How is AI changing product discovery?
The shopping query moved. Instead of scanning ten blue links, buyers ask ChatGPT, Gemini, Perplexity or Google AI Overviews comparative, situation-shaped questions: best espresso machine under five hundred, is DTC brand X legit, what do runners with knee problems actually buy. The engine returns a synthesized shortlist with reasons — and the reasons are assembled from product data, review corpora, comparison articles and forum sentiment. The numbers are stark: AI Overviews now serve 2.5 billion+ monthly users (Alphabet, Q2 2026); when an overview appears, position-one CTR drops roughly 58 percent (Ahrefs, February 2026). If your products are not inside the synthesized answer, a growing share of demand never reaches a page where you can compete.
The four surfaces engines read
- Product data quality. Merchant feeds, Product schema with real attributes (materials, sizing, compatibility), accurate pricing and availability. Engines answering which is best for X need attributes to match against; vague marketing copy gives them nothing.
- Review corpora. On-site reviews plus independent platforms. Texture matters: reviews mentioning the use case (wore them for a marathon, survived two toddlers) are the synthesis material for fit questions.
- Comparison and editorial content. Buying guides, versus pages, honest roundups — yours and third parties’. Engines lean on independent corroboration, so digital PR and creator coverage feed the same answers. See digital PR for GEO.
- Forum sentiment. Reddit-class discussions are heavily cited for is-X-worth-it questions; you cannot control them, but you can participate honestly and fix the product issues they surface. See the Reddit and Quora strategy.
What about marketplaces versus your own store?
Both matter and they interact. Engines frequently cite marketplace listings and marketplace review corpora when comparing products, so Amazon-class presence with complete attributes and healthy review texture corroborates your own site rather than cannibalizing it. The mistake is dependency: if the engine can only find your product facts on a marketplace, the answer links there. Publish full attribute data, honest comparisons and buying guides on your own domain so the citation — and the margin — can land with you.
What to do first
- Audit your top fifty products for attribute completeness and valid Product schema.
- Publish honest comparison content — including where competitors win; one-sided puffery is uncitable.
- Build review operations that surface use-case texture, never incentives.
- Verify Bing indexation and IndexNow — ChatGPT’s shopping answers lean on Bing.
- Measure monthly: a fixed set of category and brand questions scored on the four-dimension standard.
Where does paid fit?
Nowhere, currently — ChatGPT has no ad slots as of November 2026, and AI Overviews citations are earned, not bought. That makes the answer layer one of the few remaining channels where content quality and corroboration, not budget, decide presence. Paid still matters around the edges: branded search and shopping ads capture the demand the answer creates.
What GEO cannot do for e-commerce
It cannot make a mediocre product win is-it-worth-it answers — the review and forum corpus will out-vote your content. It cannot guarantee placement; engines are stochastic and no honest vendor promises slots. What it can do: ensure your real strengths are legible, structured and corroborated wherever the engines look. That is usually the difference between being named and being absent.