IMPLICATOR
Meta Platforms is testing a shopping research feature inside its Meta AI chatbot that recommends products based on a user's location and inferred gender, Bloomberg reported Monday. Select US desktop users can already see a "Shopping research" button in the Meta AI query box. Type a query, and the chatbot spits out product carousels, each card showing the brand, price, and a link to the merchant's site. Meta confirmed the test. The company said nothing about when it ships broadly or how it plans to make money from it.
ChatGPT, Gemini, and Perplexity all shipped AI shopping features late last year. Meta is arriving after the land rush. But Meta arrives with something the others lack: a user graph covering 3.2 billion daily active people across Facebook, Instagram, and WhatsApp.
The Breakdown
- Meta testing AI shopping for select US desktop users, recommending products based on location and inferred gender
- No checkout or payment in Meta AI; users click through to merchant sites with no confirmed referral commissions
- 41% of consumers used AI for product discovery as of January 2026; a third fully replaced their old methods
- Some Meta AI queries route through Google's Gemini 3 while Meta builds its own Avocado model for H1 2026
What the tool actually does
Ask Meta AI for puffer jackets, and it checks what it knows about you first. Bloomberg's tester got women's options from shops that ship to New York, tailored to the name and location on the account. Each recommendation comes with a bullet-point explanation of why that product was selected.
There is no checkout. No payment processing. No buy button. Users click through to a merchant's website to complete a purchase, which keeps Meta in the product discovery lane rather than the transaction lane. The company would not say whether it earns referral commissions on those clicks or whether brands advertising on Facebook and Instagram get preferential placement in results.
That silence is telling. An emboldened Zuckerberg told investors on a January earnings call that the company's "new agentic shopping tools will allow people to find just the right, very specific set of products from the businesses in our catalog." The phrasing matters. "Businesses in our catalog" means advertisers. Product discovery that routes through an existing ad network is not neutral recommendation, it is paid distribution wearing a different interface.
Late to a crowded counter
Meta is the last of the major AI platforms to add shopping. OpenAI bolted a shopping assistant onto ChatGPT before Black Friday 2025, then started working on PayPal integration for in-chat purchases. Google rolled out its own shopping tools for Gemini around the same time and has since built a Universal Commerce Protocol with Shopify and Walmart to standardize how AI agents interact with retailer inventories.
Amazon's Alexa+ supports conversational shopping through partners like Expedia and Yelp. Microsoft is testing Copilot Checkout. Everyone wants to be the last click before the credit card.
Each platform is trying to become the first screen you open before buying something. PYMNTS Intelligence data from January 2026 shows 41 percent of consumers have used AI platforms for product discovery, and a third of those say they have fully replaced their old methods. Not supplemented. Replaced.
Perplexity pushed harder, building an agent that could buy things on its own. Amazon dragged it to court last November, and that case became the first real legal test of whether AI bots can transact without a human saying yes. Meta, by comparison, is staying well behind that line, offering carousels and outbound links instead of anything resembling a transaction.
Stay ahead of the curve
Strategic AI news from San Francisco. No hype, no "AI will change everything" throat clearing. Just what moved, who won, and why it matters. Daily at 6am PST.
No spam. Unsubscribe anytime.
The data advantage and the model question
Meta's pitch rests on personalization. Google's AI shopping features draw on search history. ChatGPT's draw on conversation context. Meta draws on years of social behavior, likes, follows, location check-ins, and demographic inference built into a platform that knows who your friends are and what they buy.
That is either a competitive moat or a privacy liability, depending on how consumers react to a chatbot that infers their gender from their name and shops accordingly.
None of that matters if the model underneath isn't yours. According to Testing Catalog, some Meta AI queries are currently routed through Google's Gemini 3, not Meta's own infrastructure. The company is building an in-house model codenamed Avocado through Meta Superintelligence Labs, expected in the first half of 2026. Meta also acquired Manus, an autonomous AI agent platform, to prop up its agentic ambitions.
Running your shopping recommendations through a competitor's model while building your own is an awkward arrangement. It works for a limited test with select US users. It does not work at the scale Zuckerberg described to investors.
A storefront without a register
The gap between what Zuckerberg promised and what Meta shipped is worth watching. "Personal superintelligence" was the phrase he used on the January call. "Uniquely personal experience" based on history, interests, content, and relationships. What showed up is a product carousel with price tags and outbound links. Useful, maybe. Superintelligent, not yet.
But the infrastructure is there. Three billion daily users and an ad network that already matches businesses to buyers, all fed by a recommendation engine trained on a decade of social signals. The shopping feature is a prototype today. The question is whether Meta can build the transaction layer before Google's Universal Commerce Protocol or OpenAI's PayPal integration locks in the merchant relationships that matter.
Wells Fargo analyst Ken Gawrelski raised his price target on Meta to $856 in late February, maintaining an Overweight rating. The stock sat at $648 when he published the note. Wall Street is pricing in a version of Meta's commerce future that does not exist yet, one built on closed-loop shopping where product discovery, recommendation, and purchase all happen inside Meta's apps.
Right now, the loop stays open. You ask Meta AI for puffer jackets. It shows you options. You leave to buy them somewhere else. That somewhere else is where the money changes hands, and Meta still won't say whether it gets a cut.
Frequently Asked Questions
How does Meta AI's shopping tool personalize recommendations?
The chatbot uses data Meta already has, including location and gender inferred from the user's name. In Bloomberg's testing, a query for puffer jackets returned women's options from shops shipping to New York based on the tester's account profile. No opt-out mechanism has been disclosed.
Can you buy products directly through Meta AI?
No. The tool is strictly product discovery. Users see carousels with brand names, prices, and links to merchant websites. To complete a purchase, you click through to the retailer's site. Meta has not disclosed plans for in-chat checkout, unlike ChatGPT's PayPal integration or Google's Universal Checkout Platform.
Does Meta earn money when users click merchant links?
Meta won't say. A spokesperson declined to confirm referral commissions or whether Facebook and Instagram advertisers get priority placement. Zuckerberg's January earnings call referenced 'businesses in our catalog,' suggesting the tool may favor existing advertisers.
What AI model powers Meta's shopping feature?
Some queries currently route through Google's Gemini 3. Meta is building its own model codenamed Avocado through Meta Superintelligence Labs, expected in the first half of 2026. The company also acquired Manus, an autonomous AI agent platform.
How does Meta's shopping tool compare to competitors?
Meta arrived last. ChatGPT launched shopping before Black Friday 2025 and is adding PayPal checkout. Google built a Universal Commerce Protocol with Shopify and Walmart. Perplexity built an autonomous buying agent that Amazon sued to stop. Meta's tool is the most conservative: product carousels with outbound links, no transactions.
Harkaram Grewal
Freelance correspondent reporting on the India-U.S.-Europe AI corridor and how AI models, capital, and policy decisions move across borders. Covers enterprise adoption, supply chains, and AI infrastructure deployment. Based in New Delhi.
