Picture the moment an affiliate commission is born. Someone reads a review, clicks the link inside it, and a cookie stamped with the publisher's ID lands in their browser. Days later they buy, the store reads the cookie, and the publisher gets paid. Every part of that chain depends on one physical event: a human clicking a tracked link.
Now picture the same purchase run by an AI assistant. The shopper asks ChatGPT or Perplexity to find the best option, the assistant compares products, and either the shopper or the agent completes the buy. Nobody clicked a review link. No cookie was dropped. The store sees a sale appear with no referral attached to it.
This is Part 2 of a two-part series. Part 1 covered the agentic commerce and agent-payment protocols, what OpenAI, Google, Visa, and Mastercard are standardizing, and why they matter. This part answers the narrower and more uncomfortable question those protocols raise for one channel: does a tracked, commissionable affiliate link actually survive an agent-mediated purchase, and if it does not, where exactly does the credit go?
What the affiliate link was always carrying
An affiliate link looks like a URL, but its real job is to carry three pieces of information from the moment of influence to the moment of sale. It identifies the publisher. It records a timestamp so a cookie window can expire. And it gives the merchant something to read at checkout. Strip any one of those out and the commission cannot be paid.
For thirty years that worked because the human did the carrying. They clicked, the browser stored the cookie, the browser arrived at checkout, and the cookie was still there. The publisher never had to do anything after the click. The shopper's own browser was the courier.
Server-side tracking and conversion APIs changed where the record lives, moving it from the browser to the merchant's own systems, but they did not change the trigger. Something still has to fire a click event, or pass a click ID, or apply a coupon code, for the merchant to know a partner was involved. The affiliate channel is built on the assumption that discovery ends with a tracked click. An assistant breaks that assumption at the root.
Where the chain actually snaps
When an agent transacts, the break is not vague. It happens at specific, identifiable points.
The first is discovery. The shopper describes what they want to the assistant instead of searching, clicking several review pages, and choosing. The assistant reads sources, including affiliate content, and synthesizes one answer. The publisher's words may have shaped the recommendation, but no link was clicked, so no cookie exists. Influence happened; tracking did not.
The second is the handoff to checkout. Under the Agentic Commerce Protocol that OpenAI and Stripe published and now maintain as an open spec, the agent sends the merchant a structured checkout request: the items, the buyer's payment credentials wrapped in a permissioned token, and shipping details. The merchant stays the merchant of record and can accept or decline. The first published versions of that spec moved a cart and a payment cleanly and nothing else. They carried no field for an affiliate ID or publisher attribution, so the data a tracked link was supposed to deliver had nowhere to sit. That gap was real, and it was the channel's loudest warning sign.
The third is the arrival itself. As one agentic-commerce infrastructure company put it bluntly, agents do not follow affiliate links, do not carry tracking cookies, and reach checkout independently of the referral chain. There is no referral header, no click ID, no cookie drop. The attribution infrastructure does not get bypassed by accident. It simply never fires, because nothing a human used to do happens.
So the honest answer to the title question, under today's plumbing, is no. A conventional tracked affiliate link does not survive an agent-mediated purchase. The click it depends on was the whole mechanism, and the agent removed the click.
What the platforms have actually said and done
The platforms building these assistants have been clearer in their product moves than in their press releases, and the moves matter more.
OpenAI launched Instant Checkout inside ChatGPT on September 29, 2025, letting US users buy without leaving the chat. It opened with US Etsy sellers, with more than a million Shopify merchants flagged as coming soon. It said product results were organic and unsponsored, ranked on relevance. It also said it would take a fee from merchants on completed purchases. Then the retreat. In March 2026 OpenAI pulled back to a discovery-first model. Its own explanation was that the first version of Instant Checkout lacked the flexibility it wanted, so merchants would use their own checkout while OpenAI focused on product discovery. Reporting filled in why: adoption was thin, with only a small number of merchants ever live against a pool of millions of eligible Shopify sellers, and conversion inside the chat ran far below merchants' own sites. Daniel Danker, Walmart's EVP of product and design, said purchases made inside ChatGPT converted at about a third of the rate Walmart's own site achieved. OpenAI's newer shopping research feature builds a buying guide and sends the shopper onward; it does not promise to carry a publisher's tag.
That retreat is the most important signal for affiliates, and it cuts two ways. A discovery-first assistant that hands off to the merchant's own site reopens the door for a tracked link, because there is a destination to send the shopper to. But OpenAI has also said it expects ads and commissions to become a meaningful share of its revenue over time. A platform that monetizes placements and merchant fees directly has little reason to also pay an affiliate publisher for the same sale.
Perplexity went further on transactions, building Buy with Pro and the Comet agentic browser, and it ran into a wall. In March 2026 a federal judge in San Francisco granted Amazon a preliminary injunction blocking Comet's agent from accessing logged-in Amazon pages to shop on a user's behalf, finding Amazon likely to succeed on Computer Fraud and Abuse Act claims. The court's distinction was sharp: the agent may have had the user's permission, but not Amazon's authorization. That ruling tells affiliates something about the terrain. The largest retailer in the world would rather litigate than let a third-party agent roam its logged-in store, which means agent access to merchants will be negotiated and gated, not open.
Perplexity has been more constructive on the publisher side than on checkout. Its Publishers' Program shares revenue with the sources its answers cite, an early admission that content which informs an AI answer has value even when no click occurred. It is revenue sharing for citation, not an affiliate commission, but it accepts the principle that the channel needs.
