Agentic Shopping Isn't Magic Yet: Product Data Is Why

What Wired Actually Found

In November 2025, Wired published a detailed look at the industry's scramble to make AI agents shop for you: "You Won't Be Able to Offload Your Holiday Shopping to AI Agents Anytime Soon." Its conclusion was blunt: the technology is not the blocker. The plumbing is.

Reporting from executives at seven tech and ecommerce companies, Wired found that OpenAI, Google, Amazon and their retail partners were still negotiating over "the amount of product data … that has to be exchanged to make these agents successful." The result, as Wired put it, is that current features "require significant input from users and operate slowly or only for a limited number of items."

"I haven't yet felt a super magical agentic experience in commerce," Talia Goldberg of Bessemer Venture Partners told Wired.

That is the crux. Agents are capable of reasoning about what a shopper needs. What they lack is clean, structured, machine-readable product data to reason over. When the data is missing, the agent slows down, asks the user to do the work, or simply narrows to the handful of products it can actually understand.

The Proof That It's a Data Problem, Not an AI Problem

The same Wired piece contains the counter-example. Expedia's app for ChatGPT feeds the agent real-time flight and hotel pricing: genuine, structured, current data. Users still book manually; no autonomous agent completes the purchase. And yet:

The feature is "leading to greater sales than Expedia anticipated. That means there's something in these tools that works," Expedia's Clayton Nelson told Wired.

Nothing changed about the underlying AI between the products that feel like magic and the ones that do not. What changed was the data. Give an agent structured, live, machine-readable information about what it is recommending, and it performs, and it sells. Withhold that data and the same agent stumbles.

The demand is not in doubt. Wired cites surveys in which 60 percent of US consumers plan to use AI to assist with shopping and 20 percent say they would let an agent fully handle everyday purchases, and points to McKinsey's estimate of up to $1 trillion in agentic-shopping sales in the US alone by 2030.

Since Then, the Channels Have Switched On

The eight months since that article have closed much of the gap on distribution. In July 2026, Shopify announced Agentic Storefronts, giving millions of merchants one-click access to sell inside ChatGPT, Microsoft Copilot, Google's AI Mode and the Gemini app. In the same week, Salesforce made its Agentforce Commerce agents generally available, with native integration into ChatGPT, Google AI Mode and Gemini.

The open standards are live too: the Agentic Commerce Protocol (OpenAI and Stripe) and the Universal Commerce Protocol (Google, Shopify and a coalition of retailers and payment networks). And the early numbers are striking: Salesforce reports that retailers running their own shopping agents grew sales 59 percent faster, and that AI-referred traffic converts at roughly eight times the rate of social.

In other words: the doors into ChatGPT and Gemini are now open to almost everyone. Which surfaces the real question for merchants.

Being Present Is Not the Same as Being Recommended

An AI agent does not return ten blue links. It gives one answer, maybe a short shortlist, and explains why. Turning on an agentic storefront makes your products eligible to appear. It does nothing to make yours the product the agent actually names.

That decision comes down to exactly what Wired identified as the bottleneck: the quality and structure of your product data. When a shopper says "I need a gift for a family who likes light strategy games, under £30, that plays in an hour," the agent has to match that intent to a product. It will pick the one whose data most clearly, and most machine-readably, answers the question.

We wrote about this decision in detail in How AI Shopping Agents Choose Products and Answer Engine Optimization. The short version: agents match on questions, problems and use cases, not on keywords or page rank.

What Agents Actually Need From Your Catalogue

To be the recommendation rather than a missed opportunity, a product needs data structured the way an agent reasons:

• Intent, mapped explicitly. The real questions buyers ask, the problems the product solves, and the scenarios it fits, stated in plain language, per product, not implied through marketing copy.
• Answers, not just prompts. Each question paired with a short, factual answer grounded in the product, so the agent can respond, not just retrieve.
• Live price and stock. An agent that recommends an out-of-stock or mispriced item looks unreliable, so it learns to avoid stale sources.
• Machine-readable discovery. A clean feed and a way for agents to query it directly, so they never have to scrape a human web page and guess.

This is precisely the "product data … that has to be exchanged" that Wired reported the platforms wrestling with. The merchants who publish it in a form agents can consume are the ones who will be exchanged, and recommended.

How Agenticfeed Provides the Missing Piece

Agenticfeed is built to be exactly that data layer. It turns your catalogue into a structured agentic feed:

• Every product mapped to the questions, problems and use cases it answers, each with a short, grounded answer, solution and fit, so agents get reasoning, not just labels.
• Live price and stock, kept current so recommendations stay accurate.
• A machine-readable feed plus an MCP server, so compliant agents can discover and query your products directly.
• UTM attribution on every product URL, so you can measure AI-driven traffic and sales in your own analytics.

It is an open standard, so the work you do once applies across ChatGPT, Gemini, Claude and whatever launches next, the same "syndicate your data into the channels agents consume" approach we cover in How to Make Your Products Visible to ChatGPT and Gemini.

What to Do Now

The window that Wired described, where the experience is not yet magic because the data is not yet there, is the window to move. The platforms have handled distribution. The differentiator left on the table is data quality.

1. Structure your catalogue around buyer intent (questions, problems, use cases), each with a grounded answer.
2. Keep price and stock live so agents trust and reuse your feed.
3. Make it discoverable and queryable: a feed and an MCP endpoint, not a JavaScript-rendered page.
4. Measure it, so you can prove the channel is working.

Do that, and when the "super magical agentic experience" Bessemer is waiting for finally arrives, your products will be the ones the agent reaches for.

Frequently Asked Questions

Why isn't agentic shopping fully working yet?

Because, as Wired reported, the platforms are still resolving how much product data has to be exchanged for agents to work reliably. Today's experiences are slow and limited largely because clean, machine-readable, intent-rich product data isn't widely available. It's a data-plumbing problem more than an AI problem.

Isn't being listed in ChatGPT or Gemini enough?

No. Being present in a channel makes you eligible; it doesn't make you the recommendation. The agent gives one answer and chooses the product whose data most clearly matches the shopper's intent. Presence is table stakes; data quality is the differentiator.

What's the single most important thing to fix first?

Map each product to the questions, problems and use cases it answers, with a short grounded answer for each. That's the layer agents reason over. See What Is an Agentic Feed to see the structure.

The Takeaway

Wired's headline said you won't be offloading your shopping to agents anytime soon. The reason it gave was not that the agents are not smart enough; it was that the product data is not ready. Eight months on, the channels are open and the money is flowing, but the data gap is still the deciding factor in who gets recommended.

That gap is the whole point of Agenticfeed: the structured, intent-rich, answer-bearing, live product data that turns "not magic yet" into a recommendation of your product.