A product page has a harder job in AI search than an article does. An article needs to be crawlable, indexable and quotable. A product page needs all of that plus a current price, a stated availability, a return policy and a shipping rate, and it needs them in a form a machine can read without visiting the page at all. That second channel is what most catalogues have not built, and it is why generative engine optimisation for ecommerce looks less like content work and more like data work.
Key Highlights
- Two channels feed the same result: Google states that providing both structured data on web pages and a Merchant Center feed maximises your eligibility, so markup alone leaves value on the table.
- AI features need no special schema, but shopping surfaces do, and the two requirements are often confused into one.
- The Shopping Graph holds more than 50 billion product listings, two billion of which are updated every hour, which sets the standard for how fresh a feed has to be.
- ChatGPT selects products independently and labels them as not ads, ranking merchants on availability, price, quality and whether the merchant is the maker or primary seller.
- India is in scope for some surfaces and not others: the shopping knowledge panel includes India, while popular product carousels remain United States mobile only.
What Makes a Product Page Different in AI Search?
Editorial pages are judged on what they say. Product pages are judged on what they assert, and every assertion is checkable. Price, stock, delivery window, return window, size, material, warranty: each is a fact that either matches reality or does not, and an engine that quotes a wrong one damages its own answer. That is the whole reason AI product discovery leans on feeds rather than on prose.
So the work splits in two. One half is ordinary answer engine optimisation, making the page reachable, renderable and quotable. The other half is supplying the same facts as structured data, as a merchant feed, or as both, so that a shopping surface can compare your item against forty others without rendering a single page.
Google’s documentation is unusually direct about wanting both. Its introduction to product structured data says you can use web page markup, a Merchant Center feed, or both, and that providing both maximises your eligibility. It also notes that some experiences combine the two, with product snippets drawing pricing from a merchant feed when the page itself does not supply it.
Does AI Search Need Special Product Markup?
No, and the confusion here costs budgets. Google’s guidance on AI features states that pages are eligible if they are indexed and snippet-eligible, and that there are no additional technical requirements. There is no AI-specific schema type, no separate file to publish and no markup that makes a page more quotable in an AI answer. Any AEO agency selling one is selling a fiction.
What markup does is different. It decides how your product is represented in the shopping experiences that sit beside and inside AI surfaces, which is a question of representation rather than of eligibility. Miss it and you can still be summarised. You simply will not appear with a price, a star rating and a delivery estimate where a comparison is being drawn, which is the moment that converts. AI Overviews optimisation follows the same split, which is why it belongs inside one AI SEO strategy rather than sitting as a line item of its own.
That distinction is the single most useful thing a team can hold onto when briefing AEO services or GEO services. Eligibility is earned by being crawlable and useful. Representation is earned by supplying clean data. The same argument settles most of the GEO vs SEO confusion for ecommerce, because neither half is new and only the consumption has changed.
Merchant Listing or Product Snippet?
Choosing the wrong one is the most common markup error on Indian ecommerce sites, and it is usually an inherited theme setting rather than a decision.
| Merchant listing | Product snippet | |
| Use when | A shopper can buy the product from you on that page | The page reviews a product sold elsewhere |
| Offer type | Requires Offer, not AggregateOffer | Accepts Offer or AggregateOffer |
| Supports | Shipping, returns, apparel sizing, member and strikethrough pricing | Richer review detail, including pros and cons |
| Can appear in | Shopping knowledge panel, Google Images, popular products, product snippets | Product snippets |
| Common mistake | Using it on a page that links out to other retailers | Using it on your own purchasable page and losing shipping and returns |
Three technical rules sit underneath the choice, and any AEO agency worth retaining will check all three in the first week. Each product variant needs its own URL, and variant markup tells Google which items are versions of the same parent. Each currency needs its own URL, which catches Indian sellers running multi-currency storefronts on one page. And a page should focus on one product or variants of the same product, which is why a category page marked up as a product rarely earns anything.
Return policies and loyalty markup should sit under your Organization markup, not be repeated on every product. Add a product-level policy only when it needs to override the merchant-level one. This makes everything easier to maintain and reduces the chance of information becoming outdated. That matters because feed-reading platforms can penalise inconsistent or stale data.
Which Product Attributes Carry the Most Weight?
