GEO for E-commerce: Get Cited by ChatGPT & Perplexity

Why e-commerce brands can't ignore AI search

Shoppers are increasingly asking ChatGPT and Perplexity questions they used to type into Google: "best waterproof hiking boots under $150," "which robot vacuum handles pet hair." Nobody has published a reliable, engine-specific figure for how much shopping research now happens this way, but the direction is clear from platform usage alone. What matters more than the size of the shift is its shape: these tools don't return ten blue links. They return a short, specific answer with two or three named products.

That changes the metric that matters. Ranking on page one used to be the goal. Now the goal is narrower and higher-stakes: does the AI say your brand's name out loud, or does it recommend a competitor while you sit un-cited in the training data it never surfaced. There's no partial credit for "close."

How ChatGPT picks which products to recommend

ChatGPT's product recommendations aren't tied to a single pipeline. OpenAI publishes its own product feed spec, accepts feeds through the Agentic Commerce Protocol (ACP), pulls from Shopify's merchant catalog, and can draw on public product pages directly. Google Merchant Center is one possible input, not the officially stated primary source, so treating it as the only feed that matters is a good way to miss the other channels ChatGPT actually uses.

Here's the part that surprises people: one study of a 43,000-product ChatGPT carousel sample found that around 83% of the products it recommended also appeared in Google Shopping's top 40 organic listings. That's a correlation, not proof that ChatGPT is pulling straight from Merchant Center. The more useful read is that ChatGPT's picks and strong organic Shopping performance move together, so if your organic Shopping presence is weak, feed work alone probably won't fix that. Classic Shopping SEO (accurate GTINs, competitive pricing, strong CTR, review volume) is still doing a lot of the work upstream of the AI layer.

How Perplexity picks which products to recommend

Perplexity works differently. It doesn't require a merchant feed submission at all. PerplexityBot crawls and indexes pages the way a conventional search crawler does, and Perplexity-User can open a page live in response to a specific query. Perplexity hasn't publicly documented Product Schema as a dedicated ranking factor, and it hasn't confirmed that JSON-LD is prioritized over visible page content.

That doesn't make structured data pointless. Accurate Product and Offer markup can reduce ambiguity and help a system parsing your page interpret price, availability, ratings, shipping, and return information correctly, whether that system reads JSON-LD, visible HTML, or both. Treat schema as a way to remove doubt, not as a confirmed Perplexity ranking lever.

From AI product discovery to merchant-owned checkout

OpenAI first announced Instant Checkout on September 29, 2025, letting US ChatGPT users complete a purchase from Etsy and, later, Shopify sellers without leaving the chat, built on its Agentic Commerce Protocol (ACP). By March 24, 2026, OpenAI said it was moving away from that standalone in-chat checkout and refocusing ChatGPT on product discovery, with the purchase itself completing on the merchant's own site or app. That's the current, official framing: ChatGPT helps a shopper find and narrow down what to buy, then hands off to your store to close the sale.

The practical upshot doesn't change much for feed and schema work. Whether checkout happens in-chat or on your site, ChatGPT still needs accurate product data to surface your item in the first place, and a clean handoff (correct price, working link, current availability) is what keeps that discovery step from turning into an abandoned cart on your own domain.

Fields to include in an OpenAI product feed

These are feed fields, not on-page markup: this is data you submit through OpenAI's product feed spec, ACP, or Shopify's catalog integration, separate from anything a crawler parses off your live page. Getting this feed complete is what lets ChatGPT surface and describe your product accurately during discovery.

• Core identifiers and content: title, description, brand, GTIN/SKU/MPN, images, condition.
• Price and availability, kept current, since stale pricing is one of the fastest ways to get a product dropped from consideration.
• Reviews and Q&A, which OpenAI's feed spec supports as fields in their own right, not just a number bolted onto the listing.
• Return policy, spelled out at the feed level so "can I send it back" has a direct answer.
• Popularity and engagement signals (fields like a popularity score or return rate), plus video and 3D model links where you have them, since OpenAI's feed spec has room for richer product media that a plain Google Shopping feed doesn't ask for.
• Related products, so the assistant has a legitimate alternative to offer instead of substituting a competitor's item.

Fields to include in Product structured data

This is the separate layer: Schema.org markup embedded in your page's HTML (usually as JSON-LD), which any crawler, including PerplexityBot and OAI-SearchBot, can parse directly from the page itself, independent of whatever feed you submit elsewhere.

• Product and Offer, the two base types almost everything else nests inside.
• GTIN, SKU, and brand, matched exactly to what's on the page and in any feed you also submit.
• price and availability, inside the Offer object, kept in sync with what a shopper actually sees.
• AggregateRating, with reviewCount and ratingValue kept distinct. A bare average rating without a review count is a weaker trust signal than the same rating backed by a visible sample size.
• shippingDetails and hasMerchantReturnPolicy, which answer the "will it arrive in time and can I return it" questions directly in structured form.
• ProductGroup, if you sell variants, so size or color options are represented as one connected group instead of disconnected listings.

Reviews, Reddit, and other trust signals

AI models lean hard on external validation before recommending a product, which is exactly why Trustpilot reviews, Reddit threads, and independent comparison blogs carry weight that your own product description can't replicate. A page that only ever talks about itself is a weaker citation candidate than one backed by people who don't work for you.

One caution worth building into your strategy: these weightings shift without warning. In September 2025, OpenAI adjusted its retrieval weighting and Reddit's share of ChatGPT citations reportedly dropped from around 60% to about 10% within six weeks, with no announcement. If your entire trust strategy lived on one platform, that kind of change would have gutted your visibility overnight. Spread your review and mention footprint across several channels (Reddit, Trustpilot, category-specific forums, comparison sites) so no single algorithm update can zero you out.

The source mix also differs by engine. Some analyses of AI citation patterns found ChatGPT leaning heavily on Wikipedia (around 48% of citations in one study), while Perplexity favored Reddit-style community discussion (around 47%). Treat those exact numbers as directional rather than fixed, since they shift with every retrieval update, but the pattern itself is a reason not to build your entire trust strategy around one type of source.

GEO vs. traditional e-commerce SEO

The two aren't competing strategies. GEO for e-commerce currently sits on top of Shopping SEO, not instead of it. The 83% overlap between ChatGPT's picks and Google Shopping's top listings is a correlation, not proof of a direct pipeline, but it's still a strong signal that the two are tightly linked in practice.

How to get started: a practical checklist