Almost everything written about showing up in AI is built for people who sell services or publish articles: be a source, earn the citation. If you sell products, the game is different. When someone asks ChatGPT or Perplexity for «the best X under $100», it doesn't cite you in a paragraph: it puts you — or doesn't — into a carousel of product cards with a photo, a price and a buy button. And that storefront already moves real money: per Adobe Analytics, traffic from AI assistants to U.S. retail sites grew 693% over the 2025 holiday season. This guide is the map for getting into it.
In ecommerce nobody cites you in text: you make the product carousel, or you don't
The difference from any other page is the surface. An assistant answering «how do I do X» cites you in the body; one answering «which X should I buy» returns cards: image, title, price, rating and, increasingly, a pay button. Perplexity shows three to five of those cards per shopping query. Your goal isn't to sound authoritative in a paragraph — it's for your product, with its price and stock correct, to be one of those three to five. Everything else here serves that.
And to make a card, the assistant needs data it can only read if you hand it over structured: exact name, current price, availability, rating. If that data lives only in the pretty HTML of your page, AI won't extract it reliably and drops you. So ecommerce GEO is, above all, a clean-data job — not a brilliant-prose one.
ChatGPT pulls most of its products from Google Shopping: your feed is asset number one
Here's the lever almost nobody works. When ChatGPT recommends products, it doesn't crawl your store at random: per Semrush's analysis, around 75% of the product data it shows comes from Google Shopping. It runs two parallel searches per answer — one contextual for the text, one encoded against Google Shopping to build the carousel. Translated to your work: your Google Merchant Center feed is, almost literally, your application to show up in ChatGPT. If your feed is incomplete, stale or disapproved, you're not in the deck no matter how beautiful your site is.
A feed that does make the deck carries, per product, at least four things, in order:
- Correct identifiers. GTIN, MPN and brand. Without them Google can't match your product to the query, and AI won't retrieve it.
- Real-time price and availability. The assistant drops what's out of stock and distrusts a price that doesn't match the page. If your stock changes and the feed doesn't, you vanish from the card.
- Specific titles and descriptions. Not «running shoe» but the model, the use and the attribute people search: «neutral-gait running shoe, 250 g». The AI query is long and specific; a generic title won't catch it.
- Clean image and complete attributes (size, color, material, weight). These are the filters the assistant uses to narrow «under $100, size 9, waterproof». What's missing doesn't filter; what doesn't filter doesn't show.
Perplexity lets you in for free: the Merchant Program and «Buy with Pro»
Perplexity did the opposite of putting up a tollgate: its Merchant Program is free, takes about five minutes to apply for, and directly improves your products' odds of showing up in its cards. On top of that it gives you a search-and-shopping-trends dashboard and API access. There's no reason not to be in: it's one of the few AI-visibility levers you switch on by filling out a form instead of writing for months.
Its headline piece is «Buy with Pro»: for Pro subscribers in the U.S., the user checks out inside the chat, without leaving for your site, with free shipping funded by Perplexity — not by you. For you that's a conversion trigger that doesn't touch your margin: the number-one cause of cart abandonment disappears. The trade-off is that checkout lives on Perplexity's surface, so you compete to be the chosen card, not on the design of your checkout page. Sign up and work to be that recommended product.
Product schema isn't optional: without it, AI can't read you
If the feed is your application on shopping surfaces, schema is what makes your own page legible when the assistant visits it directly. Mark every product page with Product structured data in JSON-LD — the format Google recommends and the one that doesn't break when you redesign the template — and include what AI needs to extract: name, brand, sku, gtin, offer (price, currency, availability) and aggregateRating. It's not cosmetic: in industry counts, 71% of the pages ChatGPT cites and 65% of those in Google's AI Mode carry structured data. Pages with no schema barely get cited, because AI can't reliably read price or rating from loose HTML.
With the feed and schema in place, the rest of the game is four things:
- Put reviews to work. The aggregateRating and, above all, text reviews feed the summary the assistant paints over your product. Ask for them systematically, mark them up with Review schema, and treat the negative ones as raw material, not a threat: a credible rating profile gets cited more than a perfect, suspicious one.
- Add real FAQ to the page. The questions a buyer actually asks («does it run large?», «does it work for X?»), marked up with schema, win you both the classic «People also ask» and the citation in the AI answer. Two surfaces, one job.
- Write specific pages, not template ones. Unique, concrete content keeps AI from collapsing your product into a competitor's. The manufacturer's copy-paste description is invisible; your own detail (what it's for, what it isn't, what it pairs with) is what's citable.
- Keep price and stock synced everywhere. Data that doesn't match across your site, your feed and your schema is the fastest way for the assistant to drop you as unreliable. Freshness here isn't a luxury: it's the entry threshold.
Measure it differently: not «am I cited?» but «do I make the cards for these purchases?»
Measuring ecommerce GEO isn't the same as for a services brand. You don't count mentions in text: you check presence in the storefront. Define 15 to 25 real shopping queries for your category — in the form people actually use: «best X for Y under $Z» — run them once a month in ChatGPT and Perplexity, and note three things per query: whether your product shows in the cards, in what position, and with what price and rating. The month-over-month trend is what counts, not one day's snapshot.
And check the boring stuff, where people fall down: that your Merchant Center feed is approved and error-free, that your schema validates without warnings, that price and stock are synced across site, feed and page, and that your robots.txt doesn't block the crawlers that fetch pages to answer. Almost all the ground lost in ecommerce isn't lost to bad strategy: it's lost to a disapproved feed or a stale price nobody checked.