State of the agentic economy, May 2026.

Agent traffic to affiliate destinations is growing roughly five times faster than human traffic to the same destinations. Most of it is happening through shopping and research workflows, not through anything resembling a browser. Networks are starting to notice.

This is the first of what we intend to make a monthly note: a short, honest read on what we are seeing across the agentic economy and the affiliate networks downstream of it. We are writing it partly for ourselves, to force a record we can be held to later, and partly for the operators, merchants, and network teams who keep asking us the same questions. Where we have a number, we will give it. Where we only have a hunch, we will say so and label it as one. The agentic economy is early enough that most confident claims about it are wrong, and we would rather be useful than certain.

Three things stand out this month.

The shape of the catalog query is changing

A year ago, agents searched for products the way a human would: by keyword. Someone wanted a project management tool, so the query was “project management tool,” and the ranking that came back was the ranking a search box would have produced. The agent was a thin wrapper around a human habit, and the catalog behind it could afford to be a thin wrapper around a keyword index.

That is no longer what we see. The dominant query shape now is use case, constraint, and user intent, often all three at once. “A time tracker that integrates with Linear, costs under ten dollars a seat, and does not require a credit card to start a trial.” “Noise cancelling headphones for an open plan office, budget around two hundred, that a reviewer with small ears found comfortable.” These are not keyword searches. They are specifications, and they arrive pre-filtered by a model that has already discarded the options a human would have scrolled past.

The reason is structural. An agent does not have a person’s tolerance for skimming ten near-identical results and eyeballing the differences. It wants the difference encoded so it can reason over it. So the enrichment pipeline that builds our catalog is now optimized for use case tagging over keyword density. We spend more compute describing what a product is for, the constraints it satisfies, and the situations in which it is the wrong choice, than we do indexing the words on its landing page. A product page that ranks well for a human shopper and a product record that an agent can reason over are increasingly different artifacts, and the gap between them is widening every month.

The practical consequence for merchants is that the metadata you expose matters more than the copy you write. An agent will recommend a product whose marketing site is mediocre if the structured facts line up cleanly with the user’s constraints. The reverse is also true: beautiful copy wrapped around vague specifications loses to a plainer competitor that states its limits clearly. The merchants adapting fastest are the ones treating their product facts as an interface, not a brochure, and the ones who never do will quietly fall out of agent consideration sets without ever seeing the impressions they lost.

Reversals are getting cleaner

A year ago, the gap between a conversion and its eventual disposition was opaque. A sale would be recorded, credited, and then, weeks later, quietly reversed, with little explanation and a lag that made reconciliation a guessing game. For a human publisher with a few hundred conversions a month, that lag was an annoyance you absorbed. For an agent operator generating volume across many merchants at once, it is the difference between a usable ledger and noise you cannot plan against.

Networks have tightened their feedback loops this year, partly under regulatory pressure and partly because the volume finally justified the engineering investment. Reversal reasons are more specific than they were. The window between a conversion and its final state is shorter and more predictable. The data that comes back is structured enough to act on rather than just file.

We pass this through to the earnings ledger in real time. When a network reverses a conversion, the corresponding entry moves out of the pending balance on the same cycle, with the reason attached, rather than vanishing silently at payout time. We would rather an operator see a reversal the moment it happens, with its cause, than discover a phantom balance weeks later and have to reconstruct what went wrong. An append-only ledger that tells the truth about pending money is worth more than one that flatters it, even when the truth is unwelcome.

There is a second-order effect worth naming. Cleaner reversals make fraud easier to see. When the legitimate reversal signal is noisy, abuse hides inside it. As the networks sharpen that signal, the patterns that do not fit (velocity spikes, conversions with no plausible click path, operators whose reversal rate diverges sharply from their cohort) stand out earlier and cost less to investigate. That is good for everyone whose traffic is real, and it is part of why we think the long-run incentives here favor operators who run clean.

Operator tiers matter more than we expected

Tier two and above operators, those who have completed identity verification, are converting at noticeably higher rates than unverified operators. We did not predict the size of the gap, and we do not have a clean explanation for it yet.

A few hypotheses, none of which we are committing to. The simplest is selection: operators willing to complete verification are more likely to be running a serious, sustained agent rather than a weekend experiment, so their traffic is better on average for reasons that have nothing to do with verification itself. A second is trust propagation: verified provenance may be earning better treatment somewhere downstream, at the network or the merchant, in ways we cannot yet observe directly from our side of the pipe. A third is plainer still, that verification correlates with everything else an organized operator does well, and we are looking at a composite signal and crediting one input.

We are deliberately not drawing a conclusion. The honest position is that verification predicts conversion quality and we are still establishing why. We are instrumenting more of the path to find out, because if any part of the effect turns out to be causal, it changes how we think about onboarding, about which operators we prioritize, and about what we ask of new agents on day one.

A note on how we measure this

Everything above comes from our own pipeline, which means it is a view of the economy as seen through the agents and merchants already working with us. That is a real sample, but it is not a neutral one. Operators who find their way to a settlement layer this early are not representative of every agent that will exist in two years. We try to flag where a number is likely skewed by who shows up, and we will keep doing that.

We are also wary of confusing growth in our own footprint with growth in the underlying economy. The five-times figure at the top is drawn from destination traffic patterns we can attribute, not from our signup curve. When the two diverge, we will tell you, because a state-of-the-economy note that is secretly a state-of-our-company note is not worth writing.

What we are watching next month

A few open threads we expect to write about as they resolve.

Disclosure verification in the wild. Every link rifref generates carries signed disclosure metadata, and the public verification key will be published as a JWKS document at launch. We want to see who actually fetches and checks it, and whether offline verification gets used the way we designed it to be used.

The use case query trend, quantified. The qualitative shift in query shape is obvious to anyone looking. The next step is to measure it: what fraction of queries are specifications versus keywords, and how that split varies by niche and by the maturity of the agent making them.

Payout method mix. As operators span more jurisdictions, the split across Stripe Connect, PayPal, and crypto is shifting in ways that tell us something about who is actually building and where.

If you operate an agent, join the waitlist. If you sell something on a network, talk to us. The next note will be shorter on throat-clearing and longer on numbers.