The two major systems want the same facts in different envelopes, which makes an attribute audit the highest-return task in AI SEO optimisation for a catalogue.
| Attribute | Google requirement | ChatGPT feed requirement |
| Title and description | name required; description recommended | Required |
| Images | image required, crawlable, several aspect ratios | Required |
| Price and currency | Inside offers; separate URL per currency | Required, with per-field limits |
| Availability | Inside offers | Required |
| Unique identifier | GTIN, MPN or ISBN recommended | Offer ID must be unique in the feed |
| Shipping | shippingDetails with rate, destination and delivery time | Required where applicable |
| Returns | MerchantReturnPolicy under Organization | Seller policy links required when checkout is enabled |
| Apparel specifics | Colour, size, age group, gender | Colour, size, size system and gender recommended |
| Ratings | aggregateRating recommended | Reviews recommended, improving relevance |
Google’s free listings requirements add two country-specific conditions worth knowing in India, and both bear directly on AI product discovery. Shipping settings or the shipping attribute are required for Indian sellers, and return policy information is expected. The same documentation is explicit that free listings help your product pages appear, not your homepage, your policy pages or your category pages, which is a useful corrective for anyone measuring AI search visibility at domain level.
What Does Google’s Shopping Graph Actually Read?
The scale of the Shopping Graph explains why fresh data matters. Google says it contains more than 50 billion product listings, with two billion updated every hour. It powers shopping results in AI Mode and the Gemini app. Against that kind of update frequency, a daily feed might seem perfectly reasonable at first, but it can quickly leave product information out of date.
Surface availability is uneven and worth checking before a target is agreed, because an AI SEO strategy aimed at a surface your market cannot see will report failure for a year. Rich results carrying price, availability and reviews are global on mobile and desktop. The shopping knowledge panel is limited to a named list of countries that includes India. Popular product carousels remain mobile and United States only. Google Images and Lens product results are global. Gemini appears in the same list of places free listings can show, without further detail.
One technical warning in Google’s merchant listing documentation deserves more attention than it gets. The page recommends putting markup in the initial HTML and warns that markup generated by JavaScript can make Shopping crawls less frequent. For a React or Vue storefront that renders product data client-side, that single line explains a great deal of missing representation, and no amount of AI powered SEO elsewhere compensates for it.
How Does ChatGPT Pick Products?
Differently, and on stated record, which makes it the clearest published account of AI product discovery available to a merchant. OpenAI’s help documentation says product results are selected independently by ChatGPT, that they are not ads, and that they are not influenced by OpenAI partnerships, with advertising presented separately. Selection draws on structured metadata from first-party and third-party providers, on merchant and product metadata, and on data sent directly by merchants.
Its ranking factors are named, which is more than most briefs from an AEO agency will tell you: availability, price, quality, and whether the merchant is the maker or the primary seller. That last one matters for Indian brands competing with marketplace resellers of their own products, because being the manufacturer is a stated advantage rather than a hopeful one.
Two mechanics follow. Shopify merchants are already integrated through Shopify Catalog, with the documentation stating that no additional work is required from individual merchants, which quietly covers a large share of Indian direct-to-consumer stores. Everyone else supplies a product feed, accepted as TSV, CSV, XML or JSON, pushed to an allow-listed endpoint, with updates accepted as often as every fifteen minutes. The system also generates its own simplified titles, feature labels and review summaries, and OpenAI notes these are not verified, which is a reason to make your attribute data unambiguous rather than decorative. Treating that as a branch of generative AI optimisation rather than as a developer chore is the shift most teams have yet to make.
Where Does Agentic Checkout Change the Job?
Purchase is moving inside the assistant, which changes what a product page is for. OpenAI’s Instant Checkout runs on the Agentic Commerce Protocol, with three flows a merchant must implement: feed sharing, orders and checkout, and payments. The merchant validates the order, calculates tax, assesses risk and charges through its own payment service provider, with Stripe’s Shared Payment Token the first compatible implementation. Building on the protocol is open to all; Instant Checkout itself is available to approved partners.
Google is on a parallel path. Its November 2025 announcement described agentic checkout buying an item on the merchant’s site through Google Pay once a price alert is met, rolling out for eligible United States merchants with named partners and asking the user’s permission before each purchase. Both companies are building towards the same thing from different ends, one from the feed and one from the payment rail.
For an Indian seller the near-term consequence is not integration but preparation. The data layer that feeds an agentic purchase is the same data layer that feeds a recommendation, so the attributes driving AI product discovery today are the ones a checkout integration will need later, and the work has one payoff now and a second one in reserve. That is the strongest argument for treating AI SEO services and feed management as one brief rather than two.
What Breaks Product Page Eligibility Most Often?
- Markup rendered only by JavaScript: Documented as making Shopping crawls less frequent, and the most expensive five-minute fix in most catalogues.
- Markup that disagrees with the visible page: A price in schema that differs from the price on screen is a structured data violation and a trust problem at once.
- Stale availability: An out-of-stock item presented as available is the fastest route to being dropped from a comparison, whatever the AI SEO strategy says.
