Your product feed has 30+ fields. For AI shopping, about 8 of them matter. The rest are either ignored by AI systems entirely or only relevant to paid ad platforms. If you're optimizing your feed the same way for Google Shopping Ads and for AI-powered product recommendations, you're wasting effort on fields that ChatGPT, Perplexity, and Google AI Overviews never look at.
This guide breaks down every major product feed field, tells you whether AI shopping systems actually use it, and gives you the exact optimization strategy for each one. No theory. Just the field-by-field priority list.
Why AI Shopping Treats Your Feed Differently Than Ads Do
Google Shopping Ads use your Merchant Center feed to match products to search queries, then rank them based on bid, relevance, and quality score. The feed is a targeting mechanism. Fields like custom_label, promotion_id, and ads_redirect exist purely to give you control over ad campaigns.
AI shopping is different. When someone asks ChatGPT "what's the best stroller for city apartments?" or Perplexity "compare wireless earbuds under $100," the AI isn't running an ad auction. It's trying to understand products well enough to make a genuine recommendation. That means it needs rich, structured, unambiguous product data.
The AI doesn't care about your bid strategy. It cares whether it can figure out what your product actually is, who it's for, and how it compares to alternatives. This is a fundamentally different use of the same data, and it requires a different optimization approach. For a broader look at how AI systems choose which products to recommend, see our breakdown of how ChatGPT recommends products.
The Field Priority Matrix: What AI Actually Parses
I've organized every standard Merchant Center / product feed field into three tiers based on how AI shopping systems use them. Tier 1 fields directly influence whether AI recommends your product. Tier 2 fields provide supporting context. Tier 3 fields are irrelevant to AI.
| Tier | Field | AI Impact | Why It Matters |
|---|---|---|---|
| 1 | title | Critical | Primary entity identification. AI parses brand, product name, and key attributes from this. |
| 1 | description | Critical | Rich context for use cases, comparisons, and feature extraction. |
| 1 | gtin / upc / ean | Critical | Cross-references global product databases. Enables review matching and deduplication. |
| 1 | brand | Critical | Entity resolution. AI matches this against brand authority signals across the web. |
| 1 | product_type | Critical | Category taxonomy. Tells AI what comparison set your product belongs to. |
| 2 | image_link | High | Visual understanding for multimodal AI. Also used in product cards shown to users. |
| 2 | price | High | Required for comparison queries ("under $50", "best value"). |
| 2 | availability | High | AI won't recommend out-of-stock products. Stale availability damages trust. |
| 2 | Structured attributes (color, size, material, age_group) | High | Enables filtered recommendations ("red leather wallet for men"). |
| 2 | condition | Medium | Distinguishes new vs. refurbished. Relevant for price-sensitive AI queries. |
| 3 | custom_label_0-4 | None | Ad campaign segmentation only. AI systems ignore these. |
| 3 | promotion_id | None | Google Ads promotions. Not parsed by AI recommendation engines. |
| 3 | ads_redirect | None | Ad tracking URLs. Invisible to AI shopping systems. |
| 3 | display_ads_id | None | Display campaign identifier. Zero AI relevance. |
If you're spending time perfecting custom labels and promotion IDs but your descriptions are thin and you're missing GTINs, you have it backwards. For AI shopping, the hierarchy is clear: identity fields first, context fields second, ad fields never.
Tier 1 Fields: The Ones That Make or Break AI Recommendations
Title: Your Product's Identity Card
The title is the single most important field for AI product matching. When someone asks "best noise-canceling headphones for travel," the AI scans titles to build its candidate set. A title like "Sony WH-1000XM5 Wireless Noise-Canceling Over-Ear Headphones, Black" gives the AI everything it needs: brand, model, key feature, form factor, color.
A title like "Premium Headphones - Best Seller - Free Shipping!!!" gives it almost nothing.
The title optimization formula for AI:
[Brand] + [Product Name/Model] + [Primary Differentiator] + [Key Attribute 1] + [Key Attribute 2]
Keep it under 150 characters. No promotional language. No all-caps. No emoji. AI systems extract entities from titles, so every word should be a parseable attribute, not a sales pitch.
