example Shop", "image_url": "https://www.example-shop.com/img/trailhead-28l-front.jpg", "availability": "in_stock", "price": "79.99 USD", "gtin": "4006381333931", "condition": "new", "star_rating": 4.6, "review_count": 212}

Field names follow OpenAI's published feed spec: required item_id, title, description, url, brand, seller_name, image_url, availability, price. Placeholder values. Confirm current onboarding with OpenAI.

Product page consistency checklist

markdown
- [ ] Visible price, currency and stock match the markup
- [ ] Visible price and stock match the Merchant Center feed
- [ ] Visible price and stock match the OpenAI feed
- [ ] Title on the page and in the feeds says the same thing
- [ ] Specs, shipping and returns are plain text, not images or PDFs
- [ ] GTIN and brand present where they exist
- [ ] One product (or its variants) per URL
- [ ] Reviews are real, shown on the page and tied to this product
- [ ] A line that says who it is not for

Run it for each of the ten top products, and again whenever you change a template.

Which four places do these systems read?

Treat each product as having four copies of its facts: the visible page, the structured data on that page, the Google Merchant Center feed and any feed you send to OpenAI. Each one must agree. Google says supplying both structured data and a Merchant Center feed maximizes eligibility, and some experiences fill gaps in one source from the other 1.

Where product facts live and who uses them
SourceWho reads itWhat you control
Visible product pageEvery crawler, plus peoplePrice, stock, specs, shipping, returns, photos, reviews
Product structured dataGoogle Search and other parsersName, image, offers, brand, GTIN, ratings, return policy 2
Merchant Center feedGoogle shopping surfacesProduct ID, title, price, availability, brand and identifiers 3
OpenAI product feedChatGPT shoppingItem ID, title, price, availability, seller name, reviews 5

What do you need, and how long does it take?

You need edit access to your product templates, a product feed export from your shop platform and a Merchant Center account. Fixing one best-selling product page by hand takes about two hours; that is our estimate. A catalogue is a template job: fix the template once, then spot-check ten products.

Start with the ten products that earn the most money. If their pages, markup and feeds agree, you can roll the pattern out to the rest.

What should the product page itself say?

Put the buying facts in plain HTML text. Microsoft's AI-search guidance warns that AI systems may not render hidden content such as tabs or expandable menus, and advises against putting key information only in images or PDFs 9. A spec table that opens only on click, or a size chart that is a photo, may be skipped.

Then add what a copy of the manufacturer sheet cannot give. Google's guide says non-commodity content, meaning unique expert or experienced takes, matters more in the long run than content based on common knowledge 8. For a product, that means your own measurements, test notes, comparison with a named alternative and honest limits.

Weak versus strong product description (illustrative)
  • Weak: 'Premium daypack for every adventure. Durable, comfortable, stylish.'
  • Strong: '28 litre daypack, 650 g, fits a 15-inch laptop. Water-resistant fabric; the rain cover is included. Not suitable for loads above 8 kg because the frame is a foam sheet. Ships in 2 days; returns accepted within 30 days.'
  • Why it is better: it states size, weight, fit, a limit and the terms in plain sentences that can be quoted. Replace every value with your own.

Which Product schema fields matter?

For merchant listings, Google requires name, image and an offer with a price and currency. It recommends brand, a GTIN (the barcode number), ratings, reviews, shipping details, a return policy and availability 2. Product markup supports pages focused on a single product or variants of the same product, so do not mark up a category page as one product 2.

Keep the markup in step with the page. Google's product documentation says both structured data and a feed help it verify your data 1. If the page price changes at midnight, the code must change with it, which is why a template that prints both from the same database field is safer than hand-edited values.

How do Merchant Center and the ChatGPT feed fit in?

Merchant Center requires an ID, title, description, link, image link, price and availability, and a brand for most new products. Google says to match the title and availability to your landing page, and titles can run to 150 characters without promotional text such as free shipping 3.

OpenAI's feed spec asks for an item ID, title, description, URL, brand, seller name, image URL, availability and price, written like 79.99 USD. Search eligibility defaults to true, but the spec says eligibility does not guarantee display 5. Checkout is separate: OpenAI states it is not the merchant of record, and payment runs on the merchant's own systems 4.

How should you handle reviews and ratings?

Use only real reviews, and show them on the page. The US Federal Trade Commission's consumer reviews rule took effect on 21 October 2024. It bars fake reviews, buying reviews that must express a set sentiment, and review suppression through threats, and courts can impose civil penalties for knowing violations 7.

Technical rules apply too. Google says a reviewer name in markup must be a valid Person or Team 2. In the OpenAI feed, reviews should describe the same product or variant as the row 5. Do not copy a rating from one colour variant onto another product.

What goes wrong most often?

The usual failures are plain inconsistency. The page shows a sale price while the feed holds the old one. The feed says in stock while the page says sold out. Variants share one URL, so a feed row points to the wrong colour. Or the product title in the feed is stuffed with extra words that the page title does not carry.

A second group is hiding facts. Specs sit in an image, shipping terms sit in a PDF, or the return policy loads only after a click. A third is thin copy: the same supplier paragraph used by fifty other shops.

