Meta Wants to Be Your Personal Shopper

Meta just started testing an AI shopping assistant inside Meta AI, and the timing is… interesting. OpenAI is quietly backing away from direct checkout like someone who realised they volunteered to plan the office party. Meanwhile, Zuckerberg is striding into the AI commerce arena with the confidence of a man who has spent two decades cataloguing your interests, your friendships, and that suspicious amount of time you spend looking at camping gear you'll never buy.

Let's talk about what this actually means for sellers — and why the strategic play here is smarter (and scarier) than it looks on the surface.

The Feature Nobody Asked For (But Meta Built Anyway)

Here's how it works right now: you ask Meta AI something like "what are the best cat toys?" and it serves you a neat little carousel of products with prices, brand names, and links to merchant sites. It even gives you a little explanation for why it picked each one, like a sommelier for pet accessories.

What it doesn't do is let you buy anything directly. You click, you leave Meta's ecosystem, you complete the purchase somewhere else. Which, on the face of it, makes this feel like a glorified Google Shopping result wearing a conversational trench coat.

But that's the part everyone's getting wrong.

Meta doesn't need to process your transaction. Meta needs to own the moment you decide what to buy. The transaction is the boring part — Amazon can handle that. The decision is where the money lives. And Meta has been quietly building the world's most detailed map of human decision-making for twenty years.

The Data Advantage Nobody Wants to Admit

Every AI shopping assistant right now is basically doing the same thing: pulling product data, arranging it in a carousel, and hoping you click. Amazon's Rufus does it. Klarna's OpenAI-powered thing does it. Google's been doing it so long they're practically bored of it.

The difference is what sits behind the carousel.

When Amazon recommends you a product, it knows what you've bought before. When Google recommends one, it knows what you've searched for. When Meta recommends one, it knows what your friends bought, what lifestyle content you engage with, what ads you lingered on for 0.3 seconds longer than usual, and — through some algorithmic alchemy I genuinely don't want to think about too hard — what kind of person you aspire to be based on your Instagram behaviour.

That's not a product recommendation engine. That's a personality profile with a shopping cart attached.

Forrester analyst Sucharita Kodali called this a "copycat move," and she's not wrong about the surface-level product. But dismissing Meta's data advantage is like saying a new restaurant is nothing special because they also serve food on plates. The plates aren't the point. The twenty years of behavioural data simmering in the kitchen — that's the point.

The Trust Problem (And It's a Big One)

Here's where Meta's grand plan hits a speed bump the size of a small building: nobody trusts AI shopping recommendations.

A recent Gartner survey found that nearly two-thirds of consumers believe AI shopping tools will serve them biased results. Which is a polite way of saying most people assume the AI is recommending whatever product paid the most to be recommended. (And honestly? Given Meta's advertising history, can you blame them?)

This is the fundamental tension in everything Meta does with commerce. They've spent years optimising their platform to show you things that advertisers pay to put in front of you, and now they want you to trust that their AI shopping assistant is giving you objective recommendations? That's like your estate agent assuring you there's definitely no damp — while standing in a puddle.

Gartner's Brad Jashinsky made a point worth sitting with: it took sixteen years after the iPhone launched for mobile to overtake desktop as the dominant ecommerce channel. Sixteen years. AI shopping assistants aren't going to compress that trust-building timeline just because the technology is impressive. Humans are stubborn. We took decades to trust online payments, and some of us still give our credit card number to websites with the suspicion of someone handing over state secrets.

Instagram Already Does This (Sort Of)

The obvious question is: doesn't Instagram already handle product discovery? Shoppable posts, influencer content, integrated checkout for some merchants — it's basically been an AI shopping experience wearing sunglasses and pretending to be organic content for years.

And yes, that's exactly the point. The Meta AI shopping feature isn't trying to replace Instagram. It's trying to catch the intent that escapes Instagram — the moments when someone thinks "I need new running shoes" and types it into a search bar instead of scrolling their feed until an influencer tells them what to buy.

It's filling a gap. Right now, if your product discovery instinct is conversational ("what's the best X for Y?"), you're going to ChatGPT or Google or maybe Perplexity. Meta wants that query. They want you asking their AI instead of someone else's.

Whether that works depends entirely on whether Meta can make the experience good enough that people choose it over the alternatives. And right now, with no direct purchase capability, no persistent memory of your preferences, and no integration with Facebook Marketplace inventory, "good enough" is doing some heavy lifting.

More than half of customer experience leaders say they're worried inaccurate AI responses will damage the customer relationship. Meta is clearly aware of this, which is why they're rolling out cautiously to a small group of US users instead of doing the typical Zuckerberg thing of shipping it to three billion people and iterating in public.

(Progress, honestly.)

