On November 5, 2026, an ecommerce development agency in Coral Gables, Florida did something that would have been unintelligible two years ago: it published a guide to agentic commerce optimization, a service for making your store readable to software that shops. Optimum7’s offering formalizes a discipline that has been assembling itself all year: audits of catalog identifiers, pricing logic, inventory accuracy, and checkout flows, all re-engineered so that the next visitor to your product page might not be a person.

The category exists because the traffic is real. It also exists, mostly unexamined, on a foundation where every fact a store tells an agent is a fact the store tells about itself. The optimization wave now rolling through retail makes products legible to machines. Nothing in the stack makes them verifiable.

The stack merchants are now optimizing for

Three surfaces define the current era, and all three ingest the same thing: merchant-authored structured data.

Google’s Universal Cart went live on October 15, 2026. Per Google’s launch post, the cart is rolling out in the US across Google Search and the Gemini app, letting shoppers save products from multiple retailers in one place, track price drops and restocks, and check out “with Google Pay in just a few taps” through the Universal Commerce Protocol, or transfer items to the merchant’s own cart. The launch partners include Nike, Sephora, Target, Ulta Beauty, Walmart, Wayfair, and many Shopify merchants, and in both checkout paths “the merchant stays the seller of record and handles order confirmation and customer support.” Underneath it sits the Shopping Graph: more than 60 billion continually refreshed product listings. Every one of those listings is a self-description supplied by a merchant feed.

ChatGPT shopping runs on a feed spec, not a crawler. As Optimum7’s guide documents from OpenAI’s own commerce documentation, product data for ChatGPT shopping arrives through a structured feed merchants supply, with one row per purchasable item or variant and nine required fields: item ID, title, description, product URL, brand, seller name, image URL, availability, and price. Onboarding is currently open to approved partners. The specification does not reference on-page markup at all. What ChatGPT knows about a product is what the seller typed into a spreadsheet.

Amazon opened its own index to external feeds. Its Shop Direct Merchant Workspace, launched November 4, lets stores that do not sell on Amazon pipe Google-format product feeds into Amazon search results and Alexa for Shopping, no seller account required. The largest product index on earth is now ingesting inventory it does not sell, fulfill, or verify, on the strength of a feed the merchant controls.

Three competing ecosystems, one shared architecture: the agent reads what the merchant wrote.

The numbers that created the category

The urgency is not hypothetical. Adobe’s September 28 holiday forecast, built on more than one trillion visits to US retail sites, 100 million SKUs, and 18 product categories, projects AI traffic to US retail sites up 130 percent year over year this season, peaking at 159 percent growth on Thanksgiving, 95 percent on Black Friday, and 88 percent on Cyber Monday. Adobe measures this as shoppers clicking links surfaced by AI chat and browser services, and the growth comes off a base that exploded first: AI-driven retail traffic grew 693 percent during the 2025 holiday season.

The 2026 actuals show the spike becoming a channel. Per Adobe Digital Insights data compiled by commercetools:

  • Q1 2026 AI-referred retail traffic grew 393 percent year over year; January through May grew 235 percent, and July was up 62 percent.
  • On Prime Day 2026, AI traffic rose 89 percent year over year, and those visitors converted 40 percent better than non-AI channels.
  • Compare the growth asymmetry inside Adobe’s own holiday forecast: paid search revenue share is projected to rise 4 percent, while AI traffic grows 130 percent.

The honest caveat, which Digiday’s reporting keeps in view: AI-driven sessions still sit below 0.2 percent of total ecommerce traffic. The channel is small and compounding fast. Forecasts of where it lands range from Bain’s $300 to $500 billion in US agentic commerce by 2030 to Juniper Research’s $8 billion globally in 2026 growing to $3.5 trillion by 2031. Small base, steep curve, no disagreement about direction.

Structured data is the new storefront

When the reader is a machine, the feed is the page. The Alhena analysis cited in commercetools’ white paper found that 71 percent of pages cited by ChatGPT and 65 percent cited by Google AI Mode include structured data, and SE Ranking’s study found pages with structured data are cited 3.1 times more frequently in Google AI Overviews, which now appear on 14 percent of shopping queries per Search Engine Land.

The supply side is the bottleneck. Rithum’s survey of retailers found 45 percent “sometimes” face data quality issues that affect business decisions and 36 percent said it happens “often.” Duran Inci, CEO of Ecommerce.com and Optimum7, named the failure mode precisely:

“An AI shopping agent can’t reliably quote that price if those calculations are hidden inside checkout. We make the applicable pricing accessible through a feed or API, using the same rules the store applies when a customer buys.”

That is the optimization contract: expose the identifiers, resolve the price outside the cart, return one definition of sellable stock, and surface checkout conditions before payment. Necessary work. Also work with a ceiling, because a store that passes every one of those checks is only guaranteed to be legible.

What agents do with what they read

The conversion data is more interesting than the traffic data. Peer-reviewed research from the University of Hamburg and Frankfurt School (Kaiser and Schulze) found ChatGPT referral traffic converts better than paid social but below direct, organic search, and email: high intent, not magic. Walmart saw conversion rates three times lower for in-chat purchases than for redirects to its own site, then rerouted strategy around the fact that ChatGPT traffic still drives roughly twice as many new customers as traditional search.

