How do Meta Muse recommendations choose businesses?

What is Meta Muse, and why should small businesses care?

Meta Muse is Meta's personal AI agent, and it's one of the fastest-growing AI assistants ever launched in the US. On Meta's third-quarter 2026 earnings call, Mark Zuckerberg said more than 10 million people in the US and Canada had set up a Muse agent in about six weeks. He added that roughly a quarter of users come back daily, more than a quarter of weekly users shop through it, and the average purchase is over $200, travel included. Muse does more than chat. It browses retailer sites, compares options and can check out for the shopper once they approve the total. For a business owner, the question changes. It's no longer “Do people use AI to research purchases?” It's “Which agent is doing the shopping, and does it know my business exists?”

The download numbers tell the same story. App-analytics estimates put Muse past 10 million US downloads in 40 days. ChatGPT needed 123 days to hit that mark, Grok 211 and Claude 673. There are fair caveats. Meta used a lot of its own ad space to promote Muse. Muse launched on iPhone and Android at the same time, while some rivals started on iPhone only. And ChatGPT's worldwide debut was still bigger. Even so, early retention reportedly beats the rival assistants' averages, and daily active users passed 2.5 million in mid-October. Some of this is hype, but the shift is real. Millions of buyers now have a new place to ask “what should I buy?”, and the answer comes from an agent, not a page of ten blue links.

This is the part that should get your attention. Zuckerberg said Muse should stay free for most people. Meta would earn a fee on transactions it handles for businesses, and offer paid plans for heavier users. He also admitted Muse is expensive to run, so Meta plans to start charging sooner than it usually does. In plain terms, a pay-to-play layer is coming. Search went the same way. Organic visibility is cheapest and easiest to win early, before an auction sets the price. Businesses that Muse already trusts when fees and paid placements arrive will have an edge that's hard to buy later. That's why it makes sense to treat Muse as a GEO channel now and earn recommendations organically instead of waiting to pay for them.

How does the Muse shopping agent decide what to recommend?

Muse draws its recommendations from two sources at the same time: Meta's own product catalog and a live browse of merchant websites. Meta hasn't published its ranking rules, so anything beyond that comes from outside testing and reports on how the agent is instructed. Those reports line up well, though. When a shopper asks for something, Muse first checks what it remembers about them. That can include size, budget, past preferences and even saved reels from connected apps. It then searches the catalog behind shopping on Instagram and Facebook while a browser agent visits retailer sites in parallel. Finally, it filters and ranks everything it found. If you've read our breakdown of how Meta's shopping assistant builds its shortlist, you'll recognize the pattern: the catalog gets you considered, but the open web gets you chosen.

The website half matters more than most merchants expect. One early test ran 50 prompts in a single retail category. The catalog returned around 60 candidate products per query, but the final answer usually held about six picks, and roughly four of those came from the browser rather than the catalog. The browser agent reportedly visits no more than three merchant sites per request. It stops early once it has a varied enough list. Three sites is not many. If your page loads slowly, hides prices behind scripts or buries the product details, the agent moves on and you never find out you were in the running. The sample was small and exploratory, but the takeaway is clear enough to act on. Your own website is where most Muse picks appear to be won or lost.

Two more reported quirks affect who wins. First, Muse reportedly favors well-known merchants for its top picks, and it prefers brands that sell direct over resellers. “Well-known” isn't a secret score. It's everything the wider web already says about you: reviews, mentions, a verified business profile and consistent listings. Second, Muse acts like an ordinary visitor and, by default, doesn't show its sources in the chat. Your page can be the one that convinced it, and you get no visible credit. That's a real difference from assistants where organic citations keep working after ad spend stops and show up as clickable links. Our advice: don't pay anyone who promises a guaranteed “Muse ranking.” Nobody outside Meta knows the algorithm. What you can control is how easy your business is to read, trust and find.

What makes a website easy for AI agents like Muse to read?

An AI agent can only recommend what it can read quickly and without guessing. The pages that win are easy to crawl, fast, and put the answer first. Muse reportedly reads sites through the browser's accessibility tree, the same structured layer that screen readers rely on. That makes clean HTML, real text labels and proper headings necessary, not optional. Some common setups leave the agent with gaps: a product name inside an image, a price that only appears after a script runs, or a “Request a quote” button with no details nearby. Gaps make an agent hesitant, and a hesitant agent that only visits three sites will pick someone else. Treat every important page as if a fast, impatient, very literal shopper were reading it, because that's what's happening.

