What Gemini Looks for on a Product Page
A founder-voice guide to the on-page signals Gemini reads for ecommerce shopping queries, the Merchant Center signal that tilts answers, and the fastest fixes.
If you sell a premium product on Shopify and Gemini has been recommending competitors when buyers ask about your category, the gap is usually two things. A Merchant Center feed that's stale, or a product page where the price and availability live inside JavaScript that runs after the page loads. Gemini reads the same Google plumbing that powers Search, so the fixes are familiar. They just have to actually ship.
What does Gemini look for on a product page?
Gemini reads a product page through three signals at once. The live page (HTML, Schema.org Product JSON-LD, Core Web Vitals). The Google Shopping feed if Merchant Center is connected. And visual signals from Google Lens. The pages that get cited inside Gemini's shopping answers ship clean Schema.org Product JSON-LD, sync a healthy Merchant Center feed, and have product images Lens can match against the open web. Brands that win on Gemini also win on Google AI Overviews.
The retrieve-vs-cite asymmetry that hits every assistant hits Gemini too. A Search Engine Land write-up of an AirOps study found that only 15% of pages an LLM retrieves end up in the cited answer. The fetch is wide. The cite is narrow. Pages that survive the cut have a clean answer at the top, server-rendered structured data, and a feed that confirms the live data.
Where Gemini differs from ChatGPT and Perplexity
The on-page floor is the same across all four assistants. The differences sit in the discovery and feed layers.
Merchant feeds. Gemini pulls from the Google Shopping feed through Merchant Center. Google's product data spec lists the required attributes: id, title, description, image link, price, availability. Brands that already run Google Shopping ads have most of this in place. Brands that have never touched Merchant Center are starting from zero on Gemini even if their PDPs are perfect.
ChatGPT has its own OpenAI Commerce feed, separate from Google. Perplexity has neither and relies on live page retrieval. Claude uses MCP for tool calls, which is a different shape of integration. How ChatGPT Shopping picks products, what Claude looks for on a product page, and what Perplexity Shopping looks for cover the other three.
Visual search. Gemini is the only one of the four with a deep visual pipeline through Google Lens. A buyer can point a phone camera at a couch in a friend's apartment, and Lens can match it to a SKU on the open web. None of the other assistants have this surface.
AI Overviews tie-in. Gemini and Google's AI Overviews share signal. The work that gets a brand cited in Gemini also gets it cited in the AI Overview at the top of a Google search result. That's two surfaces for one body of work.
The Merchant Center signal that tilts Gemini's answers
Most premium DTC brands have a Merchant Center account from running Shopping ads, but the feed is stale or partial. That's the gap.
Google's product data specification sets the floor. Required attributes are id, title, description, image link, price, and availability. The image link must be a full URL, not a path. The price must be a number with a currency. Availability has to be one of in_stock, out_of_stock, preorder, or backorder.
Three things tilt Gemini's answers when the feed is healthy:
- Image quality. The image must be at least 100 pixels for non-apparel and at least 250 pixels for apparel. In practice, ship 800 pixels wide or larger. Low-res images get filtered.
- Title and description match the page. If the feed says "Wool Coat" and the PDP says "The Heritage Wool Overcoat", that's friction. Match the names.
- Freshness. Submit the feed at least every 30 days. Fresh feeds get pulled into more shopping answers than stale ones.
If you can do one thing this week, log into Merchant Center and check the Diagnostics tab for any disapproved or warning items. Fix those first.
The live-page signals Gemini reads
The feed is one input. The page is another. Gemini cross-checks both.
Signal one: Schema.org Product, Offer, AggregateRating. Server-rendered JSON-LD in the document head, not injected by JavaScript. The Schema.org Product type lists the fields. The product schema markup guide covers exactly which to ship. The price has to match the feed. If the feed says $249 and the JSON-LD says $279, Gemini may skip the page or surface the wrong number.
Signal two: Direct answer in the first 150 words. Open the body with 2 to 3 sentences that name what the product is and why a buyer would want it. Move the brand origin story below the spec table. Gemini extracts these sentences when composing a citation.
Signal three: FAQPage JSON-LD. A short FAQ block with 3 to 6 buyer questions and answers, marked up with FAQPage schema. The questions are the ones a buyer would ask out loud. The answers stay in 1 to 2 sentences. This is the same signal that earns FAQ rich results on Google Search.
Signal four: Image alt text and file names. Alt text says what the image shows in plain words. File names use the SKU, not IMG_2845.jpg. Both help Lens match a buyer's photo to your SKU.