Description: Where AI Learns What Your Product Actually Does
This is where most feeds fall apart. The typical Shopify product description is written for a human browser: marketing fluff, brand voice, emotional language. That's fine for your product page. But your feed description needs to serve a different master.
AI systems use descriptions to answer specific questions: What problem does this solve? Who is the target user? What are the exact specs? How does it compare to alternatives? If your description doesn't contain these answers, the AI can't recommend your product for those queries.
My opinion: The best feed descriptions read like a knowledgeable friend explaining the product, not a copywriter selling it. "Grain-free dog food formulated for dogs with sensitive stomachs. Made with wild-caught salmon and sweet potato. 30lb bag serves a 50lb dog for approximately 6 weeks. No artificial preservatives. Vet-recommended for dogs with food allergies." That's what AI can work with.
GTIN / UPC / EAN: The Underrated Powerhouse
Most merchants treat GTIN as an annoying compliance field. For AI shopping, it's one of the most powerful signals you can provide. Here's why.
A GTIN lets AI systems do something they can't do with titles and descriptions alone: cross-reference your product against every other listing, review, and mention of that exact product across the entire internet. When ChatGPT has a GTIN, it can pull in Wirecutter reviews, Reddit threads, YouTube comparisons, and Amazon ratings for that specific item. Without a GTIN, the AI has to guess whether your "wireless earbuds" are the same ones that got a glowing review on RTINGS.
If you sell branded products, add GTINs to every single SKU. If you sell private-label products, apply for GS1 barcodes. The investment pays for itself in AI discoverability. For more on how structured data connects to AI visibility, see our guide on schema markup for AI and ChatGPT.
Brand: The Entity That Carries Authority
The brand field links your product to your brand's overall authority signal. AI systems don't evaluate products in isolation. They consider the brand's reputation across Reddit, YouTube, review sites, and news coverage. A strong brand field means the AI can connect your product to all those positive mentions.
Keep it consistent. "Nike" on one product and "Nike, Inc." on another fragments your brand signal. Use the exact same brand string across every product in your feed.
Product Type: Your Category Taxonomy
product_type tells AI systems which comparison set your product belongs to. When someone asks "best running shoes for flat feet," the AI needs to know which products in its index are running shoes. Google's product taxonomy has over 6,000 categories. Use the most specific one that applies.
"Apparel & Accessories > Shoes > Athletic Shoes > Running Shoes" is infinitely more useful to an AI than just "Shoes." The more specific your category path, the more precisely the AI can match your product to niche queries.
Is your product data AI-ready?
Check how AI shopping systems currently see your brand. Our free AI Authority Checker scans your visibility across ChatGPT, Perplexity, and Google AI Overviews in 30 seconds.
Check Your AI Visibility Score →Tier 2 Fields: Supporting Context That Strengthens Recommendations
Structured Attributes: Color, Size, Material, Age Group
These fields are how AI handles filtered queries. "Red leather wallet for men" requires the AI to match on color, material, and age/gender group simultaneously. If your feed only has a title and description, the AI has to infer these attributes from unstructured text, which it will sometimes get wrong.
Fill in every applicable attribute field. It takes 10 minutes per product type to set up in your feed management tool, and it dramatically increases match rates for specific queries.
| Attribute | Query Type It Enables | Example |
|---|---|---|
color | Color-specific searches | "navy blue backpack for school" |
size | Size-filtered comparisons | "13-inch laptop sleeve" |
material | Material preference queries | "stainless steel water bottle" |
age_group / gender | Demographic targeting | "gifts for teenage girls" |
pattern | Style-specific searches | "plaid flannel shirt" |
energy_efficiency_class | Sustainability / efficiency queries | "energy-efficient dehumidifier" |
Price and Availability: The Trust Gatekeepers
AI systems won't recommend a product with a stale price or an out-of-stock flag. That sounds obvious, but the implications are bigger than you'd think.
If your feed updates once a week and your prices change frequently, there's a window where AI systems have wrong price data for your products. When someone asks "best wireless earbuds under $100" and your feed says $89 but your site says $109, the AI either shows the wrong price (damaging user trust in the AI itself) or, more likely, drops your product from the recommendation entirely.
