GEO is the new SEO. Nobody pays the publisher when it converts.
Claude Opus 5 launched on July 24. Within 48 hours, Anthropic's status page was reporting elevated API errors from the volume of developers integrating the new model. That isn't a bug story. That's a signal: when a frontier model ships, a significant portion of the commercial internet immediately routes through it.
Right now, those agents are recommending products. Some of those recommendations convert. And the publishers whose content shaped those recommendations? They're earning nothing from it.
There's a discipline emerging called Generative Engine Optimization. GEO, if you like the acronym. The premise is simple: AI models are increasingly where product research happens. ChatGPT, Claude, Perplexity. People are asking them what to buy, which vendor to use, which software to choose. Publishers who want to stay relevant are figuring out how to appear in those answers.
The tactics are real. Structured content. Direct, citable answers. Clear attribution and sourcing. Coverage in venues models are known to cite. The goal is the same as traditional SEO: be the source the algorithm trusts. But the algorithm is now a language model, not a PageRank formula.
Publishers are doing this work. The affiliate industry is tracking it. Everflow published a dedicated "Citation Economy" content track this year, calling affiliate managers "the new architects of AI visibility." CJ Affiliate launched an AI Visibility product for merchants who want to know where their programs show up in AI answers. These aren't experiments. They're product launches responding to the same structural shift.
Here's where it falls apart.
When Google drives traffic to your site, the signal arrives clean. A session starts. Analytics fires. If there's an affiliate link in your review, the click happens, the cookie sets, and when the purchase completes, the commission clears. The whole chain is visible.
When an AI model cites your article and that recommendation drives a purchase, none of that happens.
There's no session. No cookie. No click event. The user asked the model, the model synthesized an answer shaped by your content, the user went directly to the merchant and bought, and your contribution to that transaction is invisible to everyone, including you.
This isn't a configuration problem. The infrastructure doesn't exist. There's no mechanism in the current stack for a publisher's influence on an AI recommendation to survive through to a commercial event.
GEO doesn't fix this. It makes it more expensive.
If you do GEO well, you show up in more AI answers. More people get product recommendations shaped by your content. More of those recommendations convert. Your contribution to the merchant's revenue goes up. Your commission from it stays at zero.
The better you are at GEO, the worse the math gets.
This is structurally different from the traditional SEO-to-affiliate funnel. In that model, traffic came to your site first. The visit happened. The affiliate cookie could set. The browser was the session container, and the session container was the attribution surface.
AI-mediated discovery removes the publisher's site from the user's path entirely. The recommendation happens inside the model's response. The user gets an answer and acts on it. The publisher never sees the user, never sets a cookie, never captures the signal.
The Claude Opus 5 adoption surge matters here because it compresses the timeline.
Every frontier model upgrade is a capability step that makes agent-mediated recommendations more reliable and more commercially significant. Better reasoning means better product analysis. Better reasoning means more merchants and users trust the recommendation. The commercial weight of "what the AI recommends" increases with each generation.
That means the gap between publisher influence and publisher compensation doesn't stay constant. It widens with every model launch. Publishers optimizing for AI citations are doing work whose commercial value is rising faster than any mechanism exists to capture it.
The fix isn't a new cookie. It isn't a UTM parameter smuggled into a prompt.
What's missing is an attribution layer that operates at the protocol level: a signed referral credential that travels with the agent's reasoning rather than with the user's browser. When a model synthesizes an answer from a specific publisher's content, the citation becomes a referral event. When that referral drives a purchase, the commission logic has something to attach to.
The technical requirements: the publisher needs a persistent identity the merchant can verify. The referral needs to survive across model inference and into the merchant's conversion event. The payout needs to trigger without a browser in the loop.
None of that is exotic. The pieces exist at the protocol level: signed tokens, API-layer attribution, event-driven settlement. What's missing is infrastructure that assembles them for this specific purpose: connecting AI-cited content to downstream commercial events.
Publishers doing GEO right now are operating on faith that the infrastructure will eventually arrive. The reasonable bet is that it will, because the commercial pressure to build it will eventually outweigh the convenience of ignoring it.
But eventually isn't a payout rail.
The commission layer for AI-mediated commerce needs to ship before publishers have optimized themselves into a structural disadvantage they can't negotiate their way out of. Every month of GEO without attribution infrastructure is a month of measurable commercial influence accumulating with no accounting system attached to it.
The money is moving. The publisher layer is doing real work to make that happen. The infrastructure that connects the two is the open problem.
More at syndicatelinks.co.