Getting Products Found in AI Search
Getting found in ChatGPT, Perplexity, Gemini, and Claude rests on much of what already makes SEO work: clean product data, real reviews, fast pages, and a brand people talk about. Two parts are newer: structured feeds built for AI shopping, and a greater reliance on what gets said about you somewhere other than your own site.
The four systems do not work the same way, so it helps to know which one you are optimizing for.
How Each Platform Finds Products
ChatGPT draws on a structured merchant feed if you have supplied one through OpenAI’s merchant program, plus live web retrieval. OpenAI documents four separate crawlers, and the distinction matters more than most guides suggest: OAI-SearchBot makes a site eligible for ChatGPT search features. GPTBot collects training data. ChatGPT-User fetches a page when someone asks for it directly. Blocking GPTBot to keep your catalog out of model training does not remove you from search answers, and blocking OAI-SearchBot does. Those are separate decisions. Stores sometimes block one crawler while meaning to block the other.
The web index behind that retrieval is less settled than many guides suggest. OpenAI replaced its Bing-branded browsing feature in late 2024 and describes using third-party search providers without naming them. Bing remains a reasonable inference because of the partnership. Keep your Bing Webmaster Tools listing healthy regardless, but treat “ChatGPT runs on Bing” as an educated guess rather than something OpenAI has confirmed.
Perplexity is citation-first by design. Every product recommendation traces back to a source, whether a review site, a retailer page or a forum thread, and Perplexity states its product cards are not sponsored. It draws from merchants enrolled in its shopping program and from open-web crawling of everything else. Because it reads and cites actual pages, a mention on a well-regarded review site can matter more here than in a conventional search result.
Google’s Gemini and AI Mode run on the Shopping Graph, which Google describes as “more than 50 billion of the world’s products, 2 billion of which are updated every hour”. A Merchant Center feed is how most stores get product data into that system, with Product markup on your pages doing corroborating work. SEO coverage often calls Merchant Center the canonical source of the Shopping Graph. Google does not make that claim in the documentation I found. Feeds still matter; the stronger claim belongs to someone else.
Claude has no dedicated shopping surface, and people ask it for buying advice anyway. It answers from training data plus live web results where that is enabled, which means there is no feed to submit and no program to join. What reaches it is your public product information and what other people have published about you. If your specs and reviews exist only inside a storefront widget, Claude is more likely to expose that gap.
| Platform | Feed program | What sits behind the answer | What to submit |
|---|---|---|---|
| ChatGPT | OpenAI merchant program | Third-party search providers; Bing not confirmed | Merchant feed; allow OAI-SearchBot |
| Perplexity | Perplexity shopping program | Citation-first open-web retrieval | Merchant feed; product pages others can cite |
| Gemini and AI Mode | Google Merchant Center | Google Shopping Graph | Merchant feed; corroborating Product markup |
| Claude | None | Training data plus live web search, where enabled | Nothing to submit; publish clear specs and earn third-party reviews |
For a store doing real volume, the mechanical version is short. Get a feed live in Google Merchant Center, then OpenAI’s and Perplexity’s merchant programs, and keep Product markup accurate underneath all three.
What Your Product Pages Need in the HTML
The baseline is unchanged: Product markup with name, brand and a GTIN or MPN; Offer markup with price, currency and availability; and Review or AggregateRating tied to reviews a visitor can see on the page.
Four details determine whether the markup can be used reliably:
- Name the product identically in your H1 and your structured data. A mismatch gives a parser two competing answers.
- GTIN or MPN is how these systems match your listing to the same item sold elsewhere. Without it you are harder to place in a comparison.
- AggregateRating must reflect real, visible reviews. Inflated ratings can risk a Google manual action and undermine trust as soon as someone compares the markup with the page.
- Individual Review markup, not only the aggregate, gives these systems the specific language people use. That is what gets matched against a question like “hiking boots for wide feet.”
One newer risk deserves an early check on a modern storefront. Independent testing through 2026 has reported that OpenAI’s crawlers do not execute JavaScript, and OpenAI’s own documentation says nothing either way. Nobody should treat that as settled. You can check the condition on your own site in about a minute. Open a product page, view source rather than inspecting the rendered DOM, and search for your price, your specs and your review text. If they are absent from the HTML and appear only after scripts run, a crawler that does not render JavaScript may see little or no usable product information. That matters regardless of which crawler turns out to render what.
If you have already done careful e-commerce SEO, a large share of what gets sold as AEO, GEO or AI SEO is work you have done. If your feeds and markup are a mess, this is a real project and deserves a real budget rather than an add-on line item.
Google’s Universal Commerce Protocol
Google announced the Universal Commerce Protocol on January 11, 2026, co-developed with Shopify, Etsy, Wayfair, Target and Walmart and endorsed by more than 20 other companies including Stripe, Visa, Mastercard and Best Buy. It gives AI agents one common interface for discovering products, managing a cart and checking out, instead of a bespoke integration per platform, and it builds on existing standards including the Model Context Protocol.