- Variants collapsed onto one URL: Each variant needs its own URL for the markup to resolve, and multi-currency pages need one URL per currency.
- Category pages marked as products: They earn nothing, and free listings are documented as covering product pages rather than category pages.
- Snippet restrictions inherited from an old build: A nosnippet or a tight max-snippet left over from a scraping scare removes the page from quotable surfaces entirely.
- No return or shipping data: Required for Indian free listings and visible in comparisons, so an absence reads as a worse offer.
How Should Product Copy Be Written for AI Recommendations?
Write for extraction, not for atmosphere. A recommendation engine comparing six water purifiers needs the throughput, the filtration stages, the recommended input water quality, the annual consumable cost and the service footprint, and will do nothing useful with a sentence about how handsome it looks on a counter. Specifics are what get quoted, which is the practical difference between AEO marketing copy and brand copy.
Answer the buying question on the page. Why this model rather than the next one up, who it suits, what it does not suit, what is in the box, how long installation takes. Pages that resolve a decision are the ones a model can cite without hedging, and that is as true for LLM SEO as it was for featured snippets. Comparison tables earn their place here, particularly across your own range, because they give a machine the structure it would otherwise have to infer.
Keep claims sourced and reviews real. Ratings markup must reflect genuine reviews, and because assistants generate their own review summaries from whatever they can read, thin or inflated review bodies produce summaries you would not have written. A GEO company that offers to manufacture mentions or reviews is offering a liability; Google notes in its own guidance that seeking inauthentic mentions is not as helpful as it might seem.
How Should a 60-Day Catalogue Fix Be Sequenced?
- Days 1 to 10, audit the data layer: Check markup type per template, whether it renders server-side, variant and currency URL structure, and snippet directives. Fix the render path first.
- Days 11 to 25, complete the attributes: Identifiers, images at multiple aspect ratios, shipping, returns, and apparel fields where they apply, aligned between markup and feed so the two never disagree.
- Days 26 to 40, open the second channel: Set up or clean the Merchant Center feed, confirm shipping settings for India, and decide the ChatGPT feed route, which is nothing for Shopify stores and an application plus a feed for everyone else.
- Days 41 to 55, rewrite the top pages: The forty products that carry the revenue, rebuilt as decision-resolving pages with real specifications and a comparison against your own range.
- Days 56 to 60, baseline the measurement: Record where you stand before anything moves, because retrofitting a baseline is impossible and most reporting arguments are really baseline arguments.
How Do You Measure It?
Treat representation and demand as separate numbers. Merchant Center reporting tells you what Google accepted and what it rejected, and a rejected feed item is a representation failure that no ranking report will surface. Search Console carries AI Mode and AI Overviews activity inside the Web search type rather than in its own breakdown, so read movement there as directional.
Server logs answer the question nothing else can. They show which AI crawlers arrived, what they received and whether a change to access or rendering actually took effect, and for a catalogue they show whether crawl attention is reaching the pages that matter or being spent on faceted noise.
Then measure demand with a fixed prompt panel. Twenty buying questions, run monthly across the surfaces your customers use, scored on whether your product appears, whether the attributes quoted are correct and which competitor is named instead. For India, remember that AI Mode has been generally available since 8 July 2025 in English, so a Hindi or Marathi prompt panel is testing a different and less settled surface. Sessions alone will understate all of this, which is the recurring flaw in judging AI search performance on a traffic graph.
How Does Savit Approach Product Page Optimisation?
At Savit Interactive, we start an ecommerce engagement with the data layer rather than the content calendar, because a catalogue whose markup renders client-side or whose feed is rejected cannot benefit from better copy. Our sequence is the one above: audit, complete the attributes, open the second channel, then rewrite the pages that carry the revenue. Two decades of search work from Mumbai across apparel, ecommerce, healthcare, education, logistics and manufacturing have taught us that the catalogues which win comparisons are the ones whose data is boring and correct. Brands weighing up an AI SEO agency should ask how it audits feed rejections and server logs before asking how many blogs it writes a month, and the same question separates a serious AEO agency from a reseller of templates.
What Should Change First?
Audit how your product markup renders before changing a word of copy. If it arrives only after JavaScript runs, fix that, then complete the attributes, then open the feed channel, then rewrite the pages that carry the revenue. That order is cheap and it is the difference between a catalogue that gets compared and one that gets summarised from someone else’s data.
Our view at Savit Interactive is that AI product discovery rewards the least glamorous work in ecommerce. The brands winning recommendations are not the ones publishing the most product content. They are the ones whose price, stock, shipping and returns are correct in two places at once, every day, and whose product pages answer the question a buyer actually asked.
