October 5, 2026

GEO for ecommerce: what gets products into AI shopping answers

Products reach shoppers in AI assistants through two systems. Product cards come from product data: Google's AI Mode draws on its Shopping Graph, and since a July 2026 change, brands outside an integrated merchant feed aren't considered for about 65% of ChatGPT's product recommendations, according to tracking by Profound. The answer text, which names brands and explains trade-offs, comes from pages the assistant searches and cites, often retailer listings, editorial reviews, YouTube and Reddit. So GEO for a store is three jobs: complete product data, product pages that state facts plainly, and a presence on the third-party pages each assistant reads.

This assumes crawlers can reach your store and your feeds are live. If not, start with what AI assistants need from your store and product feeds for AI shopping.

Why product cards and cited answers are different problems

OpenAI's shopping help page says ChatGPT picks products using "structured metadata from first-party and third-party providers," such as price and description. It ranks merchants on factors like "availability, price, quality, and whether they are the maker or primary seller." Google's AI Mode shopping "brings together Gemini capabilities with our Shopping Graph," and Microsoft says its Merchant Center feeds "help inform organic Copilot results."

What shoppers see	Where it comes from	What you control
ChatGPT product cards	Integrated feeds and catalogs, plus web search	Shopify Catalog or an OpenAI feed
Google AI Mode listings	The Shopping Graph	Merchant Center feed and matching pages
Copilot shopping results	Informed by Microsoft Merchant Center feeds	Microsoft Merchant Center feed
Brands named in the answer	Pages the assistant searches and cites	Product pages, lists, reviews, Reddit, YouTube

Writing in Search Engine Land in March 2026, Peec AI's Tom Wells matched 83% of 43,000 ChatGPT carousel products to Google Shopping's top 40 results. Then on July 10, feed-integrated retrieval overtook web search, rising from 8.26% to 61.54%, in Profound's tracking of 1,757,723 prompt runs that month. That's outside measurement, not an OpenAI statement, but it fits OpenAI's March 2026 announcement that merchants share feeds "so their catalogs are fully represented in ChatGPT."

What the research says gets products cited

Retailers, editors and communities supply many of the citations in product answers, and the mix depends on the assistant.

Retailers in ChatGPT, video and forums in Google. In BrightEdge's December 2025 analysis of tens of thousands of holiday-season ecommerce prompts, ChatGPT cited retailers about 36% of the time and Google AI Overviews about 4% of the time. ChatGPT leaned on Amazon, Walmart, Target and other big retailers plus manufacturer pages. AI Overviews leaned on YouTube reviews, Reddit and Quora threads, and editorial sites such as Rtings and Consumer Reports. BrightEdge didn't publish an exact prompt count.
Independent reviews over brand sites. In a September 2025 paper, University of Toronto researchers ran 1,000 consumer ranking prompts through the API version of GPT-4o search. For US consumer electronics, 92.1% of its citations went to earned media, which the paper defines as independent media, review or comparison sites, and social content was negligible. Google's regular results for the same queries were 32.9% brand sites.
"Best" lists. In a December 2025 study, Ahrefs collected 26,283 source URLs behind ChatGPT's answers to 750 software, product and agency recommendation prompts. "Best X" lists were 43.8% of all sources, and 79.1% of the 1,100 lists it dated had been updated in 2025. When a product brand's own site was listed as ChatGPT's first recommendation, it was a product page in 87.2% of those prompts.
Reddit in AI Mode. In a November 2025 study, SE Ranking ran 50,000 technical-product keywords from New York. AI Mode showed shopping results for 61.66% of them, and Reddit appeared in 97.5% of cases where it showed discussion links.

So measure each assistant separately.

What to fix on product and category pages

When ChatGPT cites a brand's own site for a product prompt, it's usually a product page, and OpenAI says its shopping research mode works by "reading product pages directly." So put the facts there, along with the criteria shoppers judge by. In Google's example, AI Mode handles a request for bags suitable for a trip to Portland, Oregon in May by running several searches at once to work out what makes a bag good for rainy weather and long journeys.