Google wrote at announcement that UCP “will soon power” checkout on eligible listings in AI Mode and the Gemini app. That is future tense. The honest status today is early. The backing is serious enough to revisit quarterly, and not mature enough to justify reordering this quarter’s work. Shopify and WooCommerce have the clearest paths so far. A custom stack, such as Suite Commerce Advanced running as a single-page app, needs separate scoping before anyone promises a timeline.
What the Research Says About Brand Mentions
Ahrefs studied 75,000 brands and found branded web mentions correlated with AI Overview visibility at 0.664, against 0.218 for backlinks. In a companion study across ChatGPT, AI Mode and AI Overviews, YouTube mentions were the strongest single factor measured, around 0.737.
The finding belongs in the plan, but it needs to be reported accurately. The popular version gets two things wrong. You will see it written as “brand mentions beat backlinks three to one,” which comes from dividing one correlation coefficient by another. Dividing those coefficients does not create a meaningful “three-to-one” comparison. And Ahrefs is explicit about the limits of its own data: “correlation ≠ causation,” and all the factors studied showed “moderate to very weak correlations on the Spearman scale.” The pattern is real. It does not show that earning a mention causes visibility, and it does not justify shifting a budget three to one.
What survives that scrutiny is still useful. Across multiple platforms, mentions correlated more strongly than links in a sample large enough to take seriously. For a store, the practical reading is that being reviewed, compared and discussed on sites you do not own belongs in the plan alongside the technical work, and that video belongs in the plan rather than at the end of it.
Reddit’s Trust and Its Risk
Reddit appears constantly in this research as a source both ChatGPT and Google’s AI systems lean on. That trust has drawn brands trying to manufacture the appearance of organic conversation, and many of those efforts have gone badly. The Wall Street Journal reported this month that brands have discovered Reddit while Redditors have not returned the feeling, and moderators have organized specifically to name and shame the attempts.
We do not offer Reddit marketing, largely because of this risk. The upside is real: a genuine, disclosed contribution from someone who actually knows the product earns the kind of mention these systems weight. The failure is public and fast, and it can cost more reputation than a mention could return. If you participate, let that person post under their own name and disclose the affiliation. Post nothing you would not want quoted back to you. If that does not fit how your team works, sitting the channel out is a legitimate choice. Reviews, comparison content and video are easier to sustain honestly.
What AI Visibility Tracking Can and Cannot Tell You
SparkToro had 600 volunteers run 12 brand-recommendation prompts through ChatGPT, Claude and Google’s AI, 2,961 runs in total. Asked the same question twice, the systems rarely named the same brand list and returned the same ordering about once in a thousand runs.
External guidance
Most coverage stops there and concludes that tracking is hopeless. SparkToro did not. They split the verdict: tracking your ranking position in AI answers is “a fool’s errand,” while “visibility % across dozens to hundreds of prompts run multiple times is a reasonable metric.” The workable half comes with a price tag.
External guidance
That is what turns a vague question into arithmetic. Twenty prompts that matter to your category, at 60 to 100 runs each, is 1,200 to 2,000 queries per measurement cycle, and it has to repeat to show a trend. For a large retailer that is a defensible line item. For most stores, it buys a number that can move for reasons nobody can attribute. The same budget may be more useful for product data, reviews, or video, where the work itself is visible.
There is also a ceiling no vendor can raise. AI answers are shaped by the person asking, including their history and context, and tracking tools query from clean accounts that strip exactly that out. No tracker can show everything these systems tell every customer, because there is no public record of what your customers asked.
Checking in beats monitoring. A few times a quarter, ask each platform about your brand, your main category and your service area, and read the answer for the things that hurt: wrong price, discontinued product presented as current, confusion with a competitor, or your absence from a category you should own. Those problems are visible in a manual check and often fixable.
Where to Start
Three take an afternoon. The rest need to become routine.
This week
- Get a current feed live in Google Merchant Center. Add OpenAI’s and Perplexity’s merchant programs if e-commerce is a meaningful revenue channel.
- View source on a product page and confirm your price, specs and reviews are in the HTML rather than injected by script.
- Check your robots.txt. OAI-SearchBot and its equivalents need access, and blocking training crawlers is a separate decision from blocking search retrieval.
Ongoing
- Keep Product, Offer and Review markup clean, with the name matching your H1, GTIN or MPN populated, and ratings tied to reviews visible on the page.
- Write product copy that answers the questions people ask, in plain language. This has not changed. It matters more now because these systems synthesize an answer rather than hand back a list of links.
- Invest in being discussed somewhere other than your own site. Reviews, comparison coverage, and video have the clearest support in the research reviewed here, and many stores underinvest in them.
- Check in on the platforms a few times a quarter. Skip the daily monitoring subscription until you can name the decision it would change.
What Stays the Same
Being findable in AI search rewards what good SEO has always rewarded: accurate information, evidence that other people trust you, and a fast, well-built site underneath. What has changed is part of the mechanism. Feeds and structured data now do measurable work, and what gets said about a store somewhere other than its own homepage may matter more than it ever has.
The Universal Commerce Protocol belongs on the roadmap. The correlation research belongs in the plan, provided it is read carefully, because most of what is written about it overstates what it found. If a platform changes something that affects a client, we will say so and adjust rather than pretend the story was cleaner than it was.