83% of the products ChatGPT recommends in its shopping carousels are sourced directly from Google Shopping organic results.

That number comes from a study of 43,000 carousel products across 10 verticals. Researchers found base64-encoded Google Shopping parameters hidden in ChatGPT's source code, confirming a direct data pipeline between your Google Merchant Center feed and ChatGPT's product recommendations (Search Engine Land, Tom Wells, March 2026).

Your Google Merchant Center feed, the thing most ecommerce teams treat as an operational task to avoid disapprovals, is already the primary input determining whether ChatGPT recommends your products to 900 million weekly users. It also feeds Google AI Overviews (2 billion monthly users), Perplexity Shopping (via Shopify feed integration), Microsoft Copilot, and Amazon Rufus.

Most feeds were built to pass Google's compliance checks. They were never designed to serve as the context layer that AI agents reason over, compare against competitors, and use to make purchase decisions on behalf of shoppers. That gap is where revenue leaks.

This guide covers how AI shopping agents source products, exactly what to fix in your feed, and how to connect feed optimization to the broader AI visibility system that turns product data into revenue. For how agentic commerce works at the platform level, see our agentic commerce guide.

How ChatGPT, Perplexity, and Google AI Mode select which products to show

ChatGPT does not crawl your website to find products. It uses "shopping query fan-outs": short, intent-specific sub-queries (averaging 7 words) that retrieve products from Google Shopping's organic index. When a user asks "best waterproof hiking boots under $150," ChatGPT generates a shopping fan-out, retrieves the top Google Shopping results for that query, and populates a carousel from those results.

The data from the Search Engine Land study shows:

• 60% of ChatGPT carousel products come from Google Shopping's top 10 organic positions

• Products outside the top 40 are effectively excluded from the selection pool

• One page of Google Shopping results is typically enough to populate an 8-product carousel

• ChatGPT uses an average of 1.16 shopping fan-outs per prompt (versus 2.4 search fan-outs for contextual information)

Each AI shopping platform sources slightly differently:

The common requirement across all platforms: structured, complete, accurate product data refreshed frequently. If your feed has gaps, AI agents skip your products before any other signal can help you.

The product feed audit every ecommerce brand should run this month

Most feeds pass Google's minimum requirements and stop there. AI shopping agents need more. Here is what to audit, organized by priority.

Critical: fix this week

Titles that match how people actually ask AI for products: Shopping fan-outs average 7 words and are conversationally phrased. "Red Dress Cotton" does not match how any human asks an AI for a product. "Women's A-Line Cotton Midi Dress in Cherry Red" does. Rewrite every product title to include the product type, key material or feature, color, and who it is for. This is the single highest-impact change for most feeds.

Pricing consistency across every channel: ChannelEngine's survey of 4,500 marketplace shoppers found that 95% notice price differences for identical products across platforms (ChannelEngine, Marketplace Shopping Behavior Report 2026). When your feed says $49.99 and your site says $54.99, AI agents drop you from consideration entirely. A price mismatch signals unreliable data. Sync pricing in real time across your feed, your site, and every marketplace where your product appears.

Complete schema markup on every product page: AI agents parse Product schema, Offer schema (price, availability, currency in ISO 4217 format), AggregateRating schema, and Review schema. Every missing field is a missed AI recommendation. Audit every product page. If schema is incomplete or absent, your products are invisible to AI shopping regardless of how good they are.

GTIN and product identifier accuracy: AI agents use GTINs to match products across sources, verify authenticity, and cross-reference pricing. Missing or incorrect GTINs break the matching process and can exclude your products from carousel consideration entirely.

Important: fix this month

Reviews and ratings in your feed: The ChatGPT feed specification accepts product_review_count and average_rating as feed-level fields. ChannelEngine's research found 3 in 5 shoppers hesitate to purchase if a product has no reviews. AI agents weight this signal when deciding which products to recommend. If you have reviews, get them into your feed. If you do not have reviews, start collecting them immediately.

Shipping and return policy fields: 91% of shoppers say free shipping directly influences purchase completion (ChannelEngine, 2026). AI agents increasingly surface shipping speed and return policy transparency as trust signals. Include free_shipping_indicator, shipping_speed, and return_policy in your feed. These fields directly affect whether an AI agent confidently recommends your product or defaults to a competitor with better trust signals.