Signal five: Core Web Vitals. Largest Contentful Paint should be under 2.5 seconds. Slow pages get crawled less and quoted less. The LCP fix is usually a hero image: serve it at the size it's displayed, not 4000 pixels scaled down by the browser.
These five together are the on-page floor. Brands shipping all five plus a clean feed get cited consistently.
Google Lens and visual search
Lens is the part most brands ignore. It's also the easiest to fix.
A buyer takes a photo. Lens runs the image against the open web. It returns matches. If your PDP has a clean image of the SKU, the match lands on you. If your PDP only has lifestyle shots, the match lands on a competitor whose image is closer to the photo.
Three things make a product Lens-friendly:
- A clear product shot on a plain background, in addition to lifestyle shots.
- A high-res image (at least 1200 pixels wide) hosted on a public, indexable URL. Not behind a login. Not lazy-loaded in a way that hides it from the crawler.
- Alt text and a file name that name the SKU and the product type. "wool-coat-heritage-navy.jpg" beats "image1.jpg".
Test it yourself. Open the Google app, point Lens at one of your products, and see what comes back. If a competitor wins the match on your own product, you have your fix list.
Why Gemini pulls from Reddit and YouTube more than ChatGPT does
Google has a content licensing deal with Reddit. That means Gemini gets first-party access to Reddit threads in a way other assistants don't. Reddit is one of the strongest third-party signals for Gemini shopping answers.
YouTube is also Google. Gemini pulls product context from YouTube reviews and unboxings. A 5-minute review of your product on a category-relevant YouTube channel earns Gemini signal in a way that's hard to replicate.
What this means for product page strategy:
- Engage on Reddit as the brand. Once a month, answer category questions in the relevant subreddits. Don't shill. Help.
- Send your product to two or three category YouTube channels. Even a small channel with 5,000 subs and a real audience moves the needle for Gemini more than a big banner ad.
- Host the YouTube embed on the PDP. A video review embedded near the bottom of the product page gives Gemini a clean cross-reference between the page and the video.
This is signal compounding. The on-page work is the floor. Reddit and YouTube are the lift.
The three-day fix for a Gemini-invisible brand
If your brand is invisible on Gemini and you want to fix it without a full rebuild, run this 3-day plan.
Day one: audit Merchant Center. Log in. Open Diagnostics. Fix every disapproval and warning on the feed. Confirm price, availability, and image link are correct on the top 10 SKUs by revenue. If you don't have a Merchant Center account, set one up and submit the feed today. The product data spec covers what to ship.
Day two: ship clean Schema.org Product markup on every PDP. Validate every PDP against Google's Rich Results Test. Confirm Product, Offer, and AggregateRating come back without errors. The price in the JSON-LD has to match the feed and the visible page. Fix the template, not individual pages.
Day three: rewrite the first 150 words and the alt text. Open every PDP with a 2 to 3 sentence direct answer to the buyer's question. Move the brand origin below the spec table. Update image alt text and file names to use the SKU. This is the move that lifts both Gemini citation and Lens matching.
Three days. The brand goes from "not pulled" to "pulled" by Gemini. Citation rate ramps up over the following weeks as the feed refreshes and the schema gets re-crawled.
How to watch citations
Track the work. Otherwise you're guessing.
- Type the buyer query into Gemini. Go to gemini.google.com and ask "best leather backpack for daily commute under $400" or whatever your category question is. Note which brands appear. Run it weekly.
- Check AI Overview snapshots. Run the same query in Google Search. Check whether your brand appears in the AI Overview at the top of the page. Screenshot the result.
- Track Search Console impressions on AI surfaces. Search Console reports impressions on AI surfaces inside the Performance report. Filter by your brand-name and category queries. Watch the trend.
Three checks. Run them every Monday. Within four to six weeks of shipping the three-day fix, the citation rate should be visibly different.
Where to start
If you're invisible on Gemini, fix the feed and the schema first. That's the floor. The AEO playbook for ecommerce covers the broader signal set across all four assistants. The cluster pillar on getting product pages cited covers the cross-engine version of this work.
If your floor is shipped and Gemini still picks competitors, the gap is usually Reddit and YouTube. The Google deal with Reddit and the YouTube tie-in mean third-party signal compounds harder on Gemini than on the other assistants.
The buyer asks the question on a phone. Gemini reads the page, the feed, and the image. The brand whose page is extractable, whose feed is fresh, and whose image is matched on Lens is the brand cited. The other brand keeps spending on ads.