When AI Channels Disagree: Prioritizing Product Feed Fields for ChatGPT, Perplexity, and Google

By Steve Merrill, Founder of WRKNG Digital — April 4, 2026

Three AI channels. Three different product graphs. And if you're running a Shopify store right now, you're almost certainly feeding all of them with the same product data you set up two years ago and never touched.

ChatGPT Shopping, Perplexity, and Google AI Overviews don't pull product information from the same place. They have different architectures, different data sources, and different ideas about what makes a product listing trustworthy. When you've got limited time to improve your feed (and everyone does), the instinct is to build a checklist that covers everything. That checklist usually covers nothing well.

This is the prioritization framework I'd follow. Based on where the channels actually overlap, where they diverge, and what moves the needle fastest.

Why Are ChatGPT, Perplexity, and Google Using Different Product Data?

They're built on different infrastructure. That's the short answer, and understanding it makes the prioritization obvious.

Google AI Overviews and Google Shopping run on Google's Shopping Graph, a massive product knowledge base that ingests data from Google Merchant Center feeds, on-page Product schema, and Google's own web crawl. Your Merchant Center feed is the primary input. On-page schema is a secondary confirmation layer. When the two conflict, the feed wins.

ChatGPT Shopping, which OpenAI rolled out as part of its broader search expansion, pulls product data primarily from Bing Shopping. Bing's shopping index overlaps significantly with Google's because many brands that submit feeds to Google Merchant Center also get picked up by Microsoft's shopping index. OpenAI layers its own web crawl on top to fill gaps and validate product information in real time. OpenAI has described this crawl layer as part of how ChatGPT Search grounds responses in current, indexed content.

Perplexity works differently. It's primarily a web crawler. When someone asks Perplexity about a product, it reads pages in near real time and surfaces what it finds there. It also integrates with some shopping data partners, but the on-page experience matters more here than on either of the other channels. What's on your product page, your title, description, price, reviews, structured markup, is what Perplexity uses to build its answer.

The channel architecture, simplified:

• Google: Merchant Center feed + on-page schema (feed is primary)
• ChatGPT: Bing Shopping (overlaps with Google) + OpenAI crawl
• Perplexity: Direct web crawl + shopping integrations (on-page schema is primary)

That overlap between Google and ChatGPT is significant. Fixing your Merchant Center feed has a multiplier effect on Bing and, by extension, on ChatGPT Shopping. But Perplexity reads your pages directly, which means it's a separate track that needs separate attention.

Which Feed Fields Actually Overlap Across All Three Channels?

Start here. These fields matter for every AI channel, and getting them right affects Google, ChatGPT, and Perplexity simultaneously.

Title. The most important field in any product feed. Titles that include brand, product type, and one or two specific differentiators (material, size, color) outperform generic titles across all three channels. Google's Merchant Center product data specification treats title quality as a primary ranking factor in Shopping, and the same logic carries over to AI shopping surfaces where the title is often the first and only thing that gets read before a citation decision is made.

Price and availability. If your listed price doesn't match the price on your product page, Perplexity will surface whatever it reads directly. Google will flag the discrepancy and may disapprove the item. ChatGPT, pulling from Bing, may serve stale data for longer but will eventually reflect the mismatch. Accuracy here is the floor, not the ceiling.

Description. This one gets underestimated. AI models use product descriptions to understand what a product does and who it's for. A description that reads like a spec sheet won't answer a conversational query as well as one that explains the benefit clearly. "2.4GHz wireless connectivity, 10-hour battery life" tells a machine what something has. "A wireless mouse built for all-day desk work, with a 10-hour battery that gets through a full workday without a charge" tells a model what it's for. ChatGPT weights natural language quality when it decides which products to surface in a response.

Images. High-resolution images with clean backgrounds are required for Google Shopping. They also surface in ChatGPT and Perplexity shopping cards. Low-quality images reduce click-through rates across every channel and can trigger Merchant Center disapprovals that cascade into Bing and ChatGPT.

Brand. All three channels use brand as an entity recognition signal. "Nike" and "Nike, Inc." are not the same entity in a product graph. Pick one format and stay consistent across every field and every page.

