AI product recommendations: how to get your products picked by AI assistants

How AI chooses recommended products

AI assistants choose products by matching the shopper's request against product records they trust, then ranking those records on relevance and merchant quality. A prompt carries more detail than a search query. "Waterproof trail shoes under $150 for wide feet" holds four constraints, and the model looks for products whose data answers every one of them. OpenAI says ChatGPT ranks merchants using its availability, price, quality criteria. It also considers whether the seller is the maker or primary seller of the item.

Structured data and the open web are the models' two kinds of source; merchants supply the structured data through the platform or their own pages. According to OpenAI's launch announcement, its shopping research feature reads product pages directly and cites them. The feature avoids low-quality or spammy sites. Google's AI Mode pulls from the Shopping Graph, which holds more than 50 billion product listings and refreshes over 2 billion of them every hour.

Shoppers already use these answers to decide what to buy. In Adobe's survey of 5,000 US consumers, 38% had used generative AI for online shopping, and 40% of those users wanted product recommendations from it. Adobe Analytics then measured a 693% rise in AI traffic to US retail sites over the 2025 holiday season.

Don't expect a fixed ranking. SparkToro ran the same recommendation prompts nearly 3,000 times and found fewer than 1 in 100 responses repeated the same list of brands. So the useful question is how often your product shows up across repeated runs of a prompt. A product's position in a single answer provides little evidence of its overall visibility.

Product data signals that matter

AI uses identifiers and descriptive attributes to recommend products. Live commercial data and factual copy complete these four groups of product data. Each group answers a different question the model asks. The model identifies the product through its identifiers and checks whether it fits the request through its attributes. Commercial data tells it whether a shopper can buy it today. The copy gives it plain sentences it can quote.

Product identifiers

Identifiers let an assistant match your listing to the same product on review sites and other retailers. The main one is the Global Trade Item Number (GTIN), the barcode number a manufacturer assigns. Google's Merchant Center help warns that "products with missing or incorrect GTIN attributes may have limited visibility." Private-label and handmade goods have no GTIN. Those listings need the brand plus a manufacturer part number (MPN). Never invent a GTIN to fill the field, because a wrong number links your product to someone else's reviews.

Attributes that answer prompts

Attributes are where most catalogs lose recommendations. Shoppers ask about material and fit. They also ask about weight or compatibility with another product. When the field is empty, the model has no evidence your product qualifies, so it picks a rival whose data does. Treat optional feed fields as required wherever the value exists. Pay most attention to the attributes your buyers mention in reviews and support tickets, since those are the words they'll type into a chatbot.

Price and availability

Live commercial data decides whether a product can be shown at all. An assistant that recommends an item which turns out to be out of stock gives the shopper a bad answer, so platforms favor fresh records. The prices in your feed and schema must match the price on the page at all times. When they drift apart, the platform has reason to doubt the whole listing.

Descriptions written as facts

Product copy works best for AI when it reads like a spec sheet written in sentences. OpenAI's feed specification caps titles at 150 characters and asks for plain-text, factual descriptions up to 5,000 characters. It rejects marketing language. Copy like "This breathable merino blend weighs 180 gsm. Machine washable at 30°C" gives a model three facts to match against a prompt. "The comfiest shirt you'll ever own" gives it nothing.

How reviews shape AI picks

Reviews tell an assistant whether a product does what its listing claims, and that evidence settles which product makes the shortlist. Models read star ratings and review counts from feeds and markup. They also summarize what reviewers say. If buyers keep writing that a jacket runs small, expect the assistant to repeat that warning in its answer.

Shoppers check those opinions too, after the AI answers. Reddit's Path to Purchase 2026 survey of 13,956 US adults, run by Attest, found that 50% of US shoppers verify AI recommendations on Reddit before they buy. A recommendation has to hold up twice, both inside the model and on the forums the shopper visits next.

The sources that count for consumer electronics are professional review sites and comparison tests. For beauty and apparel, customer reviews and community threads carry weight. B2B and niche hardware buyers read forums and specialist publications. Find out where your category's opinions live by asking an assistant your buyers' questions and reading the sources it cites.

For the first of three checks on your review profile, compare review count and recency with the competitors that appear in AI answers. Second, read your low ratings for recurring complaints, because a fixable product or sizing issue will keep showing up in AI summaries until you fix it. And third, make sure on-site ratings are marked up so machines can read them.