Putting the answer first pays off with every assistant, not just Muse. Open each product, service or category page with a plain statement of four things: what it is, who it's for, what it costs or how pricing works, and where you deliver. Then back that up with specs, sizes, materials, turnaround times, guarantees and your return policy. Add FAQs written in the words customers actually use, such as “Does this fit a queen bed?” or “Do you see patients on Saturdays?” They give the agent ready-made answers to match against a shopper's request. Structured data helps too. Product, Offer, Review, LocalBusiness and FAQ markup tell machines exactly what the visible text says. None of this is new SEO advice. It just can't be skipped anymore, now that agents do the reading.

Make sure you aren't blocking the visitors you want. Some sites block automated browsing or AI crawlers outright, often by accident through a security plugin or CDN setting. Muse's browser reportedly behaves like a normal visitor. But other crawlers shape what the wider web says about your brand, including Googlebot, Bingbot, GPTBot, ClaudeBot and PerplexityBot, and they need access too. Next, make sure your product and service details match everywhere they appear. Say your site lists $89, your Facebook catalog says $95 and a marketplace listing says “out of stock.” The agent sees a conflict, and a conflict looks like a risk. For a quick read on how ready your pages are to be quoted by AI assistants, run them through a free GEO score check before you rewrite anything.

How do reviews and a clear business identity affect AI agent visibility?

Reviews and a clear identity are how an AI agent decides you're a real, trustworthy business and not just another URL. SEO people call this your “entity”: your business as machines understand it. That means one name, one category, one location or service area and one set of products. Those details should match across your website, Google Business Profile, Bing Places, social profiles, directories and marketplaces. When they do, an agent can say with confidence who you are. When they don't, it has to guess. Old addresses, different trading names and phone numbers from two moves ago all cause this. Guessing is exactly what agents are built to avoid. Muse reportedly favors well-known merchants for its top picks, so being unmistakable is the first step toward being known.

Reviews do two jobs. They're social proof for people and evidence for machines. When an agent weighs two similar products, it looks for signs that one actually delivers. That means recent reviews, specific comments about fit, quality or service, and replies that show the business is paying attention. Volume matters, but how recent and specific reviews are matters more than most owners think. Twenty fresh reviews that mention what you sell and where you work beat two hundred text-free five-star ratings from 2021. Ask happy customers right when they're happiest, make leaving a review a one-tap job, and reply to every review, including the bad ones. A calm, helpful reply to a complaint often looks more trustworthy than a wall of perfect scores.

Beyond your own pages, the wider web decides how “known” you are. Mentions in local news, industry roundups, supplier pages, comparison articles and community forums all shape the picture AI systems form of your brand. That kind of reputation can't be bought cheaply, but you can earn it steadily with genuinely useful content that other sites want to reference. Good examples are buying guides, honest comparisons, cost breakdowns and answers to the questions customers ask on the phone. That's topical authority, and it builds on itself. The same pages that help you rank on Google and get cited by ChatGPT also feed every new agent that comes along. If you sell physical products, check out which product signals get a store chosen as the destination in ChatGPT too, because there's a lot of overlap with Muse.

What should small businesses do before Muse adds paid placements?

The smartest move is to get your business ready to be recommended now, while Muse visibility is still earned rather than bought. Meta has already said it plans to charge fees on purchases Muse handles, and paid tiers are on the way. Paid placements inside answers would be a natural next step, because other assistants are already going that way. ChatGPT is testing shopping ads, and there are early clues about how sponsored units sit next to organic product results. Muse is currently only available in the US and Canada, so it's urgent if you sell there. If you don't, the same work still pays off on Google and every other assistant. Nothing in the checklist below needs a Muse-specific tool or a new budget line. It's the same organic groundwork that earns rankings and AI citations, done with a closer eye on how an agent reads.

Work through it in order of impact. Start with access and accuracy. Make sure your site is easy to crawl, fast, and matches your catalog and listings, because no amount of content can fix a page the agent can't read. Next, clean up your business identity. Claim and complete your Google Business Profile and Bing Places, fix any mismatched names, addresses and phone numbers, and add Organization or LocalBusiness markup. Then rewrite your most important pages so the answer comes first, and add FAQs in your customers' own words. Finally, build a steady review habit and publish content that earns mentions. Most small teams can get through the first two steps in a few focused afternoons. The last two never really end, and that's where most businesses quietly fall behind.

Paid placements will have their uses, but the economics should sound familiar. Pay-per-click traffic stops the day the budget stops. A well-built page that puts the answer first keeps getting read for years by Google, Bing, ChatGPT, Gemini, Claude, Perplexity and now Muse. If you're weighing the two, our 12-month cost comparison of SEO and Google Ads is a useful reality check. You can also estimate what steady organic traffic would be worth with the SEO traffic calculator. Artiql handles this groundwork on autopilot. It publishes answer-first content that describes your business consistently, in every language you sell in. Pricing is tailored to each business after a short discovery call, so the simplest next step is to book a quick demo and see where your site stands today.

Frequently asked questions