GTIN or MPN. Global Trade Item Numbers let channels match your product to a known entity in their product graph. Google requires GTINs for branded products and will limit visibility for any listing that's missing one. Bing Shopping uses them for product matching, which means ChatGPT Shopping inherits that signal. If you're selling branded products without GTINs in your feed, you're losing product-entity matches across every channel at once.

What Does Perplexity focus on That Google's Merchant Center Doesn't?

On-page schema markup. Full stop.

Google's Merchant Center is a data submission pipeline. You submit a feed, Google ingests it, the feed drives most of what surfaces in Shopping. Your schema.org Product markup on the product page is a secondary confirmation layer. When the two conflict, the feed typically wins in Google's system.

For Perplexity, the page is the source of truth. There's no feed to submit. That means your Product schema has to be right, complete, and consistent with what's displayed on the page. At minimum, Perplexity needs to find:

• name
• description
• image
• offers with price, priceCurrency, and availability
• aggregateRating if you have reviews (you should have reviews)
• brand

I've audited stores where the Merchant Center feed was clean, approved, and performing well in Google Shopping, but on-page schema was either missing entirely or pulling from an outdated theme snippet that was still populating the original launch price. Those stores weren't appearing in Perplexity for any product-level queries. The feed looked fine. The problem was on the page. Not great.

AggregateRating markup is the specific field that gets skipped most often. If you have product reviews in your store and they're not surfacing in your schema, Perplexity won't pick up that social proof signal. Perplexity surfaces rating information prominently in shopping responses, and products without visible ratings lose out to products that have them, even when the underlying product is better.

When You Can't Do Everything at Once, What's the Right Order?

Ruthless prioritization. Here's the sequence that produces the broadest coverage fastest, based on how the channels overlap.

Step 1: Fix titles on your top 50 products. Audit them. If the title doesn't include brand, product type, and at least one specific differentiator, rewrite it. This single change touches Google, Bing/ChatGPT, and Perplexity simultaneously because all three display the title and use it for product matching.

Step 2: Add or correct GTINs. Pull a Merchant Center report filtered for missing GTINs on branded products. Fill them in. This is usually a one-time fix per SKU, and it improves product-entity matching across Google and the Bing/ChatGPT pipeline at the same time.

Step 3: Rewrite descriptions for natural language. Focus on your top 20 products by revenue first. Replace spec-sheet language with benefit-oriented descriptions of 150 to 300 words. Write to answer the question a real shopper would ask an AI assistant. "What's the best wireless mouse for long work days?" should land on a page with a description that answers that question directly.

Step 4: Audit on-page Product schema. Use Google's Rich Results Test on your top 10 product pages. Check that price, availability, and brand are populating correctly. Fix any schema that's pulling stale or incorrect data. This directly improves Perplexity visibility and reinforces your Google feed data as a secondary signal.

Step 5: Add AggregateRating markup. If you have product reviews and they're not in your schema, that's a missed citation signal on every channel. This often requires a theme-level or app-level fix, but it's worth prioritizing for your top products.

Step 6: Resolve image quality flags. Pull your Merchant Center diagnostics. Any image flagged for quality or resolution needs to be fixed. The same images surface in ChatGPT and Perplexity shopping cards, and a disapproval in Merchant Center means the product may not render correctly downstream.

How Do You Know If Your Changes Are Actually Working?

Measure before you touch anything. Most stores don't have a baseline, so they can't tell whether changes worked or whether the channel just had a good week.

Before making any changes, record your Merchant Center diagnostics: approved product count, active disapproval reasons, and item quality scores. Run 5 to 10 product-specific queries in Perplexity and note whether your products appear. Check ChatGPT Shopping for your brand name plus your main product categories. Screenshot everything. Date it.

After changes, give Google 2 to 3 weeks to re-index. Perplexity moves much faster. I've seen stores go from zero Perplexity citations to consistent appearances within 10 days of fixing on-page schema. ChatGPT Shopping lags because it's downstream of Bing's indexing cycle. Give it 3 to 4 weeks before drawing conclusions.

The channels will always have some disagreement about which products to surface. That's the architecture, not a flaw in your setup. Your job is to make sure the fields they all share are accurate, complete, and written for how AI models actually use them. The bar is genuinely low right now.