Descriptions that answer buying questions, not marketing questions: AI agents parse descriptions to match conversational queries. "Elevate your wardrobe with our stunning collection" tells the AI nothing useful. "Waterproof hiking boot with Vibram sole, Gore-Tex lining, 200g insulation, fits true to size, available in men's 7-14" gives the AI every attribute it needs to match the query "waterproof hiking boots for winter, men's size 11, under $150."

Write descriptions that state what the product is, who it is for, what problem it solves, and what specific features differentiate it. Not marketing copy. Buying copy.

Advanced: ongoing optimization

Feed refresh frequency: The ChatGPT feed specification supports refreshes every 15 minutes. At minimum, update pricing and availability daily. Stale inventory data (showing "in stock" when the product is sold out) damages your trust score with AI agents and creates a poor buyer experience that leads to returns and negative reviews.

Video and 3D model links: The ChatGPT feed spec accepts video_link and model_3d_link fields. These are not widely adopted yet, which means early adopters gain a visibility advantage. Product videos (short, product-focused, hosted on YouTube) and 3D models (GLB or GLTF format, valuable for furniture, electronics, and apparel) enrich the AI's understanding of your product and improve recommendation confidence.

Multiple product images with varied context: Perplexity's Snap to Shop feature lets users photograph an item and find similar products. Products with only white-background studio shots miss visual search matches from users photographing products in real-world contexts. Include lifestyle images, scale references, and detail shots alongside standard product photography.

Localized feeds for each market: AI agents cannot simulate geolocation. A single URL with dynamic pricing based on IP address does not work for AI crawlers. Create distinct product URLs per market (yoursite.com/us/product-name vs. yoursite.com/uk/product-name) with market-specific schema, localized pricing, currency codes in ISO 4217, and region-appropriate shipping information on each page.

Why feed optimization alone is not enough

Feed optimization gets your products into the AI selection pool. But making it into the pool is not the same as being selected from it.

The Search Engine Land study authors acknowledged that "product sentiment, brand mentions in contextual sources, and product-specific signals likely influence the final selection and ranking within the carousel." Your Google Shopping ranking determines whether you enter the pool. Your brand authority determines whether AI agents pick you from it.

Three signals beyond the feed that influence AI product recommendations:

Brand mentions across authoritative sources: Ahrefs found that branded web mentions correlated 0.664 with AI visibility, higher than backlinks (0.218) or domain rating (0.326). AI agents recommend brands they encounter across multiple trusted sources: editorial publications, review platforms, industry forums, and community discussions. If your brand only exists on your own website and your product feed, you are harder for AI to trust.

Content depth on purchase-intent topics: The Princeton/Georgia Tech GEO research paper found that content with statistics, citations, and structured evidence boosts AI visibility by up to 40% (Aggarwal et al., ACM SIGKDD 2024). Build content clusters around the purchase-intent questions in your category ("best [product type] for [use case]"). These pages influence AI agents' contextual understanding of your brand alongside the structured feed data. For how to build these clusters, see our topic clusters guide.

Content freshness: Practitioners tracking AI visibility report that content updated within 30 days receives 3.2x more AI citations than stale content. This applies to both your product feed (pricing, availability, descriptions) and your site content (blog posts, comparison guides, category pages). For how content age affects rankings and AI citations, see our content decay guide.

Feed optimization and content authority are not separate strategies. They are two layers of the same system. The feed makes your products parseable. The content makes your brand trustworthy. AI agents need both to recommend you confidently.

How to monitor whether your products actually appear in AI shopping results

You can optimize every field in your feed and build comprehensive content authority. Without monitoring, you still do not know whether AI agents are actually recommending your products.

This is the missing piece in most product feed optimization strategies. The feed is an input. The AI recommendation is an output. You need to measure the output to know whether the input is working.

Passionfruit Labs tracks which purchase-intent prompts surface your products across ChatGPT, Perplexity, Gemini, Claude, and Google AI Overviews. It shows you the specific prompts where competitors appear and you do not, connects to GA4 for revenue attribution (sessions, conversions, and dollars from each AI channel), and generates prioritized action plans based on business impact rather than alphabetical gap lists. Plans start at $19/month with a 7-day free trial.

