How to optimize your product feed for AI shopping results

Google Shopping used to have one gatekeeper: the Shopping algorithm. You optimized your feed for it and that was the job.

Now there are more gatekeepers. Google's AI Overviews show product picks inside search results. ChatGPT recommends specific products when people ask it what to buy. Gemini does the same. Each one selects a short list of products to show - and your feed data is a big part of how they choose.

Here is the good news. You do not need a separate strategy for each surface. The same clean, complete, honest product data wins everywhere. This post covers exactly what that looks like.

What are AI shopping results?

The shift matters because of how few products get shown. A classic Shopping page gives you a long carousel and rows of listings. An AI answer might show three to five products. When the shelf is that small, getting picked is everything.

And here is the part most store owners miss. These AI surfaces feed from data you already control. Google's AI features pull from the Shopping Graph, which is built largely on Merchant Center data. ChatGPT reads product data from the open web and from merchant feeds. Your feed was already your targeting in classic Shopping. Now it is also your pitch to every AI shopping assistant.

If you are weighing where ad budget goes across these platforms, that is a separate question - we cover it in ChatGPT ads vs Google and Meta ads. This post is about the organic side: making your products the ones AI picks.

Complete attributes are the entry ticket

Classic Shopping was forgiving about gaps. Leave material blank and your product still served - it just matched slightly worse.

AI surfaces are not forgiving. Think about what the AI is doing: someone asks for "polarized sunglasses under 100 with a metal frame," and the system looks for products it can confirm match every part of that request. If your feed does not say polarized, does not say metal, the AI cannot confirm your product fits. So it recommends one it can confirm.

A blank attribute is not neutral anymore. It is a reason to skip you.

Go through your feed and fill everything that applies:

• Brand
• GTIN and MPN
• Color, size, material, pattern
• Gender and age group where relevant
• Product type, set as deep as the taxonomy allows
• Product highlights
• Condition, availability, shipping, and return details

Boring work. But it is the entry ticket. Products with gaps do not lose the AI auction - they never enter it.

Write titles AI can parse

Titles still lead. But the way to think about them has shifted slightly.

Classic Shopping title advice was keyword coverage: fill the 150 characters with terms buyers search. That still holds. The AI layer adds a second requirement: the title has to parse cleanly into brand, product type, and attributes.

A structure that works for both:

Brand + product type + the key attribute + supporting attributes.

• "Rayve Aviator Sunglasses, Polarized, UV400, Gold Metal Frame" - a machine can pull brand, type, lens tech, and material out of that in one pass.
• "Summer Vibes Only - Cool Shades for the Beach" - a machine gets nothing. No brand, no product type, no attribute. That title is invisible to an AI matching a specific request.

The key attribute is whatever actually decides the purchase in your niche. Polarized or UV400 for eyewear. Solid oak versus veneer for furniture. 14k solid versus plated for jewelry. Whatever your buyer filters on, it belongs early in the title.

If you want the full workflow for rewriting titles at scale with AI doing the drafting, the AI feed optimization guide walks through it step by step - prompts, editing, and how to push the results live without touching your storefront.

GTINs and identifiers: how AI knows what your product is

A GTIN is the one attribute that removes all doubt about what you are selling.

When your product has a valid GTIN, Google can connect it to everything else known about that exact product - reviews, price history, competing sellers, spec sheets. AI surfaces lean on that connected picture when deciding what to recommend. A product with a GTIN is a known object. A product without one is a string of text the AI has to interpret on its own.

If your products have GTINs, submit them. If you sell your own manufactured products without GTINs, set identifier_exists correctly and make brand and MPN airtight - do not fake a barcode, that gets products disapproved. And if disapprovals are already eating your catalog, run through the Merchant Center disapproval checklist before you do anything else. An AI surface cannot recommend a product that is not approved to serve.

The fields AI reads that classic Shopping mostly ignored

Here is where the game actually changed.

In classic Shopping, the description carried a small share of the match signal. Plenty of stores shipped one-line descriptions for years and got away with it.

AI shopping surfaces read the whole record. When ChatGPT or Gemini explains why it picked a product - "this one has a lightweight frame and comes with a hard case" - that reasoning comes from somewhere. It comes from descriptions, product highlights, and the product page itself.

