What personal shopping agents need from your catalog

For the last 2 years, AI has proved to be excellent at answering questions and recommending products. Consumers are delighted, and adoption has climbed quickly. In a May 2026 survey of 4,040 US and UK adults, 75 percent had used an AI tool such as ChatGPT, Gemini, or Claude for shopping or a purchase decision, and for the first time more of them said they would rather shop through AI than directly on a brand's site.

In August and September, a new category emerged: personal agents.

• xAI launched Grok Bot on August 11.
• A startup called Instinct passed 100,000 users and raised $1 billion at a $10 billion valuation.
• Meta released Muse on September 8, and it reached No. 1 in the App Store within days.
• OpenAI released Dots, always-on agents that keep working toward a goal without being prompted.

These agents connect to your email, calendar, messages, and payment accounts to solve problems, run errands, and buy whatever you need. If you ask Instinct to help you buy a travel stroller, it finds top options, reviews prices and features, and buys it once you approve.

AI is moving at a furious clip, and it is not yet clear what merchants need to do to adapt.

The new personal shoppers, and how they will get paid

Why do their users like having personal agents? The agents are tireless. They can check as many retailers as it takes to find the item, read return policies, and come back with a recommendation (or two). They are able to think broadly, planning both the trip and making the required purchases for it. They are accessible on the channels we use most: phone, WhatsApp, voice.

How do they compete and compare to one another? At this stage, personal agents offer similar value propositions but with variation across initial business models. Meta's Muse has paid tiers at $20 and $100 a month. Instinct is free during its beta and has not disclosed how it will make money; a cut of what people buy through it is the obvious option, which raises the question of agent incentives. All of this is early enough that it is subject to change, and whether people settle on one agent or several is still to be determined.

Beyond the pricing, the agents will compete on how much you trust them to do the job. That trust covers more than keeping your credentials safe. An agent also has to recommend the product that fits rather than the one that pays it the most, and it cannot steer you toward a pricier version of what you asked for. Losing trust will lose the assistant relationship, and switching costs are zero. For a brand, this has a practical consequence: an agent protecting its credibility will favor attributes it can verify, such as fit, compatibility, and return terms, over marketing copy.

How the agents connect to checkout, and where Amazon stands

Personal agents are all being integrated with the commerce infrastructure of the internet. Stripe's Link, a checkout wallet comparable to Shop Pay or PayPal, reaches more than 300 million consumers and now powers purchases for Instinct, Muse, and Grok Bot. (For businesses that don't accept Link, it issues a one-time virtual card.) Stripe says agentic purchases using Link grew 38X over the previous month.

Additionally, Meta added Shopify, PayPal, Instacart, and Expedia as partners, enabling the agent to buy consumer goods, fill a grocery cart, and book travel. OpenAI followed suit: its Agentic Commerce Protocol, built with Stripe, routes purchases through retailer apps connected to ChatGPT.

A hold out, Amazon has not let the agents in, citing security and transparency concerns. Beyond those, there is an obvious commercial rationale: nothing stops a vendor-agnostic agent like Muse from bypassing the Amazon marketplace and buying directly from brands. For now, Amazon continues to enrich its shopping experience with agents like Rufus, and continues to differentiate with the greatest supply chain in the history of the world. For everyone else, the consequence is simple: the agent reads your own catalog directly, so your product data is now the storefront.

What this means for retailers and brands

As personal agents become more ubiquitous, they will categorically own more of the shopping journey for a greater variety of products. This will change merchandising, brand development, and distribution strategies.

After all, AI engines, and likely AI agents, don't browse storefronts like a human. They interrogate product catalogs and content in search of specific attributes before recommending one (or two) to a customer. Brands must prioritize two things: did the AI find you? And once it did, could it accurately represent your products and brand to a shopper who may never visit your site.

So what does an agent need from a product record? In our work with brands, it comes down to a short list:

• What the product is: the category, the specifications, the materials, the dimensions.
• What it is for: the use cases and the problems it solves.
• Who it is for: the body, the household, the skill level, or the situation it suits.
• When to recommend it: the questions a shopper asks before buying, answered in plain language.
• The constraints that decide a purchase: fit and compatibility, availability, price, delivery times, and return terms.

Beginning with a taxonomy

For most of the last two decades, the structure for your data was defined by two tech companies: Google and Amazon. Google's product taxonomy covers more than 5,000 categories, and Amazon's product types define the attributes required to list on its marketplace. Retail catalogs and PIMs were set up by starting from one of these two trees.

Both taxonomies do the first item on the list well. They ask brands to describe what a product is, and they say much less about what it is for, who it is for, or when to recommend it. Google began addressing this in January, adding dozens of new Merchant Center attributes for conversational commerce, such as product questions, compatible accessories, and substitutes. That is the right direction, and it serves Google's own surfaces. A brand now has to be readable by every engine and every agent at once.

Where Fabric fits

Fabric is built to help consumer brands grow in an era of AI. We modernize a retailer's commerce data and systems for AI, whether the reader is ChatGPT or Gemini, a personal agent like Muse or Instinct, or the storefront agent a brand runs on its own site.

The first step is the catalog. Every merchandising team we have worked with has had a backlog of products carrying only a title, a price, and a few generic tags. This gap was unpleasant when a human shopper could see a photo and assume the best. When an AI agent is shopping, a data gap is business stopping. So we start together by identifying the missing attributes and rebuilding each product and category in your catalog against the Fabric Shopping 360.

The Fabric Shopping 360

At Fabric, we built the Fabric Shopping 360, an index that enriches these original taxonomies with product attributes generated from the millions of SKUs Fabric has processed across shopping verticals. Where the Google taxonomy includes shoe size, color, and gender, the Fabric Shopping 360 adds the use case, the foot type it suits, the level of support and cushioning, whether it works with orthotics, and questions people ask before they buy. Most importantly, the Fabric Shopping 360 is continuously updated from what we observe in shopper questions, AI engine answers, and market and language trends, and it publishes to every engine and agent rather than to one company's surfaces.

Where to start

Personal agents are still new and continuously evolving. But updating a catalog is required in a time when consumers rely more and more on AI, because every engine and agent needs data to decide what to recommend. To learn more about the Fabric Shopping 360, visit fabric.inc.

If you want to join our mission to support brands as they grow with AI, visit our careers page.