The monitoring feedback loop: optimize your feed, monitor whether AI agents recommend your products, identify remaining gaps, fix them, measure again. Without this loop, feed optimization is a one-time project. With it, it becomes a continuous system that compounds visibility over time.

For the complete monitoring framework, see our AI brand monitoring guide. For how to separate AI-referred traffic in analytics, see our GA4 AI traffic tracking guide.

Final thoughts

Your product feed used to be an operational input for Google Ads. It is now the front line of AI commerce. ChatGPT, Perplexity, Google AI Mode, and Copilot all rely on structured product data to decide which products to recommend to buyers who are ready to purchase.

The 83% finding makes the priority clear: if your products do not rank in Google Shopping's top 40 for relevant queries, they are excluded from ChatGPT's selection pool before any other factor can help you. Feed quality is the gatekeeper. Brand authority and content depth determine what happens after you pass the gate.

Audit your feed this week. Fix titles and pricing first. Complete your schema. Add the trust fields (reviews, shipping, returns) that AI agents use to differentiate between you and every competitor in the same carousel.

Then monitor whether it is working. Start tracking your AI shopping visibility with Passionfruit Labs (7-day free trial, plans from $19/month). If you need help building the full AI visibility system connecting feed optimization to content authority to revenue measurement, talk to Passionfruit's GEO team. See our case studies.

Frequently asked questions

How does ChatGPT source its product recommendations?

ChatGPT uses "shopping query fan-outs," short intent-specific sub-queries averaging 7 words, to retrieve products from Google Shopping's organic index. A March 2026 study of 43,000 carousel products confirmed that 83% of ChatGPT's product recommendations match Google Shopping's top 40 organic listings. 60% come from the top 10 positions. This means your Google Merchant Center feed is the primary data source determining whether ChatGPT recommends your products.

What product feed fields does ChatGPT require?

Required fields include product ID, title, description, price, availability, images, GTIN, brand, and condition. AI-critical fields most brands miss include product_review_count, average_rating, return_policy, shipping details, popularity_score, video_link, and model_3d_link. The feed specification supports TSV, CSV, XML, or JSON formats with refresh intervals as frequent as every 15 minutes.

Does my Google Shopping feed automatically work for ChatGPT?

Your Google Merchant Center feed is already being used by ChatGPT (the 83% sourcing study confirms this). But "works" and "works well" are different things. A feed built to pass Google's minimum compliance checks likely has gaps that reduce AI recommendation quality: generic titles, missing review data, absent shipping fields, and incomplete schema markup. Optimizing specifically for conversational AI queries requires title rewrites, additional feed fields, and more frequent refresh cycles.

How often should I update my product feed for AI commerce?

The ChatGPT feed specification supports updates every 15 minutes. At minimum, update pricing and availability daily. Products with stale pricing or incorrect availability status get dropped from AI recommendations or create negative buyer experiences that damage your trust signals. Weekly updates for descriptions, images, and review data. Monthly reviews of title optimization against actual shopping fan-out queries in your category.

How do I know if my products appear in ChatGPT shopping results?

You cannot see this in Google Analytics or Search Console. AI shopping visibility requires dedicated monitoring tools that query AI platforms directly and track whether your products are recommended. Passionfruit Labs tracks product visibility across ChatGPT, Perplexity, Gemini, and Google AI Overviews, shows where competitors appear instead of you, and connects to GA4 for revenue attribution from AI channels.

What is the difference between product feed optimization and GEO?

Product feed optimization ensures AI agents can parse, understand, and recommend your specific products. It focuses on structured data: titles, descriptions, pricing, availability, reviews, and schema markup. GEO (Generative Engine Optimization) ensures your brand and content are cited by AI search engines when users ask broader questions about your category. Feed optimization gets your products into AI shopping carousels. GEO gets your brand mentioned in AI-generated answers, reviews, and comparisons. Both feed your AI visibility; they target different surfaces.

What should I fix first in my product feed?

Start with titles and pricing. Rewrite titles to match conversational query phrasing (shopping fan-outs average 7 words). Sync pricing across your feed, site, and all marketplaces so AI agents do not encounter mismatches. Then complete schema markup on every product page, add review and rating data to your feed, and include shipping and return policy fields. These changes address the most common reasons AI agents skip products and can be completed within a week for most catalogs.