ChatGPT, Gemini and Copilot answer the same shopper questions but pull product information from different places, so the right first move differs for each: Google Merchant Center for Gemini and AI Mode, Microsoft Merchant Center for Copilot, and a complete catalog on your platform for ChatGPT. To the shopper they look alike. For a store, where the data comes from decides where to invest.
This post compares how the three assistants source products and what to do about each. The feed-level comparison (accounts, order of setup, free versus paid) lives in Google vs Microsoft Merchant Center vs ChatGPT Shopping. If you want the data foundation handled for you, see our Google Merchant Center agency page.
Google Gemini and AI Mode
Where products come from: Google’s own product data, built largely from Google Merchant Center feeds and the web.
What it means for stores: if your products are in Merchant Center with complete, accurate data, you are already in the pool Google’s assistants draw from. In January 2026 Google also introduced the Universal Commerce Protocol with Shopify, to let agents work with a store’s catalog and cart.
What we would check first: the same things Google Shopping has always punished or rewarded, namely titles, identifiers, attributes and a price and stock that match the page. Conversational queries are long and specific, so attribute coverage matters more here than it did for short keyword searches.
Microsoft Copilot
Where products come from: Microsoft’s shopping data, fed by Microsoft Merchant Center, plus catalogs from commerce platforms.
What it means for stores: Copilot added in-chat checkout in January 2026 with PayPal, Shopify and Stripe. A Microsoft Merchant Center feed, often imported from Google, is the direct way in.
What we would check first: whether a Microsoft Merchant Center account exists at all. Many stores run excellent Google feeds and have never opened one, which means they are absent from this assistant by default. The steps are in our Microsoft Merchant Center and Copilot setup guide.
ChatGPT
Where products come from: structured catalogs from platforms and merchant programs, plus information on the web.
What it means for stores: OpenAI launched Instant Checkout in September 2025, but has since been reported to scale back its in-chat shopping plans. Recommendations still matter: shoppers research in ChatGPT and buy elsewhere. Complete, accessible product data is what gets you recommended.
What we would check first: whether your pages are crawlable, whether each product page shows price, availability and structured data, and what your ecommerce platform already offers. The routes are laid out in how to get your products into ChatGPT Shopping.
Side by side
| Gemini / AI Mode | Copilot | ChatGPT | |
|---|---|---|---|
| Main product data source | Google Merchant Center and the web | Microsoft Merchant Center and platform catalogs | Platform catalogs, merchant programs and the web |
| In-chat checkout | Through UCP partners | Copilot Checkout | Instant Checkout, being reshaped |
| First step for a store | Google Merchant Center | Microsoft Merchant Center | A complete catalog on your platform |
| Common blind spot | Approved but thin feed | No Microsoft account at all | Blocked crawlers or missing product-page data |
How can I test what each assistant says about my products?
Run the same set of shopper questions through all three assistants and log the results. This is a diagnostic, not a ranking report: answers vary by user, session and time, so one test proves little and a pattern across several is what counts.
- Write 10 questions a real customer would ask, mixing broad («best waterproof jacket for hiking»), specific («women’s waterproof jacket under $200») and brand-led («is [your brand] worth it»).
- Ask each question in ChatGPT, Gemini or AI Mode, and Copilot, using a fresh session each time.
- For each answer, record whether your store or products appeared, which products were named instead, and what price and availability were shown.
- Mark every wrong fact: old price, discontinued item, wrong specification, missing variant.
- Trace each wrong fact to its source: the feed, the product page or a third-party site.
- Fix at the source, then repeat the same questions a few weeks later and compare.
Treat any single appearance as luck and any recurring wrong fact as a data bug. The second is the one you can fix.
What works for all three?
The assistants differ, but they reward the same thing: product data that agents can read. Titles that say what the product is, every attribute filled, real-time price and stock, identifiers and clear policies. The full comparison of the feeds behind them is in Google vs Microsoft Merchant Center vs ChatGPT Shopping.
And because many shoppers still finish the purchase on the store’s website, read Discover in AI, buy on your site. For the basics, start with agentic commerce explained.
What does this look like for a real catalog? (A hypothetical example)
Imagine a store with a few thousand SKUs of home office furniture. The 10-question test shows the same pattern in every assistant: a competitor’s desk is recommended with dimensions and material, while the store’s equivalent desk is missing or described vaguely. This is an illustration, not a real account.
Tracing it back, the store’s feed has titles like «Model D-200», the dimensions live only in an image, and the material is blank. The fix is the same for all three assistants: rewrite titles to include product type and key attribute, fill the dimension and material attributes, and confirm the product page repeats the same facts. One data fix improves every channel at once, which is why we start with the catalog rather than with any one assistant.
What mistakes should I avoid?
| Mistake | Symptom | Better approach |
|---|---|---|
| Optimizing for one assistant only | Strong in Google, absent from Copilot or ChatGPT | Fix the shared catalog, then adapt per channel |
| Judging results from one prompt | Conclusions change every time you ask | Use a fixed set of questions and compare over time |
| Fixing the assistant’s answer instead of the source | The same wrong fact keeps coming back | Trace it to the feed or page and correct it there |
| Blocking crawlers by accident | Product pages never appear as sources | Review robots.txt and bot-protection rules |
When is it not worth worrying about each assistant separately?
If your Google feed is weak, working on Copilot or ChatGPT specifics is premature, since the same gaps will appear everywhere. And if most of your sales come from a marketplace, the marketplace listing is the data these assistants will see most. In both cases, fix the base first.
What should I ask my team or agency before starting?
- Which system holds the source of truth for titles, attributes, price and stock?
- Do we have a Microsoft Merchant Center account, and who owns it?
- What does our ecommerce platform already publish to AI assistants, and do we need to opt in?
- Do our robots.txt and bot-protection settings allow the crawlers of the assistants we care about?
- How will we log the 10-question test so results are comparable month to month?
- What is the plan when an assistant shows a wrong fact, and who fixes it at the source?
How Avafa works
Avafa is a consultancy based in Spain that works remotely with US ecommerce brands, manufacturers, distributors and retailers. We audit your catalog, build the feed in Google Merchant Center, extend it to Microsoft Merchant Center and prepare the product data for AI assistants. We do not promise placement in any assistant’s answers. Our Google Merchant Center agency service covers the foundation, and the Microsoft Merchant Center agency page covers the Copilot side.
FAQs
Which assistant matters most for US stores?
It depends on your audience, but Google remains the largest source of shopping discovery, so Merchant Center comes first.
Do I need a different catalog for each assistant?
No. One well-built catalog can be distributed to all of them, with small adjustments per channel.
Why does an assistant show the wrong price for my product?
Usually because it read an outdated or inconsistent source: an old feed value, a cached page or a third-party site. Compare the price in your feed, on your page and in your structured data, and correct whichever disagrees.
Can I control which products an assistant recommends?
Not directly. You control the quality, completeness and accessibility of your data, and the assistant decides what to show.
Is Copilot worth the effort if I already have a Google feed?
The extra effort is usually small because Microsoft Merchant Center can import from Google. Whether the traffic justifies it depends on your audience, which you can check in your analytics.