Feed quality drives recommendations

Feed quality decides whether an assistant can find your products and trust them enough to show them. The signals that count most for AI product discovery are completeness and freshness. Stable identifiers and consistency with the product page follow closely in importance. Correct categories matter too. A feed that fails on any of these gives the platform a reason to rank a competitor's cleaner record above yours.

Freshness matters more than it did in classic shopping search. OpenAI's product feed accepts CSV or TSV files. It also supports XML or JSON files. Updates are accepted as often as every 15 minutes. A nightly export leaves hours where your price or stock is wrong on every platform reading the file. Stable item IDs preserve the history and review links the platform attaches to each item, which it can lose when an ID changes.

OpenAI built the Agentic Commerce Protocol (ACP) so merchants can share product feeds and promotions with ChatGPT, with the connection method dependent on their stack. Target and Best Buy are among the retailers that have integrated into ACP for discovery. Shopify merchants get their data into ChatGPT through Shopify Catalog, with no extra work. For Google, the feed goes through Merchant Center.

Run this audit on your feed before you change anything else:

1. Export the live feed and count empty values per attribute. Begin with GTIN and brand.
2. Compare a sample of feed prices and stock levels against the live product pages.
3. Confirm that item IDs have stayed the same since your last platform migration or theme change.
4. Check that every product sits in the most specific category the platform's taxonomy offers.
5. Rewrite titles and descriptions that carry promotional claims in place of attributes.
6. Set the refresh schedule to match how often your prices and stock change.

Getting cited by AI assistants

Assistants cite pages that answer the shopper's comparison question in plain, checkable text, so brands get cited by publishing that content and earning mentions on sites the models already trust. An ecommerce product page answers "what is this item." It rarely answers "which is better for wide feet, A or B." Buyers ask for recommendations in that second question, and buying guides or comparison pages on your own domain give the model a source to quote. Use-case articles on your domain provide another source.

Write those pages the way a model reads them. Put the direct answer in the first sentence under each heading. Name products with their full model names and give real specs and prices. Keep claims you can't back with a source off the page. An assistant has a reason to cite a guide with a table that sets out sizing and weight for 4 models, with a price for each model.

Host these pages on your own domain, because citations then send traffic and authority to your store. We publish client pages on the client's own domain through a DNS connection. This works on any CMS and requires no migration. The client reviews and approves each page in our dashboard before it goes live. For Agrello, an e-signature company, that work produced +125.6% organic search sessions year over year and 51.8K impressions from Snoika-created pages.

Language is the other gap we see in ecommerce catalogs. A shopper in Madrid asks ChatGPT in Spanish, and the model looks for Spanish sources. If your guides exist only in English, a local competitor wins the answer by default. We create content in 135+ languages and regional variants, which lets one catalog compete for prompts in each market it ships to.

Off-site product roundups and specialist reviews help models decide which brands belong in a category. Community threads contribute to that evidence. When Vespia, a fintech SaaS company, combined on-site content with wider coverage, it recorded a 220% increase in AI brand mentions.

Structured data for discovery

Structured data gives AI systems a machine-readable copy of your product facts, and it confirms that what the feed says matches what the page says. On a product page, that means schema.org Product markup with nested Offer details for price and stock, plus AggregateRating for reviews. Shipping and return policy properties add the details shoppers ask about before they buy.

Google uses product snippets, one of its two types of product markup, for pages that discuss a product without selling it, like a review. Its other type, merchant listings, suits pages where a shopper can buy. Google's merchant listing documentation says only those purchase pages qualify and that the price must be greater than zero. Google also says that providing both markup and a Merchant Center feed "maximizes your eligibility" and helps it verify your data.

Template errors can drop the Offer block after a theme update or leave the default variant's price in the markup while a variant selector shows another price. These errors don't show to shoppers, so they can sit unnoticed for months. ABS Technologies saw 6x SERP impressions after its technical fixes and improved keyword indexability. It also significantly increased search visibility within the first 30 days.

Validate a sample of product pages in Google's Rich Results Test after every template change. Then compare the marked-up price and availability with the feed values for the same item IDs. Variant pages need their own check, since each variant must carry its own identifiers and price.

