A brand runs an affiliate program. At the end of the quarter the network dashboard shows a clean number: revenue driven, commission paid, a return on spend that looks healthy. The program gets renewed. Nobody asks the harder question, which is what kind of partners produced that revenue.
Affiliate marketing pays every publisher the same way, a tracked link and a share of the sale, and that uniformity hides a difference that matters more than the headline return. A YouTuber whose review introduced a buyer to a product they had never heard of, and a browser extension that popped a cashback offer two seconds before a checkout already underway, both show up in the dashboard as a sale with a commission. They are not the same investment. One grew the business. The other billed for it.
That difference is the publisher mix, and it is the part of an affiliate program most worth understanding. US advertisers will spend an estimated 13.81 billion dollars on affiliate marketing in 2026, according to a September 2025 eMarketer forecast. How much of that is real growth depends entirely on which publishers it goes to.
Where the publisher types came from
When Amazon opened its Associates program in 1996 there was effectively one kind of affiliate: a website with a link. The web was small, search was primitive, and a hobbyist site that linked to a book got a cut if someone bought. Everything that followed is specialization. As the channel grew, publishers found different jobs to do, and the mix sorted itself into four broad families.
Content and review publishers were the original shape, refined. These are the sites that test products and write the comparison a buyer reads before deciding: the "best running shoes" roundup, the long camera review, the buying guide. Their job sits early and middle in the funnel, taking a buyer from "I have a problem" to "I want this specific product." Wirecutter is the famous example, but the family runs from large editorial brands down to single-topic niche sites.
Coupon and deal publishers grew up around a different moment. These are the sites a buyer hits when they pause before paying to type "[brand name] discount code" into a search box. Their job is late funnel: they sit at the checkout and supply, or claim to supply, a reason to finish.
Cashback and loyalty publishers are a close cousin with a twist in the incentive. Rakuten Rewards, the major cashback portals and the browser extensions all share their commission with the shopper as a rebate. A buyer routes through the portal, buys, and gets a slice of money back; the portal keeps the rest. This family is the largest single slice of US affiliate spend, roughly 35 percent in 2024 per Performance Marketing Association data.
Creator and influencer affiliates are the newest family and the fastest growing. A creator tags a product in a video, a story or a newsletter and earns on tracked sales. Structurally it is an affiliate deal. What differs is the source of trust: not a Google ranking, but a person the audience already follows. On Awin's network, content creators' share of affiliate revenue rose from 15.9 percent to 19.5 percent year over year, eMarketer reported.
Who pays all of them is the same: the merchant, through the network, out of one budget. How they are paid is broadly the same too: a percentage of the sale, sometimes a flat fee, on a tracked click inside a cookie window. The sameness of the payment is exactly why the mix is easy to ignore and expensive to ignore.
The question the mix is really asking
The 2024 US publisher breakdown, on Performance Marketing Association data, gives the rough shape: cashback, loyalty and rewards at about 35 percent of spend, content publishers at about 16 percent, coupon and voucher sites at about 10 percent, technology partners at about 7 percent. A separate Awin dataset shows the value-seeking pressure inside that mix. Discount and promotions publishers took 42.4 percent of US affiliate revenue in the first half of 2025, up from 39.7 percent a year earlier, eMarketer reported. Coupons are spreading: 7 of the 10 largest publisher types increased coupon use in that period.
But share of spend is not the right scoreboard. Incrementality is, and it splits the four families along one line.
Incrementality asks: if this publisher disappeared, would the sale still have happened? A publisher is incremental when it produces customers who would not have bought otherwise, and non-incremental when it attaches itself to a purchase that was already coming. Both look identical in a last-click report, which credits whoever delivered the final click regardless of who created the demand.
Content and creator publishers tend to sit on the demand-creation side. A review introduces a product; a creator's recommendation surfaces something the audience was not searching for. These are acts that change what a buyer wants, so when a brand pauses them, demand should genuinely fall.
Coupon and cashback publishers tend to sit on the demand-capture side, and the cashback browser extension is the clearest case. The shopper has already chosen the product and opened the cart. The extension activates at that moment, drops a cookie, and under last-click attribution claims the sale, having inserted itself between a committed buyer and a checkout for a commission. The coupon site that bids on "[brand name] discount code" is a milder version: intercepting a buyer who typed the brand's own name and was already on the way.
This is not a moral claim about coupon and cashback publishers. They do real work. A genuinely good deal can rescue an abandoning cart and bring a lapsed customer back, and cashback portals with real off-site reach, email lists, app placements, bank partnerships, can influence a buyer before checkout. The point is narrower: much of what gets paid in the coupon and cashback families is not creation, it is capture, and a last-click dashboard cannot tell the two apart.
What incrementality testing has actually found
This stopped being theory when brands started running the experiment. Incrementality testing has become the measurement marketers trust most: in one eMarketer-cited survey, 60 percent of US senior decision-makers ranked independent incrementality testing above all other measurement. The method is blunt: switch a publisher type off in one market, keep it on everywhere else, and measure whether revenue moves.
The measurement firm Haus published a test of exactly this kind for a global direct-to-consumer brand. The brand switched off all affiliate loyalty partners in the US for seven weeks while keeping them running in seven other countries as a control. The result: no measurable change in revenue between the US and the control regions. Last-click reporting had been crediting the loyalty channel for revenue that was going to arrive anyway. The brand cut its US investment in that channel and saw no material drop in sales.