The new claimants on the commission
If the tracked link does not survive, the commission does not vanish. It gets claimed by whoever now sits closest to the transaction. Three models are competing for it.
The first is the platform fee. OpenAI's small percentage on completed ChatGPT purchases is not an affiliate commission paid to a publisher. It is the assistant charging the merchant for being the point of sale. The platform inserts itself where the affiliate used to be and bills for the position directly.
The second is data-source attribution: paying whoever informed the agent's decision rather than whoever set the last cookie. Perplexity's revenue share is one shape of this. Another is the idea that when an agent cites a piece of content, it emits a signal that travels with the action, so a tracking platform can capture that the content influenced a sale. The unit of payment shifts in this model, from a commission on a sale toward a payment for being a source the agent used.
The third is the survival of the merchant program itself. When OpenAI hands checkout back to the merchant's site, the shopper lands somewhere a normal tracking stack can work, and a publisher whose content carried a code or a deep link can still be credited. This is the most conservative outcome and, after OpenAI's retreat, a real one.
Notice what every model has in common. Credit flows to whoever is provably nearest the transaction or the decision. The publisher who wrote the definitive review two years ago, and influenced the agent through training data or a citation, is the hardest party to pay, because their contribution is the most diffuse. That is the structural risk: not that money disappears, but that it concentrates at the checkout and the platform, away from the content that did the convincing.
Why affiliate publishers are not out of the picture
The case for publishers is not nostalgia. It rests on what the assistants actually read.
EMARKETER reported a striking figure from an October 2025 study: almost 70 percent of the sites ChatGPT cited when discussing the eyewear brand Zenni were affiliate marketing content. Brand-owned sites are a small minority of what AI assistants surface. The reviews, comparisons, and buying guides that affiliate publishers produce are doing a large share of the work inside AI answers. The assistant cannot recommend without reading, and what it reads is disproportionately affiliate content.
That gives publishers real standing, and a clear to-do list. Three moves matter most.
Get cited, not just clicked. The skill shifts from ranking for a keyword to being the source an assistant quotes. That means original testing, first-hand data, structured and clearly-sourced content an agent can parse and trust. Generic AI-spun reviews are exactly what assistants learn to skip.
Build attribution that does not need a cookie. Coupon and promo codes are durable because they are applied in the merchant's own checkout field, so the merchant's systems record them even when the journey ran through a chatbot. Deep links into a merchant program, server-side conversion APIs, and any signal the merchant captures directly all outlive the browser cookie.
Push for an attribution standard, and watch the one that already moved. The protocols are young and still being written, which is precisely when an affiliate-credit field can be argued in rather than bolted on later, and on the Agentic Commerce Protocol that argument has already landed. In December 2025 the spec added an optional affiliate attribution object to its checkout request, described in the documentation as data for crediting third-party publishers. It carries a provider, a publisher ID, campaign and creative IDs, and a touchpoint marker for first-touch or last-touch capture. A publisher ID now has a defined place in an agentic checkout, which it did not a few months earlier. The field is optional, so a merchant has to choose to populate it, and a network has to feed it. But the structural objection, that there is nowhere to put the credit, no longer holds. The fight has moved from whether the field exists to whether merchants and networks use it.
There is a role here for the implementation layer too. The merchants and platforms that will keep affiliates in the value chain are the ones whose agentic checkout deliberately carries partner metadata and supports attribution that does not rely on a cookie. Building that into an agent commerce stack from the start, rather than discovering the gap after launch, is an agentic design decision, and it is the kind of work Perform Digital does with enterprises standing up agent-mediated commerce. The channel does not have to be designed out. It has to be designed in.
What to take from it
A tracked affiliate link, as the channel has used it for thirty years, does not survive an AI agent's checkout. It carried the publisher's identity, the timing, and the merchant signal on the back of one human click, and that click is the exact event the agent removes. Pretending otherwise is how a publisher wakes up to a revenue line that quietly went to zero.
But the commission does not disappear with the click. It moves to whoever sits closest to the buy, and right now that is trending toward the platform and the merchant, not the content. The protocols have started to bend back: the Agentic Commerce Protocol now has a place to put a publisher ID, which it did not at launch. That is the easy part. The hard part is getting merchants and networks to populate that field, and getting credit to flow to a review written two years before the agent ever read it. The affiliate channel's job for the next two years is to win that, in merchant programs and in the attribution plumbing. The assistants are still reading affiliate reviews to answer shoppers. The open question is whether the people who wrote those reviews get anything for it.
Council summary
This post argues that a conventional cookie-based affiliate link does not survive an agent-mediated purchase, because the human click it depends on is the exact event an AI assistant removes. It then tracks where the commission goes instead: toward platform fees, merchant programs, and citation-based revenue share. The council verified the timeline against primary sources and corrected several facts. Instant Checkout launched on September 29, 2025 with Etsy first and Shopify flagged as coming soon, not both at once. The Walmart conversion figure is now attributed to Daniel Danker, the EVP who stated it. The largest fix: the Agentic Commerce Protocol added an optional affiliate attribution object in December 2025, so the earlier draft's claim that the spec has no place for a publisher ID was rewritten to reflect that the field now exists and the real fight is adoption. The reader takeaway is concrete: stop relying on the cookie, get cited rather than just clicked, lean on coupon codes and server-side signals, and push merchants and networks to populate the attribution field the protocol already defines.
Comments