A lab test built for stores agrees. In E-GEO, published in November 2025, Columbia and MIT researchers had GPT-4o re-rank Amazon listings for 7,151 shopping questions from Reddit. Ten of 15 rewrite styles made "negligible or even negative" difference. The best, stressing what makes a product better than others in its category, gained 0.71 places on average. A storytelling rewrite that swapped facts for narrative lost 4.03. Take the direction, not the numbers.

Open with what the product is, who it's for and the job it does, in your shoppers' words.
Put the specs that decide the purchase in an HTML table with units, matching your feed.
Add a plain "best for" and "not for" line.
Say how it compares with the obvious alternatives, including your own range.
Answer the questions support hears most. Google announced new Merchant Center attributes in January 2026 that include "answers to common product questions," so reuse the answers in your feed.
Show reviews as text on the page, with the count, the rating spread and the critical ones.
Keep price, stock, shipping and returns identical on the page, in the feed and in Merchant Center.
Open category pages with a short buying guide that names the criteria, such as waterproof rating, weight or fit, and compare your top products against them.

Off-site: where assistants read about your products

Retailer and marketplace listings

ChatGPT cites retailers heavily, and big ones gained after the July feed change: comparing Profound's July 7 to 9 and July 10 to 12 samples, the top 10 merchants' share of references rose from 22.5% to 41.8%. OpenAI ranks merchants partly on whether they're the maker or primary seller, so make that obvious: the same product names, specs, GTINs and images on your site, in your feeds and on every retailer listing.

Review platforms, lists and testing sites

ChatGPT summarizes reviews from public websites, and in SE Ranking's AI Mode sample, 88.83% of reviewed products scored 4.1 to 5.0 stars. Collect reviews where answers look: your product pages, cited retailer listings and any review platform your baseline turns up. Then check what each cited list and testing site says about you, send current specs and samples, and ask for corrections. Skip paid placements the list doesn't disclose.

Reddit and YouTube

Google's guide to generative AI search says its AI features show what's said about products "in blogs, videos, and forum discussions," and BrightEdge found YouTube and Reddit near the top of AI Overviews' ecommerce sources. Send products to YouTube reviewers with no conditions on what they say, and the FTC's influencer guidance puts their disclosure "in the video and not just in the description." On Reddit, answer the threads your baseline surfaces from a named account and heed Reddit's spam policy, which asks people who mostly post links to their own business to be "thoughtful about the frequency of posting."

Mentions versus links

Across 75,000 brands, Ahrefs measured a 0.664 correlation between branded web mentions and appearances in Google's AI Overviews, against 0.218 for backlinks (May 2025, updated April 2026). Correlation isn't causation, and Google says seeking inauthentic mentions "isn't as helpful as it might seem." Earn mentions that would exist anyway: reviews, press, gift guides and creator content.

How do you build a shopping prompt set and baseline?

Answers vary from run to run. In a SparkToro study published in January 2026, 600 volunteers ran 12 prompts, including chef's knives and headphones, 2,961 times through ChatGPT, Claude and Google's AI, with less than a 1 in 100 chance that ChatGPT or Google's AI gave the same list of brands in any two responses. In Profound's March 2026 data, about 95% of product titles appeared in fewer than 30% of runs of the same prompt. So skip rank tracking and measure appearance rate: the share of runs in which an assistant names or shows your product.