Two fields deserve real work now:

Description. You get up to 5,000 characters. Use them for facts, not fluff. What it is made of, what it does, what is in the box, what problem it solves, care instructions, fit notes. Every concrete fact is something an AI can cite when recommending you. Every line of empty marketing copy is dead weight.

Product highlights. This attribute exists for short, scannable selling points - and it maps almost perfectly onto how AI answers are written. "Polarized lenses block glare." "Folds flat for travel." "Handmade in Portugal." Write four to six per product. Most stores leave this field empty, which makes it a cheap edge right now.

Price still decides more than you think

AI shopping answers are built around user intent, and user intent almost always includes price. "Best running jacket under 150" is a price question as much as a product question.

Two things follow from that:

First, your price data has to be right. Wrong prices in the feed are a classic disapproval trigger, and an unavailable or mispriced product will not get recommended anywhere.

Second, competitiveness matters. Google knows the market price for your product category, and AI surfaces answering "best X for the money" lean toward products priced sensibly against comparable items. You do not have to be the cheapest. You do have to be explainable. If you charge more, the reason needs to live in your feed and on your page - better material, included extras, stronger warranty - so the AI has something to justify the price with.

Structured data on the page has to match the feed

AI surfaces do not stop at the feed. They also read your product pages - and structured data (schema markup) is how a machine reads a product page.

The rule is simple: your Product structured data must say the same thing as your feed. Same price, same availability, same GTIN, same title meaning. When the feed and the page agree, the machine trusts both. When they disagree, that is a data-quality problem - and it can trigger Merchant Center price and availability mismatch issues on top of hurting your AI visibility.

Check three things on every template:

1. Product schema exists on every product page and validates.
2. Price and availability in the schema update when the product changes - not hardcoded values from launch day.
3. The identifiers in the schema match the identifiers in the feed.

Most Shopify themes handle the basics, but supplemental feed overrides are a common gap: if a tool rewrites your feed title, the page schema keeps the original. That is fine - titles may differ in wording - but price, availability, and identifiers must never disagree.

Why feed hygiene now compounds

Here is the frame that makes all of this worth doing.

Every fix above - complete attributes, parseable titles, GTINs, real descriptions, matching structured data - was already best practice for classic Shopping. Better data meant better matching, fewer disapprovals, stronger Smart Bidding signals.

AI surfaces did not replace that. They stacked on top of it. The same clean record now works two jobs: it targets your Shopping and Performance Max ads, and it gets you picked by AI Overviews, ChatGPT, and Gemini. One hour of feed work now pays out across every surface where products get discovered.

That also means the gap between clean feeds and messy feeds is wider than it has ever been. A messy feed used to cost you some match quality. Now it costs you match quality and a seat on the AI shortlist.

If you want to do this with tooling instead of by hand, we broke down what each tool category actually does well in the feed optimization tools guide. And if you would rather have the whole thing run for you - feed engineering, campaign structure, tracking - that is what our Google Shopping management covers day to day for 200+ ecom brands.

Frequently asked questions

How do I get my products into ChatGPT shopping results?

ChatGPT builds its shopping answers from product data it finds on the open web and from merchant product feeds. You cannot pay your way in - it works from data quality. The play is the same one that wins on Google: a complete, accurate feed, strong titles, real descriptions, and structured data on your product pages that matches what the feed says. Clean data gets read. Messy data gets skipped.

Does Gemini use my Merchant Center feed?

Google's AI shopping experiences - AI Overviews with product results, AI Mode, and shopping answers in Gemini - pull from Google's Shopping Graph. The Shopping Graph is built largely on Merchant Center data. So yes, the feed you already maintain for Shopping ads is the same data these AI surfaces read when they pick products to show.

Do I need a separate product feed for AI shopping?

No. There is no separate "AI feed" to upload. AI surfaces read the same product data as classic Shopping - they just read more of it and punish gaps harder. The work is raising the quality bar on the feed you already have, not building a second one.

What is the single most important feed fix for AI shopping results?

Completeness. Fill every attribute that applies to your products - GTIN, brand, color, size, material, gender, product highlights. AI systems answering a specific question pick products they can fully understand. If your feed leaves an attribute blank, the AI cannot confirm your product matches, so it recommends one it can confirm.