Claude Goes Agentic. Your PDP Is Not Ready.

Anthropic shipped agentic commerce features for Claude this quarter. Not a roadmap. Not a beta waitlist. Shipped. The new capability lets Claude execute purchases on behalf of users: compare options, select a SKU, and complete a transaction without the shopper ever touching a search bar. That changes the referral model entirely. You are no longer optimizing for a human who clicks. You are optimizing for a model that reads, ranks, and decides.

The Traffic Source You're Not Measuring

AI referral traffic is already moving product. Digital Commerce 360 has tracked the shift across online retailers in 2026. The platforms sending that traffic are not static. Claude joining the agentic tier alongside existing players means the referral cohort just got more competitive. Most operators are still measuring sessions from Google and Meta. They have no SP-API pull, no UTM taxonomy, no cohort tag for AI-origin orders. That is a measurement gap that costs you attribution clarity and, eventually, budget allocation accuracy.

Start there. Before you touch your catalog, tag AI traffic sources as a discrete channel. Build the baseline. You cannot optimize what you cannot see.

What an Agent Actually Needs From Your Listing

A human shopper will forgive a vague bullet point. They will squint at a blurry secondary image. They will decode ambiguous sizing copy. An AI agent will not. Agents pull structured attributes. They compare fields. They disqualify listings that return ambiguous or incomplete data on the query variables the user specified. Your title, your attributes, your spec table, your material callouts — these are not merchandising decisions anymore. They are qualification gates.

Run a field audit on your top 20 ASINs by velocity. Not your full catalog. Your top 20. Check every attribute field the category schema supports. Flag anything blank, generic, or copy-pasted from a manufacturer sheet that predates your current positioning. Agents rank on specificity. 'Premium cotton blend' loses to '94% Supima cotton, 6% elastane, 210 GSM.' Every vague field is an invitation for a competitor to beat you on a structured query.

The NetPPM Case for Acting Now

Catalog remediation has a cost. An agency hourly rate, internal headcount hours, SP-API development time if you are pulling and pushing at scale. Operators who delay will face a different cost: being excluded from the AI-referred order pool during the window when agent-assisted shopping is growing fastest and the incumbent list is shortest. The brands that appear in the first wave of agent recommendations build velocity signals that compound. Those signals affect organic rank. Organic rank affects your cost-per-acquisition on paid. The NetPPM impact is not hypothetical. It is lagged and it is real.

The remediation window is narrow. Agents are live. Buyers are using them now, in smaller numbers, but the cohort is growing. A brand that completes structured catalog work in Q4 2026 will carry cleaner data into 2027 when agent-assisted commerce reaches a scale that shows up in your board deck. The brands still running on legacy flat-file imports will be playing catch-up on a moving target.

Three Questions to Pressure-Test Your Readiness

First: If an agent queried your category right now using the five most common attribute filters, how many of your top ASINs would pass without a missing or generic field response? Second: Does your current UTM and attribution setup let you isolate AI-referred sessions from organic and paid — and if not, what is the specific build required and who owns it? Third: When your catalog data was last audited for structured completeness, was the standard 'does it look fine on the PDP' or 'does every schema-supported field return a specific, accurate, machine-readable value'? Answer those honestly. Then assign an owner and a date. Do not schedule a working group. Assign a single owner and a date.