Meta's Muse Pulls 67% of Picks From 2 Sites: What Now?

Why is Meta's Muse ignoring most retailers?

Meta's Muse AI shopping agent is supposed to help users discover products across the web. In practice, it's picking favorites fast. Profound's test ran 50 shopping prompts across a catalog of 60 products from 30 merchants and found that Muse returned 6 picks total, with 4 of those coming from just 2 websites. That's a 67% concentration rate from 6.7% of the merchants tested.

For SMB retailers, that number should stop you cold. This isn't a traffic distribution problem. It's a visibility problem at the agent layer, and it's one most small operators don't even know exists yet.

How does an AI shopping agent actually choose products?

AI shopping agents like Muse don't browse the way a human does. They parse structured signals: schema markup, product data feeds, page load speed, canonical URLs, structured descriptions, and how cleanly a page communicates what a product is, who it's for, and what it costs.

If your product pages were built to rank on Google, they were built for a different system. Google reads pages and infers meaning. AI agents need meaning handed to them explicitly. A page that says "premium handcrafted leather goods" without a clear product type, price, material, and use case structured in machine-readable format is going to get passed over.

The two sites that dominated Muse's picks in Profound's test almost certainly have one thing in common: their product data is clean, complete, and structured in a way agents can parse without guessing.

Can retailers even tell when Muse visits their pages?

This is where it gets operationally uncomfortable. According to the Profound research, there's genuine uncertainty about whether retailers can even detect Muse's agent crawl in their analytics. Standard web analytics tools like GA4 log human sessions and most known bots, but AI shopping agents may not identify themselves in ways your current stack catches.

If you can't see the agent, you can't measure whether your pages are performing for it, and you can't run experiments to improve. You're flying blind in a channel that's already making purchase recommendations to real users.

This is not a hypothetical future problem. Muse is live. The picks are happening now.

What does this mean for SMB retailers specifically?

Large retailers have dedicated SEO and feed management teams who will adapt to this fast. They'll update their structured data, test their product schema, and monitor agent behavior. Many are already doing it.

SMB retailers typically don't have that infrastructure. Most are running Shopify or WooCommerce with whatever schema the theme ships with, which is often incomplete. Product descriptions are written for humans, not for machine parsing. Feed hygiene is inconsistent.

The risk isn't that AI shopping agents will hurt your existing traffic. Your existing Google traffic won't disappear tomorrow. The risk is that a new, high-intent discovery channel is forming right now, and the early winners will be whoever gets their product data right first. Concentration effects in algorithmic systems tend to compound. The two sites dominating Muse's picks today will accumulate more signals, more clicks, and more data advantage over time.

What signals actually matter for AI shopping agent visibility?

Based on what we know about how LLM-based agents parse product pages, here are the factors most likely to influence whether your product gets picked:

| Signal | Why It Matters | Quick Check | |---|---|---| | Product schema (JSON-LD) | Agents read structured data before page copy | Use Google's Rich Results Test | | Price and availability in schema | Agents need current, machine-readable pricing | Check schema output matches live price | | Product type specificity | Vague categories hurt matching | "Running shoe" beats "footwear" | | Page load speed | Slow pages may get skipped or partially parsed | Core Web Vitals in Search Console | | Canonical URL cleanliness | Duplicate signals confuse agent resolution | Audit for canonical errors | | Clear, literal product descriptions | Agents don't infer; they match | Rewrite flowery copy to be direct |

None of this is exotic. It's mostly table stakes technical SEO that many SMB product pages haven't fully implemented. The difference now is the stakes are higher because you're optimizing for a system that makes direct purchase recommendations, not just rankings.

How should retailers think about feed management going forward?

If you're selling physical products, your product data feed is now a marketing asset, not just an ops file. The same feed that powers your Google Shopping ads, your Meta catalog, and your Bing placements is increasingly the input that AI shopping agents will pull from.

Feed quality variables that matter:

• Title structure: Lead with the most specific descriptor. "Men's Merino Wool Base Layer, Midweight, Navy" outperforms "Cozy Winter Top."
• Description completeness: Include material, dimensions, use case, and compatibility where relevant. Every field an agent might use to match a query should be filled.
• GTIN and MPN fields: These unique identifiers help agents confirm product identity across sources. If you're not populating them, do it.
• Category taxonomy: Use Google's product taxonomy as your baseline. It's what most agent systems were trained against.

The retailers who win in AI-mediated shopping won't necessarily have the best products. They'll have the best-described products.

What we'd actually do

• Audit your product schema this week. Run your top 10 product URLs through Google's Rich Results Test and Schema.org's validator. Fix any missing or malformed fields before anything else. This is the highest-leverage single action you can take right now.
• Clean your product feed before your next catalog update. Prioritize title specificity, GTIN/MPN population, and complete attribute fields. If you're on Shopify, tools like Feedonomics or DataFeedWatch can automate a lot of this. If you're managing feeds manually, build a QA checklist and run it on every SKU.
• Set up agent traffic monitoring. Work with your developer or analytics provider to flag unusual crawl patterns in your server logs. You want to know if and when Muse or similar agents start hitting your pages so you can measure response rates and iterate. If you want to go deeper on building the systems to do this well, the AI For Business community at skool.com/aiforbusiness is where we work through exactly this kind of operational challenge with SMB operators.