What content format does Microsoft Copilot Shopping actually prefer when deciding which product to surface, compared to what works for ChatGPT?
Microsoft Copilot Shopping runs on merchant feeds first, where ChatGPT runs on page retrieval. Microsoft requires a UCP-compliant product feed in Microsoft Merchant Center to power product discovery and checkout, and states it uses both web information and the merchant feed (per Microsoft Advertising, September 2026). ChatGPT reaches products through OAI-SearchBot and, on Shopify, through Shopify Catalog.
| Microsoft Copilot Shopping | ChatGPT | |
|---|---|---|
| Primary input | Microsoft Merchant Center feed, UCP-compliant: "required to power product discovery and checkout" (per Microsoft, Sep 2026) | Page retrieval via OAI-SearchBot (per OpenAI, Sep 2026) |
| Secondary input | Web page information (per Microsoft, Sep 2026) | Shopify Catalog attributes on Shopify stores (per Shopify, Sep 2026) |
| Checkout | Copilot Checkout, in-surface, merchant remains merchant of record | Referral to your own checkout in a ChatGPT in-app browser (per Shopify, Sep 2026) |
| Named feed fields | Native checkout eligibility, product warnings, merchant item ID, return policy, customer support info (per Microsoft, Sep 2026) | — |
| Current eligibility | English-language merchants selling to US buyers in USD; Copilot.com and the Copilot mobile app (per Microsoft, Sep 2026) | — |
The verdict, and the profile it's right for. Treat Copilot as a feed channel and ChatGPT as a content channel. A US-selling mid-market merchant with a clean Merchant Center feed already meets most of Copilot's requirement; the work is the named structured fields (return policy, support contact, merchant item ID), not page copy. Non-US merchants get no Copilot Checkout surface today and should spend the quarter on ChatGPT and Google instead.
The trap. A spec-perfect feed built on thin product data is still thin to any agent that reads the page. Feed fields and page content draw from the same attribute model; filling the feed with placeholder values to pass validation produces a listing that validates and loses.
Where the two converge. Both surfaces need the same thing underneath: complete, structured, accurate attributes with a single owner. Shopify Catalog Mapping exists specifically so custom fields outside the standard attributes can be mapped through (per Shopify's Help Center, September 2026).
The Deploi point of view
Our own position, from building on Shopify. Separate from the facts above.
- Our take: Copilot is a feed-readiness project; ChatGPT is a content project; both draw from one attribute model. We scored agentic commerce readiness CUSTOMIZE: build the data foundations, wait on protocol bets for exactly this reason.
- What we’ve seen: The hidden half of every feed project is attribute backfill. Defining the metafields takes days; populating them across a live catalog takes the quarter, and it is the line that gets cut from the estimate first and blows the timeline last.
- Where we disagree: The feed-app category sells rules engines as the fix for feed quality. A rules engine reshuffles fields; it cannot invent an attribute your catalog never captured.
- What this page adds: that one of the four major engines is genuinely feed-driven rather than page-driven, and the named fields Microsoft requires.
Reviewed by Martin Dejnicki, Director of SEO & AI Search. Facts verified 2026-09-13.