Prompt type	Example (illustration)	What it tells you
Category	best trail running shoes for wide feet	Whether you make the shortlist
Constraint	waterproof hiking boots under $150 that ship to the UK	Whether price, stock and shipping facts get through
Use case	what to pack for a week of wet-weather hiking	Whether criteria content leads to you
Comparison	[your brand] vs [competitor] trail shoes	How assistants describe your trade-offs
Brand fact	[your brand] return policy and warranty	Whether your facts are current

Write 40 prompts, eight per type, from site search, Search Console queries, support tickets and reviews, in shoppers' words.
Run each five times in ChatGPT, Google AI Mode, Gemini, Copilot and Perplexity, from the country you sell to, within one week. Turn off Memory and custom instructions, since OpenAI says ChatGPT's picks consider both.
Log per run whether cards appeared, whether yours was there and whose buy link it carried, whether the text named you and three competitors, and every cited URL.
Score appearance rate per assistant for you and each competitor, and group cited URLs by type: your site, retailer, editorial, review platform, Reddit, YouTube or other.
Rerun monthly with the same prompts and settings, keeping new prompts in a separate set.

A worked example with made-up numbers: 40 prompts run five times in ChatGPT is 200 runs. If you appear in 46, your appearance rate is 23%. If your main competitor appears in 88, theirs is 44%. If most of the top cited URLs are lists and threads that name them and not you, that's your off-site list.

Bing Webmaster Tools' AI Performance report, in public preview since February 2026, adds Copilot citations of your pages and the grounding queries behind them. For traffic and revenue, see tracking AI assistant traffic in GA4.

What doesn't work, or isn't proven

llms.txt. Google says Search ignores the file, so it "will neither harm nor help" your visibility there. SE Ranking's study of 300,000 domains, published in November 2025, found its model predicted AI citations better without the llms.txt variable.
Hidden text and prompt injection. Google's spam policies define hidden text as content placed "solely to manipulate search engines and not to be easily viewable by human visitors," which would cover instructions to AI in white text. Bing's guidelines now warn that attempts to add content designed to manipulate or interfere with language models used by Bing or Copilot "may result in reduced visibility or removal from search experiences," in a section Search Engine Journal covered in February 2026. OpenAI says shopping research avoids "low-quality or spammy sites."
Fake or incentivized reviews. The FTC's final rule bans fake reviews, including AI-generated ones, incentives conditioned on a particular sentiment, undisclosed insider reviews and review suppression, and lets the FTC seek civil penalties. Ask every buyer and tie no reward to the rating.
Scaled AI pages. Google's spam policies call using generative AI tools "to generate many pages without adding value for users" scaled content abuse, and in E-GEO most generic rewrites didn't help anyway.

A 90-day plan with owners

Days 1 to 10, marketing lead: run the baseline against three competitors, ending with appearance rates and the 20 most-cited third-party URLs.
Days 1 to 14, ecommerce manager: confirm your route into Shopify Catalog or an OpenAI feed, Google Merchant Center and Microsoft Merchant Center, and fix any price, stock or spec that differs between page and feed.
Days 7 to 21, customer service lead: turn six months of pre-purchase questions, chats and return reasons into the 30 questions that decide a purchase, with answers.
Days 15 to 45, content lead with the ecommerce manager: rewrite the top 20 product pages and top five category pages against the checklist. The developer confirms specs, reviews and answers are in the page HTML.
Days 21 to 60, content lead: publish a dated buying guide for each main category, built on the criteria your prompts surface, saying when a competitor fits better.
Days 15 to 75, founder or PR lead: work the cited-URL list, correcting facts with editors, sending products to reviewers and answering Reddit threads.
Days 30 to 60, ecommerce manager: match retailer and marketplace listings to your site on names, specs, identifiers and images.
Days 30 to 90, customer service lead: ask every buyer for a review after delivery, on your site and on retailer listings that show up in answers, and reply to critical ones.
Day 45, marketing lead: rerun the baseline and trace any wrong fact in an answer back to the page it came from.
Day 90, marketing lead with the founder: rerun the baseline, compare it with day one, check AI assistant revenue in GA4, and set next quarter's priorities.

If you want help

Our Agentic SEO audit costs $1,500 and takes two to three weeks. It checks how findable and recommendable your products are in ChatGPT, Google AI Mode, Gemini, Copilot and Perplexity. See how we work with DTC and ecommerce brands, or book a 30-minute call.

Sources

Facts checked on 4 October 2026.