AI Search Optimization for Ecommerce: How Online Stores Get Recommended

AI engines find products three ways: live retrieval from the open web, merchant feeds, and what they learned in training or read about you elsewhere. Each engine mixes these differently, and the documented routes are narrower than most advice implies.

• ChatGPT: OpenAI describes the Agentic Commerce Protocol as an open standard that lets ChatGPT ingest structured catalog data and surface relevant products. OAI-SearchBot surfaces sites in ChatGPT search.
• Google AI Overviews and AI Mode: Google says no extra requirements or special optimizations apply beyond being indexed and snippet-eligible, and asks commerce sites to keep Merchant Center data current. In January 2026 it announced checkout inside AI Mode for eligible U.S. retailers.
• Perplexity: a free Merchant Program takes product catalogs from stores that sell and ship to the U.S. (as reported March 2026), and PerplexityBot fetches pages for search results.

Our position: a feed makes you eligible, but it does not give an engine a reason to name you over the next store selling a similar product. That reason comes from your pages and from independent sources that confirm them. Start with the page, then the feed.

They need price, availability, identifiers, specs and reviews as plain text on the page, repeated in matching JSON-LD. Google's merchant listing guidance requires price, currency and an Offer, and recommends GTINs, brand, ratings, reviews, shipping and return policy.

Google's structured data policies say markup must be a true representation of the page content, and Google notes it may verify merchant data before showing it. A schema price that differs from the visible price gives any engine reason to distrust both.

Checklist for each product template:

1. Product JSON-LD in the server-rendered HTML, with name, brand, description, a GTIN or MPN, and an Offer with current price, currency and availability.
2. Review and AggregateRating only for reviews shoppers can read on the same page. Never mark up invisible reviews.
3. Organization on the homepage, with sameAs links to your real profiles. In our crawler study, only 28 of 61 e-commerce homepages (45.9%) had sameAs links.
4. Specs in HTML text or tables, never only in an image. An engine cannot quote a JPEG.
5. One name per product everywhere: your site, marketplaces, press and schema.

Our structured data guide covers the markup, and the product page answer goes template by template.

Write each product and category page so it answers the question a shopper would type into an AI tool: who is this for, who should skip it, how does it differ from the obvious alternative, and what are the real specs. Plain facts get quoted. Brand mood does not.

Compare two descriptions of one boot:

• Vague: "Built for adventure. Premium quality, made to last."
• Quotable: "Waterproof leather, 1.1 kg per pair, runs half a size small, best for narrow feet, resoleable."

Only the second gives an engine something to repeat or verify. Apply the same rule across the page:

• Open with the answer. Fit, material and use case come first, before any brand voice.
• Name who it is not for. Honest exclusions make a recommendation believable.
• Publish buying guides with comparison tables for each category, covering competitors fairly. "Best X for Y" and "X vs Y" are the shapes AI shopping answers already take.
• Add a visible FAQ on sizing, returns, durability and compatibility, with matching FAQPage markup.

Earn mentions on sites you do not own: marketplace reviews, editorial roundups, forum threads and comparison articles. In our reading, engines are more willing to name a product that independent sources confirm. Your own star rating alone proves little.

Practical moves, in order of effort:

1. Collect detailed reviews on third-party platforms, not just your site, using consistent product names.
2. Answer questions honestly where your buyers already ask them, such as relevant subreddits and communities. Our note on why AI engines keep citing Reddit explains the pattern and the limits.
3. Pitch category roundups with samples and real specs.
4. Fix what the record already says. If marketplace reviews contradict your on-site rating, engines can see both.

One caution from our data. The publishers who write roundups are the sites most likely to restrict AI crawlers: in AEOeye's crawler access study (October 2026), 27 of 38 news publishers blocked at least one AI retrieval bot, against 4 of 90 e-commerce and consumer brands. A roundup mention helps shoppers either way, but do not assume it reaches every AI answer. Test it.

Three faults do most of the damage: bot protection that treats AI retrieval bots as scrapers, product data that only appears after JavaScript runs, and robots.txt rules that block the wrong bots. In our study, the first was the common one among stores.

Bot walls. When our self-identifying research crawler fetched homepages, 24 of 88 reachable e-commerce and consumer homepages (27.3%) challenged it: 20 returned 403 and 4 returned 429. Cloudflare says verified bots are checked by methods such as published IP lists and reverse DNS, and customers can set AI bot policies.

OpenAI recommends allowing OAI-SearchBot and its published IP ranges. Check your CDN or WAF logs for those user agents. A challenge to our crawler is a warning sign, not proof a store was hidden from AI.

JavaScript-only content. In a December 2024 analysis, Vercel found that major AI crawlers, including OAI-SearchBot, ChatGPT-User, ClaudeBot and PerplexityBot, do not render JavaScript. Google also notes that not all bots can run JavaScript. Open a product page with Ctrl+U: if price, specs and review text are missing from the source, render them on the server. Late-loading review widgets are a common culprit.

Wrong robots.txt rules. Only 4 of 90 e-commerce and consumer sites blocked any AI retrieval bot, so this is rarer than people fear. If you object to training use, OpenAI says OAI-SearchBot and GPTBot are independent settings, so you can block GPTBot and keep ChatGPT search. Note that ChatGPT-User may not follow robots.txt, and Perplexity-User generally ignores it, because a person triggered the visit.

Skip llms.txt as a fix. 33 of 70 e-commerce sites we could check had one (47.1%), yet Google says AI features need no new machine-readable files. We publish one because it is cheap; see why we call it overrated.

Ask the engines your shoppers' questions and record who gets named. Write 10 to 15 unbranded questions, such as "best waterproof hiking boots for wide feet", and run them in ChatGPT, Perplexity, Google AI Mode, Gemini and Claude. Repeat monthly.

Log date, engine, question, products named and your position. Track mention rate first, then which competitors appear and which sources the engine cites for them.

Position matters. In AEOeye's own audit records, a small, self-selected set of 127 audits, when a brand was named and ranked, it came first in 71.4% of 147 ranked mentions. Treat it as a hint, not a benchmark; the data and caveats are public.

To skip the manual work, the free AEOeye audit asks ChatGPT three unbranded buyer questions, shows the full answers and lists the competitors it names instead. The $29 report adds Claude, Gemini, Google AI and Perplexity across 10 questions.

For daily tracking across hundreds of prompts, dedicated trackers do that and we do not. More on method: measuring AI visibility. On Shopify? See our Shopify guide.