How do you check it worked?

Check the machine side first. Run a sample of live product URLs through the Schema Markup Validator, then compare the price and stock on the page with your feed for the same ten products. Any mismatch is a bug to fix before anything else.

Then watch the result. Google's Search Console has a Generative AI performance report that shows impressions in AI Overviews and AI Mode by page, country and device 10. It shows impressions, not clicks. For ChatGPT, ask real buying questions in a fresh chat each month and log whether your product appears, using how to set up an AI visibility tracking spreadsheet.

When is this not worth the effort?

Skip the feed work if you sell only through a marketplace that controls the listing, if your catalogue changes hourly and you cannot automate updates, or if your products are identical to every competitor's. In the last case, first create something that is yours: testing, sizing guidance or a comparison. For related tracks see how to get recommended by ChatGPT and how to write comparison pages that get cited.

Frequently asked questions

Do I need a Merchant Center account to appear in Google AI Mode shopping?

Google does not say it is mandatory in the sources we read, but it says supplying both structured data and a Merchant Center feed maximizes eligibility 1. Its generative AI guide also tells ecommerce sites to keep Merchant Center details current 8. For a shop, setting it up is a low-risk step.

Does Product schema make ChatGPT recommend my product?

We found no source that says so. OpenAI documents a product feed as the way to share structured data with ChatGPT 4. Schema is still worth adding for Google and for clean data, but treat it as accuracy work, not a guaranteed route into ChatGPT results.

Can I pay to appear in ChatGPT shopping results?

OpenAI's feed spec includes an ads eligibility flag that is off unless a feed-level ads default applies 5. We could not read OpenAI's announcement pages, so we cannot say how ads and organic results are shown. Check OpenAI's current policy before assuming either way.

How often should I update my product feed?

Update it whenever price or stock changes. OpenAI's spec says to submit the current price and update the feed when a sale starts or ends, and that uploaded dates do not schedule changes 5. Google says more than 2 billion listings refresh every hour, so stale data stands out 6.

About the author

Rahil Jain

Rahil Jain is an AI SEO (answer engine and generative engine optimisation) consultant and co-founder of Chalk Labs.

LinkedIn · Method and credentials

Sources

  1. Introduction to Product structured data · Google Search CentralOpened October 20, 2026. Supports: Product snippets vs merchant listings; providing both structured data and a Merchant Center feed maximizes eligibility and helps Google verify data. Page last updated 2025-12-10.
  2. Merchant listing (Product) structured data · Google Search CentralOpened October 20, 2026. Supports: Required properties name, image and offers with price and currency; recommended brand, GTIN, ratings, reviews, shipping details, return policy, availability; single product page rule; reviewer name must be a valid Person or Team.
  3. Product data specification · Google Merchant Center HelpOpened October 20, 2026. Supports: Required attributes id, title, description, link, image_link, price, availability, brand for new products; 150-character title limit; title and availability must match the landing page.
  4. Key concepts (Agentic Commerce Protocol) · OpenAI DevelopersOpened October 20, 2026. Supports: Merchants share structured product data so ChatGPT can accurately surface products; OpenAI is not the merchant of record; checkout is separate and processed on merchant systems.
  5. Product feed spec · OpenAI DevelopersOpened October 20, 2026. Supports: Required feed fields and formats; availability values; price format; search eligibility defaults to true and does not guarantee display; update feed when price changes; reviews must match the row's product; ads eligibility flag.
  6. Shopping in AI Mode: new ways to shop with virtual try-on and agentic checkout · Google (The Keyword)Opened October 20, 2026. Supports: Shopping Graph has more than 50 billion product listings; more than 2 billion refreshed every hour (post dated 20 May 2025).
  7. Consumer Reviews and Testimonials Rule: Questions and Answers · US Federal Trade CommissionOpened October 20, 2026. Supports: Rule went into effect 21 October 2024; prohibits fake reviews, sentiment-conditioned review purchases, review suppression; courts may impose civil penalties for knowing violations.
  8. Optimizing your website for generative AI features on Google Search · Google Search CentralOpened October 20, 2026. Supports: Non-commodity content with unique expert or experienced takes; maintain Merchant Center and Business Profile details; structured data not required for generative AI search. Page last updated 2026-07-10.
  9. Optimizing Your Content for Inclusion in AI Search Answers · Microsoft Advertising (Bing)Opened October 20, 2026. Supports: AI systems may not render hidden content such as tabs or expandable menus; avoid key information only in images or PDFs. Post dated 8 October 2025.
  10. Generative AI performance report · Google Search Console HelpOpened October 20, 2026. Supports: Report shows impressions in AI Overviews and AI Mode by page, country, date and device; rolled out to all sites as of 31 August 2026.
How this page was made. Built on Rahil Jain's AI SEO method and drafted with AI research assistance. Every statistic links to the page we read it on (opened October 20, 2026 or nearby). Published October 20, 2026; last updated October 20, 2026. Chalk Labs sells AI SEO services, so we have a commercial interest in this topic. Spotted an error? Tell us and we will fix it.