The Market Everyone's Projecting Into

Grand View Research projects the AI shopping assistant market will grow from $3.36 billion in 2024 to $28.54 billion by 2033. That's a roughly 27% compound annual growth rate, which sounds incredible until you remember that market projections for emerging tech categories are basically astrology with better formatting.

What those numbers do tell us is that serious money is flowing into this space. Meta, Amazon, Google, OpenAI, Perplexity, Shopify, Klarna, eBay — everyone's building some version of an AI shopping layer. The field is crowded enough that calling it a "race" understates it. It's more like one of those mass-participation marathons where half the runners don't know where the finish line is.

The consolidation of this market is going to take years, not months. And in the meantime, we're all going to be bombarded with AI shopping features that range from genuinely useful to "we added AI to our app because our investors asked about our AI strategy during the earnings call."

What This Actually Means If You Sell Things

Let me cut through the noise on this one, because the practical implications are narrower than the headlines suggest.

Right now, today: Meta's AI shopping assistant is not a threat to your existing acquisition strategy. Its reach is limited, it adds friction to the conversion path (you still have to leave the platform to buy), and consumer trust in AI recommendations is low enough that adoption will be gradual.

In the medium term: This is another signal that product discovery is fragmenting across more platforms. The query that leads to a purchase used to live almost exclusively on Google and Amazon. Now it's spreading across AI chatbots, social platforms, and conversational interfaces. Each one of these surfaces pulls product information from different sources and prioritises different signals.

What to do about it: Your product data quality matters more than ever. AI shopping assistants synthesise information from multiple sources — your listings, your reviews, your structured data, your product descriptions. Products described in clear, natural language with rich attribute data will surface better across these systems than products optimised purely for Amazon's A9 algorithm or Google's keyword matching.

Think about it this way: you used to optimise for one search engine. Now you're optimising for a dozen AI systems that each interpret product information slightly differently. The sellers who win will be the ones whose product data is so clean and comprehensive that it works everywhere, not just in one walled garden.

The Bottom Line

Meta entering AI shopping isn't the earthquake — it's another tremor in what's becoming a very long seismic event. The ground is shifting under product discovery, and it's shifting slowly enough that most sellers have time to adapt but fast enough that ignoring it is genuinely risky.

Zuckerberg's play here is characteristically ambitious: use Meta's unmatched behavioural data to build an AI shopping experience that's better at understanding you than any competitor. Whether consumers will trust the company that pioneered targeted advertising to also give them unbiased product recommendations is, let's say, an open question.

But here's the thing that keeps nagging at me: Meta doesn't actually need you to trust the recommendations. They just need the recommendations to be good enough that you click. And with twenty years of data on what makes people click, they might be annoyingly well-positioned to pull that off.

P.S. — If Zuckerberg's AI assistant eventually knows my shopping preferences better than I do, I'm going to need it to also explain to my partner why I need a fourth pair of noise-cancelling headphones. That's the real test of artificial intelligence.

P.P.S. — The fact that Meta's shopping AI can't remember your preferences between sessions is either a privacy feature or an engineering limitation, and I genuinely cannot tell which. Either way, it means the AI will recommend you the same cat toy twice. Which, come to think of it, is exactly how Instagram already works.

Anthropic’s 21 Claude Cowork plugins reveal a brutal truth: a few are game-changing, many are half-baked, and the winners already hint at how AI starts replacing pricey SaaS workflows.

Claude Cowork is less chatbot, more desktop chief of staff, but only if you set up context files, instructions, plugins, and connectors properly so it can actually finish work while you step away.

Claude just gained auto-memory, meaning it can now carry your preferences, project context, and working patterns across sessions instead of starting from zero every time.

Google is giving all AI Max advertisers text guidelines, letting brands steer AI copy with plain-language rules so creative stays on-brand while still chasing stronger search performance.

Google has made Nano Banana 2 its default image generator, bringing faster, sharper visuals, cleaner text, and far better consistency across Gemini, Search, Flow, AI Studio, and the API.

A new Claude Code skill called Nimble turns live websites into clean, structured tables, making real-time web data collection far more useful than the usual wall of scraped text.

OpenAI’s new GPT-5.4 lands in ChatGPT, the API, and Codex with stronger reasoning, better knowledge work, native computer use, and sharper outputs across docs, decks, and spreadsheets.

📸 Kondor for iOS - ChatGPT for your phone camera (link)*

⚠️ Norm AI - Eliminate compliance risks for your business at scale with AI (link)

🎨 Visual Electric - An image generator built for the creative process.

💬 CopilotKit - Build in-app AI chatbots and AI-powered text areas (link)

🎥 Video interviews by Outset - Conduct user surveys with video and AI.

📱 Galileo AI: Generate mobile and desktop UIs quickly with AI.

😍It was great, all I needed

🙂Good, not great

🙄It sucked