Here's a framing shift that changes everything: your Microsoft Merchant Center feed is no longer just a Shopping campaign data source. It's your Copilot presence. When a user asks Copilot for product recommendations in your category, Copilot reaches into your feed to assemble its response.

The quality of that response — what products Copilot recommends, how it describes them, whether it recommends you at all — depends almost entirely on how well your feed is structured.

How Copilot Reads Your Feed

Copilot's product recommendation engine synthesises from multiple sources: your Merchant Center feed, organic web content, and user conversation context. For in-conversation commerce (Showroom Ads, Copilot Checkout), the feed is the authoritative source. For organic recommendations (where Copilot mentions your brand without a paid placement), both the feed and your organic Bing presence contribute.

Think of the feed as the "product brief" you give to Copilot. Vague, incomplete briefs produce vague, unconvincing recommendations. Precise, structured briefs produce specific, compelling product placements.

Title Optimisation: The Highest-Impact Field

Product titles are the single most impactful feed field for both Shopping performance and Copilot recommendation quality. The problem: most advertiser titles are optimised for how they look to a human scanning a search results page. AI reads them differently.

Copilot-optimised title structure:

[Brand] [Product Type] [Key Attribute 1] [Key Attribute 2] [Variant]

Example: "Sony WH-1000XM5 Wireless Noise-Cancelling Headphones Over-Ear Bluetooth — Black"

What to avoid:

  • All-caps brand names that disrupt natural language parsing
  • Keyword stuffing that reads unnaturally in AI-generated sentences
  • Truncating important attributes to stay under character limits — expand titles to use the full 150 characters
  • Model numbers without descriptive context (Copilot can't explain why XYZ-4420 is relevant to a user's question)

Description Optimisation: Write for AI Comprehension

This is where most feeds fall shortest, and where the opportunity is greatest. Product descriptions in the average feed are either:

  • A single-sentence marketing line ("The most advanced widget yet!")
  • A keyword list ("blue widget, best widget, discount widget, widget sale")
  • A copy-paste from the manufacturer that's already present on hundreds of other sites

None of these serve Copilot well. Write descriptions as structured paragraphs that AI can cite directly:

Template: "[Product name] is a [category] designed for [primary use case]. It features [key attribute 1], [key attribute 2], and [key attribute 3]. [Product] is compatible with [relevant platforms/standards] and is available in [variants]. [One sentence on what makes it distinctive vs alternatives]."

Structured Attributes: The Often-Ignored Goldmine

Beyond title and description, these fields have a disproportionate impact on Copilot recommendation matching:

  • product_type: Use your full category path ("Headphones > Wireless > Noise-Cancelling"). Copilot uses this for intent matching.
  • google_product_category: Use the most specific matching category from the Google Product Taxonomy. This is the structured ontology Copilot uses to understand what your product is.
  • age_group, gender: Critical for apparel and lifestyle products. Copilot uses these to filter recommendations by user profile.
  • material, pattern, size_type: Any attribute that a user might specify in natural language should be in the feed. "I need a waterproof hiking boot in a wide fit" requires material:waterproof and size_type:wide to match.

Feed Freshness: The Underestimated Signal

Copilot's product recommendations weight recently-updated feed data more heavily than stale data. This is because stale data (especially stale prices) creates bad user experiences, and Microsoft has built freshness as a quality signal.

Implementation hierarchy by update frequency need:

  1. Price and availability: Real-time via Content API if possible, minimum daily
  2. Promotions and sale prices: Real-time or daily with sale_price_effective_date populated accurately
  3. Descriptions and attributes: Weekly or when changed
  4. Images: Update when product photography changes; stable images don't need frequent refreshing