GEO for Ecommerce: How to Get Your Products Recommended by AI

GEO for ecommerce is the work of making your products easy for AI engines such as ChatGPT, Google AI Mode, Gemini, Perplexity and Copilot to find, verify and recommend when a shopper asks for something. For an ecommerce site, that comes down to seven jobs: a complete product record, matching schema.org markup, feeds sent to the engines that accept them, crawler access, third-party citations, comparison and FAQ content, and fresh price and stock data. Then you measure it, which is still the weakest part.

This guide takes each job in turn and says what to do on the merchant side. It assumes you already sell online and have some SEO in place. If you want the background on why discovery is moving into AI answers, start with how AI search is replacing ecommerce SEO.

What GEO for ecommerce actually means

The term comes from a research paper. Aggarwal et al., "GEO: Generative Engine Optimization", written at Princeton and IIT Delhi and accepted at KDD 2024, tested how rewriting a web page changes how often a generative engine uses it in an answer. Across a benchmark of 10,000 queries, the best methods lifted visibility by up to about 40%. Adding quotations, adding statistics and citing sources worked best. Keyword stuffing did worse than doing nothing.

Two caveats matter for a store. The benchmark queries were mostly informational, not shopping requests, so the paper tells you how to write a buying guide, not how to get a product into a shopping carousel. And product answers do not come only from page text. Google says its AI Mode shopping answers draw on the Shopping Graph, which it put at more than 50 billion listings in May 2025. ChatGPT reads merchant feeds and Shopify's catalog. Those are structured records, not prose.

So GEO for an ecommerce website splits into two halves: the record you control, and the evidence other people publish about you. An AI engine weighs both before it names a product.

Step 1: Make the product record complete

An AI shopping answer is a filter. The shopper says "waterproof hiking boots, wide fit, under $180", and the engine drops every product that cannot prove waterproof, wide and under $180. A product with an empty width field is not ranked lower. It is left out.

Start with the attributes a shopper uses to rule products in or out in your category: dimensions, material, compatibility, fit, capacity, certifications. Fill them in the same units and spellings across the catalog. Add a GTIN wherever the product has one, because identifiers let an engine match your offer to reviews and listings of the same item elsewhere. Group variants under one parent so the engine sees one product in five colours, not five unrelated products. Our guide to product attributes and structured data covers the field-by-field work.

Google added a set of optional "conversational attributes" to Merchant Center for exactly this kind of question. As of October 2026 the help page lists six: question_and_answer, document_link, related_product, item_group_title, variant_option and popularity_rank. They do not affect approval, and Google recommends sending them as a supplemental data source. The question and answer attribute takes up to 30 pairs per product and is meant for AI Mode. Use it for the questions your support team actually gets: does it fit, does it work with, can it be washed.

Step 2: Mark up product pages with schema.org Product and Offer

Your product page is the record a crawler reads when it visits. Google's merchant listing documentation requires a Product with name, image and an Offer that carries a price and currency. It recommends gtin, brand, sku, aggregateRating, review, availability, shippingDetails and hasMerchantReturnPolicy. Ship all of them if you have the data.

Do not expect markup to work as a secret lever. Google's guidance on AI features in Search, last updated in December 2025, says there are no extra requirements for AI Overviews or AI Mode, and no need for special schema or new AI text files. What markup does is remove ambiguity: the price on the page, in the markup and in the feed should be the same number. When they disagree, an engine has a reason to distrust all three.

Step 3: Send feeds where the engines read them

Feeds are the most direct line into AI shopping answers, and each engine has its own door.

• Google. Products in Merchant Center feed the Shopping Graph that AI Mode, AI Overviews and Gemini shop from. Organic free listings are eligible; ads are a separate decision.
• ChatGPT. OpenAI publishes a product feed specification with required fields such as item_id, title, description, url, brand, image_url, availability and price, in JSONL, CSV or TSV. Shopify stores are covered through Shopify's catalog; other merchants apply. The full route, including what changed after checkout moved back to merchant sites, is in how to get your products into ChatGPT shopping.
• Copilot. Microsoft's agentic commerce page tells merchants to start in Microsoft Merchant Center, add return and support policies, and make the feed ready for Google's Universal Commerce Protocol. Copilot Checkout is US, English and USD only.
• Perplexity. Perplexity runs a free program for retailers at perplexity.ai/merchants. Third-party guides disagree on how selective it is, so apply and see.

Treat one master feed as the source and generate each channel's version from it. Separate hand-edited feeds drift apart within weeks. For the field-level work, see product feed optimization for AI agents.

Step 4: Let the right crawlers in

AI engines use several bots, and they do different jobs. Blocking the wrong one removes you from answers without anyone noticing. As of October 2026:

• OAI-SearchBot finds pages for ChatGPT search. OpenAI's bot documentation says sites that block it will not be shown in ChatGPT search answers.
• GPTBot collects content that may train OpenAI models. Blocking it does not remove you from ChatGPT search.
• ChatGPT-User visits a page when a user asks ChatGPT to; OpenAI says robots.txt may not apply to it.
• PerplexityBot indexes for Perplexity search and, per Perplexity's docs, is not used for training. Perplexity-User fetches pages for a live question and generally ignores robots.txt.
• Googlebot is the crawler behind AI Overviews and AI Mode; there is no separate AI Mode bot.
• Google-Extended is not a crawler but a robots.txt token. Per Google's crawler docs, it controls whether your content trains Gemini models and is used for grounding in Gemini, and it does not affect inclusion or ranking in Google Search.