Update frequency matters. Daily at minimum. Real-time if you can manage it. Google's Shopping Graph processes over 45 billion product listings, and fresher data gets prioritized.
Image Link: Visual Understanding in Multimodal AI
This one is becoming more important fast. ChatGPT, Gemini, and Perplexity all have multimodal capabilities now. They can look at your product image and understand what the product is, what condition it's in, and whether it matches the description. A clear, well-lit product photo on a white background isn't just good for conversions. It's good for AI product understanding.
Use high-resolution images (at least 800x800px). Show the product clearly. Avoid heavy text overlays or collage-style lifestyle shots as your primary image. The AI needs to see the product itself.
What to Remove From Your AI Feed Strategy
My opinion: half the "feed optimization" advice online is written for Google Shopping Ads, not AI shopping. Here's what you can safely deprioritize if your goal is AI visibility rather than ad performance:
- Custom labels are purely for ad campaign segmentation. AI systems don't read them.
- Promotion IDs control Google Merchant Promotions in paid ads. No impact on organic AI recommendations.
- Ads redirect URLs are tracking parameters for ad clicks. AI shopping surfaces use your canonical product URL.
- Display ads ID and similar fields exist for remarketing campaigns. Completely invisible to AI.
- Shipping weight (as a standalone field) rarely factors into AI recommendations unless the query specifically involves shipping concerns.
I'm not saying to delete these fields. They still serve their ad-side purpose. But if you're spending time optimizing custom labels while your descriptions are two sentences and you have no GTINs, your priorities are wrong for the AI era.
Feed Optimization Checklist: AI vs. Ads
This is the comparison most merchants need. It shows exactly where ad-focused and AI-focused feed strategies overlap and where they diverge. Use this to audit your current feed setup.
| Optimization Area | For Google Shopping Ads | For AI Shopping |
|---|---|---|
| Title structure | Keyword-front-loaded, search-term focused | Brand + product + attributes, entity-focused |
| Description style | Marketing copy with keywords | Spec-rich, question-answering, structured |
| GTIN / UPC | Recommended for matching | Critical for cross-platform entity resolution |
| Custom labels | Essential for campaign segmentation | Irrelevant |
| Product type | Broad categories often sufficient | Most specific taxonomy path possible |
| Structured attributes | Nice to have for filtering | Essential for filtered AI queries |
| Update frequency | Daily for price/availability | Daily minimum, real-time preferred |
| Image requirements | Meets Merchant Center specs | Clear product shot for multimodal AI parsing |
| Promotion fields | Active for merchant promotions | Irrelevant |
| Review / rating data | Star ratings in ads | Sentiment signals that influence AI confidence |
The Description Formula That AI Systems Love
Good feed descriptions for AI follow a consistent structure. Here's the formula I'd recommend:
- Opening sentence: What the product is and its primary use case. One sentence. No fluff.
- Key features: 3 to 5 specific, measurable features. Materials, dimensions, capacity, performance specs.
- Target user: Who is this for? What problem does it solve? Be specific.
- Differentiator: What makes this different from competing products in the same category?
- Social proof hook: If applicable, a factual reference to awards, certifications, or expert endorsements.
This structure mirrors how AI systems break down product information when generating recommendations. They need to extract the "what," "who for," "why this one," and "is it credible" from your description. If those answers are embedded in the text in a clear, parseable way, the AI can work with it.
For a deeper look at how these structured product attributes connect to ChatGPT's Shopify product recommendations, that guide covers the full pipeline from feed to AI output.
Connecting Your Feed to On-Page Schema
Your product feed and your on-page Product schema markup should tell the same story. AI systems pull from both sources, and inconsistencies create confusion.
If your Merchant Center feed says "Navy Blue" but your Product schema says "Dark Blue," an AI system has to decide which one to trust. If your feed price is $49.99 but your schema says $54.99, the AI will likely drop your product rather than recommend potentially wrong information.
Sync your data sources. Your Shopify product data, Merchant Center feed, and on-page JSON-LD should all use identical values for title, price, availability, brand, GTIN, and attributes. Tools like Feedonomics, DataFeedWatch, and native Shopify feed apps can automate this sync. Our schema markup for AI guide covers the full implementation.