The demand side has a specific shape. YouGov finds about 65 percent of Americans trust AI to compare prices, but only 14 percent trust AI to place orders on their behalf. Just 13 percent mostly or completely trust AI for shopping advice, versus 53 percent for personal recommendations. Adobe’s survey of 5,000 consumers, fielded in July 2026, found 77 percent of AI-assisted shoppers feel more confident in their purchases, and 69 percent say they are less likely to return an item bought that way.

Sit with that pair of numbers, because it is the most important dataset in the sector. Confidence is up. Nothing external changed to earn it. The shopper did not gain access to better evidence; the summary got more fluent. An agent that reads 4.7 stars from an unverified corpus will recommend with the same conviction as one reading a verified one, and the confidence transfers to the human, who then returns the item less often. Fluency is being booked as trust.

Every field is a self-portrait

Walk the specifications. Google’s merchant listing requirements mandate a product name, an image, and an offer with a price, and recommend GTIN, MPN, SKU, availability, shipping details, and a return policy. ChatGPT’s feed wants its nine fields. Amazon’s Shop Direct accepts a Google-format feed verbatim.

Across all three: the title is written by the seller. The description is written by the seller. The brand, the availability, the price, the stock count: seller, seller, seller. A Universal Cart price-drop alert is only as honest as the listing it watches. There is no field, in any spec, for whether the review corpus behind the product is authentic, whether the rating has survived manipulation analysis, or whether any third party has examined the product at all. Google’s launch post says, without apparent irony: “In online shopping, trust is everything.”

The adversarial context makes the omission load-bearing. A Visa report cited in security commentary documents a 25 percent spike in malicious bot-initiated transactions over six months, 40 percent in the US. Darwinium’s research finds 97 percent of organizations experienced an increase in AI-facilitated attacks in the past year, at an average direct loss of $4.5 million. Accenture finds 87 percent of payments leaders expect trust to be the most significant barrier to agentic payments, and the industry’s answer, “Know Your Agent” frameworks for verifying who is acting, verifies the actor while leaving the merchandise unexamined. Meanwhile the FTC’s fake review rule is in its second holiday enforcement season, prosecuting an economy of manufactured consensus that predates agents entirely.

The system being assembled for 2026 verifies the shopper, the agent, the payment, and the merchant’s identity. It takes the product’s reputation on the merchant’s word.

The fifth assessment area

Optimum7’s audit checks four areas: catalog data, pricing logic, inventory accuracy, checkout flow. Every retailer racing to become agent-readable should add a fifth: evidence.

Legibility gets you considered. Veracity gets you recommended, because the agent layer is where review manipulation finally meets a counterparty that can be duped at scale. A human shopper scrolling past 3,000 products might skim five reviews. An agent generating a three-product shortlist weights the entire corpus, fake and real alike, with equal confidence, in milliseconds, for millions of users. Unverified review data was always pollution. In an agent channel it becomes the water supply.

An evidence layer for agentic commerce needs three properties the feed specs lack:

  • Independence: scoring computed outside the seller’s control, on review quality rather than review quantity, with manipulated reviews filtered down and authentic ones weighted up.
  • Machine legibility: queryable per SKU at recommendation time, not rendered on a page an agent has to scrape. This is why GoBuy exposes its trust data through an MCP server at gobuy.ai/api/mcp: an agent building a shortlist can pull a Smart Score from 0 to 100, computed from review quality, before it recommends, and consult a curated top-7 rather than re-ranking an unbounded catalog.
  • Persistence: trust that survives a season. Products scoring 80 or above over a 90-day window carry the GoBuy Verified badge, a signal designed to be read by machines and humans alike. For the human holding the phone, the Chrome extension renders the same panel directly on Amazon product pages.

The merchants who will win the agent channel are not just the most legible. They are the ones whose claims survive an independent read.

The uncomfortable summary

Agentic commerce optimization is a legitimate, necessary discipline, and November 2026 is the correct moment for it: Universal Cart is live, ChatGPT feeds are expanding through approved partners, Amazon’s index is open to the open web, and the referral curves all point the same way. But observe what the industry has collectively optimized. It has made products findable, quotable, purchasable. It has not made them knowable. The 77 percent confidence number is the tell: the market is pricing in trust that no layer in the stack actually produces.

Every previous era of commerce built its trust institutions after the fraud waves, not before. The agent era has the unusual option of building one in advance, and the MCP infrastructure to distribute it already exists. Before your next agent-assisted purchase, before your store’s next feed submission: the question that decides everything is not “can the machine read this product?” It is “should the machine believe it?”

GoBuy is the trust layer before buying on Amazon: fake reviews filtered, authentic reviews weighted up, a Smart Score built on review quality, and a curated top 7 instead of a thousand maybes. Try it at gobuy.ai, install the Chrome extension, or if you are building shopping agents, wire in the evidence layer at gobuy.ai/agent-docs.