Monitoring recommendation visibility

Track AI product recommendation citations with a fixed set of buyer prompts run across assistants on a schedule. Record whether your brand appears and where it's cited. Also record how it's described. Use real purchase questions about categories ("best running shoes for flat feet") and comparisons in the prompt set. Add questions that name a use case. Run each prompt more than once per check, because the SparkToro data shows single answers vary too much to judge from.

Record mention rate, which tells you how often your brand appears, as one of four measures for each run. Share of voice compares that rate with your competitors' rates on the same prompts. Citations show which URLs the assistant used as sources, and sentiment shows whether its description helps or hurts you. Citations matter most for action, since they point to the exact pages and third-party sites that shape your answers.

Your analytics identify ChatGPT traffic by the utm_source=chatgpt.com tag it adds to the links it sends, while other assistants show up as referrers in Google Analytics 4 (GA4). Compare that referral traffic with your mention rate over time. A rising mention rate with flat traffic means the assistant names you but links elsewhere, such as to a marketplace or review site.

Manual tracking in a spreadsheet works for a 20-prompt baseline. It stops working once you cover several assistants and markets every week. We use weekly prompt sets to track mentions and share of voice in ChatGPT and Google AI Overviews, where most AI search traffic happens. The same tracking measures sentiment and citations. 1StopVAT, in tax and fintech, saw 10x growth in ChatGPT visibility in 7 weeks.

Managing AI product visibility

An in-house team or an agency provides two of the four setups ecommerce brands use to handle AI visibility. The other options are a monitoring-only tool or an all-in-one platform with services. The right choice depends on who will fix the problems that tracking reveals. Monitoring tells you that a competitor wins a prompt. Someone still has to repair the feed and write the comparison page. The review coverage also has to be won.

Snoika's plans are priced by strategy slots: Starter runs one strategy at a time for €124 a week, Standard runs two in parallel for €248 a week, and Pro runs three or more for €495 a week. Enterprise is priced on request. Enterprise adds multi-brand management, custom integrations and API access.

Our verdict is that a large retailer with a strong feed team and its own writers can run generative engine optimization in-house and add a monitoring-only tool. Smaller brands rarely have the hours to act on what monitoring finds. For them, a GEO services setup where the same team tracks the prompts and ships the fixes saves the handoff where most of the work stalls.

ChatGPT product recommendations

To get your products recommended by ChatGPT, make your catalog readable to OpenAI's systems, then give the model independent evidence that your products fit the prompts buyers type. ChatGPT's shopping results are organic and reflect relevance to the request. Product metadata and the sources the model finds on the web also shape these results. Work through these steps in order:

1. Allow OpenAI's search crawler, OAI-SearchBot, in your robots.txt file so ChatGPT can read your product pages.
2. Submit your product feed through ACP, or confirm your Shopify Catalog connection if you sell on Shopify.
3. Follow OpenAI's allowlisting process so your store can appear in shopping research results.
4. Fix feed and markup gaps from the audits above. Address GTINs and missing attributes first.
5. Publish comparison and buying-guide pages on your domain for the prompts where you're absent.
6. Earn coverage on the review sites and forums ChatGPT cites in your category.
7. Track your mention rate weekly and repeat the cycle on the prompts you still lose.

Expect results to build over weeks. ChatGPT's answers come from retrieval at the moment of the question, so a fixed page or a new review thread can change answers soon after it's indexed. Broader category authority takes longer to form, because it depends on mentions across sites you don't control.

The payoff is inbound demand from buyers who arrive with a decision half made. 3DPrinterOS achieved #1 visibility in ChatGPT for industry-specific queries and turned AI search into a steady inbound lead source of 3–4 qualified inbound leads/day. It also reported an 87% increase in LLM brand mentions to Startup Reporter. 3DOS reached top visibility in ChatGPT for targeted industry queries and saw 2x registered users growth.

Next steps

Measure your current AI visibility before you touch the feed. Without a baseline, you won't know which fix moved your mention rate, and AI answers vary too much from run to run to judge by memory. Pick 20 to 50 real buyer prompts. Record which products the assistants recommend today and the sources they cite. That list identifies gaps in feed data or reviews. It also reveals missing comparison content.

We start every engagement with onboarding, where we learn about your category and buyers. We also learn your goals during that process. We ask for GA4 and Google Search Console access. We also need access to Google Tag Manager. Your baseline for AI product recommendations comes from that first step. Get a free AI visibility report to see where your products appear today.