How Agentic Shopping Changes the Feed Game
We're entering a phase where AI shopping agents don't just recommend products. They browse, compare, and purchase on behalf of users. Shopify's push toward agentic storefronts means your product feed isn't just a data source anymore. It's the interface between your store and autonomous shopping agents. Our agentic storefronts guide covers this in full, but the feed implications are worth noting here.
My opinion: within the next 12 to 18 months, the quality of your product feed data will matter more than your website design for a growing segment of purchases. Agentic shoppers won't see your beautiful product page. They'll parse your structured data, cross-reference reviews, compare specs, and make a purchase decision programmatically. The stores with the cleanest, most complete product data will win those transactions.
This isn't speculative. Google Shopping Graph already processes over 45 billion product listings. OpenAI's shopping features in ChatGPT continue to expand. Perplexity launched a Buy button. The infrastructure for agent-driven commerce is being built right now, and your product feed is the foundation it sits on.
The 15-Minute Feed Audit
You can assess your feed's AI readiness in about 15 minutes. Pull up your Merchant Center feed export (or your Shopify product CSV) and check these items:
- GTIN coverage: What percentage of your SKUs have valid GTINs? If it's under 80%, that's your highest-priority fix.
- Description length: Are your feed descriptions at least 150 words? Anything shorter likely lacks the structured detail AI needs to recommend confidently.
- Product type specificity: Are you using the most granular Google product category available? Check against the full taxonomy.
- Attribute completeness: For apparel and accessories, do you have color, size, material, gender, and age group filled in? For electronics, do you have specs like capacity, connectivity, and compatibility?
- Title structure: Do your titles follow the Brand + Product + Attributes pattern? Or are they stuffed with promotional language?
- Price/availability freshness: When was your feed last updated? Is it syncing at least daily?
- Schema consistency: Do your feed values match your on-page Product schema exactly?
If you find gaps in three or more of these areas, your feed is probably invisible to AI shopping systems. The good news: these are all fixable within a week.
See how AI shopping systems view your brand right now
Your feed is only part of the picture. Brand authority, review sentiment, and content presence all factor into AI recommendations. Run a free scan to see your full AI visibility profile.
Run Your Free AI Visibility Scan →Frequently Asked Questions
Which product feed fields matter most for AI shopping recommendations?
Title, description, GTIN/UPC, product_type, brand, and structured attributes like material, size, and color are the highest-impact fields. AI systems parse these to understand what a product is, who it's for, and how it compares to alternatives. Fields like custom_label or ads-specific columns have zero impact on AI recommendations.
Do AI shopping assistants use Google Merchant Center feeds?
Google AI Overviews and Shopping Graph pull directly from Merchant Center feeds. ChatGPT and Perplexity rely more on on-page structured data (JSON-LD Product schema) and web-scraped product information. Optimizing both your feed and your on-page schema covers all major AI shopping surfaces.
How should I write product titles for AI shopping?
Lead with the brand name, then the product name, then the most important differentiating attributes (size, color, material, use case). Keep titles under 150 characters. Avoid keyword stuffing or promotional language like "best seller" or "free shipping" in the title field. AI systems extract entity-level data from titles, so clarity beats cleverness.
Does adding GTIN/UPC to my product feed help with AI visibility?
Yes. GTINs let AI systems cross-reference your product against global product databases, match it to reviews from other platforms, and deduplicate listings. Products with valid GTINs are significantly more likely to appear in AI shopping comparisons because the AI can verify they're real, specific items rather than vague listings.
What is the difference between product feed optimization for ads vs AI?
Ad-focused feed optimization prioritizes fields like custom_label, promotion_id, and sale_price to control bidding and ad targeting. AI-focused optimization prioritizes description quality, structured attributes, GTIN, product_type taxonomy, and review data because AI systems use these fields to understand and recommend products. The two strategies overlap on titles and images but diverge on almost everything else.
How often should I update my product feed for AI shopping?
At minimum daily for price and availability changes. For description and attribute updates, push changes within 24 hours of any product modification. AI systems like Google Shopping Graph refresh frequently, and stale data (especially wrong prices or out-of-stock items) can damage your trust signals across all AI surfaces.