SEO had one gatekeeper. Agent commerce has five, each armed differently

If you run marketing for a brand that sells physical goods, you spent the last decade learning one discipline: optimize for one search engine, and the gains showed up everywhere a buyer might look. That bargain is gone. Digiday's roundup of recent agent-shopping data, pulling figures from Jellyfish, Tinuiti, Adobe, Dentsu and HUMAN Security, shows what replaces it, and the replacement is worse for your workload: not one gatekeeper to satisfy but several, each reading a different web by different rules, with no single action that moves you across all of them.

Why one playbook no longer works

Start with the mechanics, because they are stranger than they look. Microsoft's Copilot cites retailers more than any other assistant: Tinuiti found the top 100 e-commerce sites made up 34% of its citations in July, triple their share on Google's AI Mode and roughly eight times their share on ChatGPT. The reason is not that Copilot loves stores. It is that Reddit blocks Copilot's crawlers, so forum pages drop out and retailer pages fill the space they leave.

Sit with that. Your discoverability on one major assistant is partly decided by a crawling dispute between two companies you have nothing to do with. There is no content you can write, no schema you can add, that resolves it in your favor. Meanwhile Amazon's Alexa, asked for toy recommendations, returned 177 brands sourced almost entirely from Amazon's own store, while the same prompt gave ChatGPT 24 retailers and Google's AI Mode 37. Three assistants, three incompatible maps of where products are allowed to exist.

Under classic SEO, these would collapse into one ranking you could chase. Here they do not. Dentsu's April survey of 1,950 senior marketers found 59% already tuning content for AI search, tying it for the most-used innovation strategy. But that number hides the trap: those marketers are optimizing for a thing that is really five things, and effort spent winning on Copilot buys you little on Alexa.

The money arrives before the standard does

The incentive to figure this out is real. Adobe, analyzing more than a trillion visits to U.S. retail sites, found AI-referred shoppers spent 53% more per visit than other traffic in July. That is a genuine pull toward these channels.

Read the figure carefully, though. A year earlier, non-AI visits were worth more than twice as much per visit, so the premium has narrowed from the other side, not widened in AI's favor. And a person who asks a chatbot for a specific product is often already close to buying, which means the assistant may be collecting intent rather than creating it. Per-visit value also says nothing about total volume. Lucrative visits that barely happen still barely matter.

Consolidation might rescue the workload, or not

There is one way the five-gatekeeper problem eases: if one agent swallows most of the shopping. HUMAN Security began tracking Meta's Muse on Sept. 22 and logged 5.6 million requests inside two days, a bar it says ChatGPT took 11 months to clear. Muse already averages 72% of the daily agent requests HUMAN sees, 84% of them hitting product and search pages, and in one sample it reached checkout 25% more often than other agents.

If Muse keeps that lead, the fragmentation inverts: instead of spreading effort thin across five assistants, you optimize hardest for whichever one owns the phone. That is a different kind of trap, concentration rather than splintering, and it hands Meta the leverage. The honest caveat is that these numbers come from one bot-detection vendor's opening window, and an agent reaching checkout is not a sale the brand gets to keep. For now, plan for the harder case: several readers of the web, no common feed, and no single lever.