A sensible default for a store that wants to be recommended: allow OAI-SearchBot, PerplexityBot and Googlebot, and decide on GPTBot and Google-Extended as a separate content-licensing question. Then check your CDN, not only robots.txt. Cloudflare began blocking known AI crawlers by default for new domains in July 2025, and a bot rule in a firewall overrides whatever your robots.txt says.

Step 5: Earn third-party citations

The record proves what the product is. Third-party sources prove that people who bought it agree. AI answers lean heavily on them. A March 2026 study by Peec AI, a vendor of visibility software, counted 30 million citations in US answers from ChatGPT, AI Mode, Gemini, Perplexity and AI Overviews: Reddit, YouTube, LinkedIn, Wikipedia and Forbes led overall, and each engine had its own favourites. Treat it as vendor research, but the direction is consistent with other studies.

What a merchant can do about it:

• Reviews. Collect reviews on your own site and on the third-party sites your category uses, and mark up the ones on your site. Never buy or fake them: the FTC's rule on consumer reviews has allowed civil penalties since October 2024.
• Editorial lists. Find the "best X for Y" articles that AI answers cite for your category and send those editors product to test. One honest review on a site that gets cited is worth more than ten on sites that do not.
• Reddit and forums. Answer questions in the communities where your buyers ask them, under a disclosed brand account. Threads where real owners compare products are exactly what engines quote.
• Video. Product demos and independent reviews on YouTube carry facts about fit and use that text often leaves out.

An AI engine will not recommend a product it cannot verify, and it verifies in two places: your data and other people's words.

Kris Krokos

Step 6: Publish comparison and FAQ content the engines can quote

Shoppers ask AI engines comparison questions: "X vs Y", "best for flat feet", "does it work with". Pages that answer those questions directly give the engine text it can lift. This is where the GEO paper's findings apply: specific numbers, attributed quotes and cited sources made pages more likely to be used.

Good candidates on an ecommerce site are buying guides per use case, honest comparisons with named alternatives (including the cases where the other product is the better pick), size and compatibility pages, and FAQ blocks on product pages. Keep the facts in these pages identical to the feed. A guide that says 1.2 kg while the feed says 1.4 kg teaches the engine that your data cannot be trusted.

Step 7: Keep price and stock fresh

AI shopping answers are only as current as the data behind them. Google said in May 2025 that more than 2 billion Shopping Graph listings are refreshed every hour. OpenAI's feed spec asks merchants to update the feed whenever price, sale status or availability changes. An engine that sends a shopper to an out-of-stock page, or quotes a price you no longer charge, has every reason to prefer a competitor whose data matches the page. Sync feeds at least daily, and on every price change if your platform allows it.

How to measure GEO for an ecommerce site

Measurement is still patchy, so combine sources.

• Google Merchant Center.AI performance insights show share of voice against your Merchant Center competitors, top terms, popular attributes your data is missing, and how many of your products show up. As of October 2026 it covers English queries in the US, Canada, Australia, India and New Zealand, organic traffic only.
• Bing Webmaster Tools. The AI Performance report, in public preview since February 2026, counts how often your pages are cited in Copilot and Bing's AI summaries.
• Search Console. AI Overview and AI Mode traffic is counted inside the normal Web search type, not broken out.
• Analytics. Segment referrals from chatgpt.com, perplexity.ai, copilot.microsoft.com, copilot.com and gemini.google.com. ChatGPT tags links with utm_source=chatgpt.com, but a tagging bug in mid-October 2026 briefly sent those visits to Direct, Search Engine Roundtable reported, so watch for sudden drops that are not real.
• A prompt panel. Write 30 to 50 shopper prompts for your category, run them monthly in each engine, and log which products are named and which sources are cited. Our comparison of AI visibility tools for ecommerce covers the paid trackers and a free manual version.

More on the shift behind all of this is collected under AI search and discovery.

FAQ

What is GEO in ecommerce?

GEO (generative engine optimization) in ecommerce is the practice of making products easy for AI engines to find, verify and recommend. It covers product data, feeds, markup, crawler access, third-party reviews and citations, and content that answers shopper questions directly.

Is GEO different from SEO for an ecommerce site?

They overlap heavily: crawlable pages, good content and authority help both. GEO adds feeds to AI engines, attribute completeness, third-party citations and new ways of measuring. Our guide to GEO vs SEO for ecommerce compares them side by side.

Does schema markup help AI recommend my products?

It helps an engine read your price, stock, ratings and identifiers without guessing, and it must match your feed. Google says no special markup is required for its AI features, so markup supports eligibility rather than guaranteeing a recommendation.

Should I block GPTBot to protect my content?

You can block GPTBot without leaving ChatGPT search, because search uses OAI-SearchBot. Blocking OAI-SearchBot is what removes your pages from ChatGPT search answers.

How long does GEO take to show results?

Crawler and feed changes can show within days; OpenAI and Perplexity say robots.txt changes take up to about 24 hours. Third-party citations and reviews build over months. Track a fixed prompt panel monthly so you can see the trend.