That single result should not be over-read into a verdict on every cashback partner everywhere. The honest counter-evidence runs the other way for the channel as a whole. CJ Affiliate's data science team ran what it described as the largest affiliate incrementality study of its kind, analyzing more than 21 million consumers and 5.5 million transactions with a test-and-control design. It found that an affiliate touch in the path to purchase was associated with 88 percent more revenue per shopper, for both new and returning customers. The channel clearly does real work.
Both findings can be true at once, and holding them together is the lesson. The channel is incremental. Specific publisher types inside it, often the lower-funnel ones, frequently are not. An average across the whole program hides that. A publisher-mix review exists to stop the averaging.
A framework for reading your own mix
The useful output of all this is a way to see what a program actually is, not a ranking of publisher types from good to bad. Four steps get you there.
Start by sorting publishers by funnel position, not by revenue. Group them into demand creators, content sites, creators and genuine comparison publishers, and demand capturers, cashback extensions, coupon-code sites and toolbar partners. A program that looks healthy on total return but is 70 percent demand-capture revenue is not a growth channel. It is a discount applied after the fact.
Then look at timing, the cheapest tell available before a full test. Pull the gap between the affiliate click and the sale. Clicks that land seconds before checkout signal capture: nobody researches a purchase in four seconds. Clicks that land days earlier, on a different device, point toward genuine influence. A publisher that is consistently the first touch is doing introduction work. One that is consistently and only the last touch is doing interception work.
Then run the actual test on the families you are unsure about. Switch one publisher type off in one market for four to six weeks, hold comparable markets steady, and watch revenue. If it does not move, you have found spend that can be cut or repriced with no loss. An affiliate program is one of the easiest places to do this cleanly, because the publisher types are already separable.
Then price by contribution, not by last click. Most affiliate platforms now support it. Commission tiers can pay a content or creator publisher a full rate and a checkout-stage coupon partner a reduced one. If a content publisher introduced the buyer and a coupon extension took the final click, the payout can split between the two. First-order commissions can be set higher than repeat-order ones so the money concentrates on genuine acquisition. The mechanics exist. What is usually missing is the decision to use them.
None of this means firing the coupon and cashback partners. It means knowing which is doing real work and which is collecting a toll, and paying each accordingly.
Where the mix is heading
Two forces are about to make the publisher mix matter more, not less.
The first is the collapse of last-click attribution, and the publisher mix is where that collapse is felt first. PayPal acquired the coupon extension Honey for about 4 billion dollars in 2020. In December 2024 the YouTuber MegaLag published an investigation alleging Honey overwrote other affiliates' tracking cookies at checkout and, under last-click attribution, claimed commissions on sales those partners had originated, as documented in the lawsuits and reporting that followed. Honey shed roughly 8 million Chrome users through 2025, Google changed Chrome extension policy in March 2025 to bar extensions from claiming affiliate commissions without delivering a discount, and in January 2026 Rakuten Advertising removed Honey from its network. Strip away the drama and the lesson is structural: a payout rule that rewards the last click will be optimized against, and the publisher type best placed to do that is the one sitting at the checkout. The fix is paying for contribution, not final position.
The second force is AI search, reshaping who can even do the demand-creation job. As buyers ask AI assistants to compare options instead of clicking through review sites, the organic traffic that funded content publishers is falling. But the assistants have to read something, and what they read is heavily affiliate and creator content rather than brand-owned pages. Creator publishers, whose trust comes from an audience rather than a search ranking, are partly insulated. The likely result is a mix that tilts further toward creators and toward content built to be cited by AI, while the pure search-traffic review site gets harder to sustain. The demand-capture families are not exposed the same way, because they never relied on search-driven discovery. They will keep capturing, and the question of whether that capture is worth full price only gets sharper.
For any brand putting AI agents into its commerce stack, the publisher-mix problem becomes a design requirement. If an assistant or a brand's own agent walks a buyer from research to purchase, the system has to record which partner shaped that journey, not just which link delivered the final click, or the program quietly defaults back to paying whoever stood nearest the checkout. Designing agent-mediated journeys so contribution survives the handoff is the kind of measurement plumbing that decides whether an affiliate budget buys growth or a rebate. It is the sort of work Perform Digital does when it helps enterprises put agents into a commerce stack.
The takeaway is small and it is the whole thing. Do not judge an affiliate program by its total return. Open it up, sort the publishers by what they do in the funnel, and ask of each one whether the customer would have arrived anyway. That answer, publisher type by publisher type, is the difference between a channel that grows the business and a line item that gives margin back.
Council summary
This post argues that an affiliate program's headline return hides the question that matters: which publisher types created demand and which merely captured a sale that was already coming. It sorts the channel into four families, content, coupon, cashback and creator, and shows that last-click attribution pays a demand creator and a checkout-stage extension identically even though only one grew the business. Every figure was checked against its source: the 13.81 billion dollar 2026 eMarketer forecast, the Awin creator share moving from 15.9 to 19.5 percent, the Haus loyalty on/off test that found no measurable revenue change, the CJ Affiliate study of 21 million consumers showing 88 percent more revenue per shopper, and the Honey timeline through the January 2026 Rakuten removal all verify. The swapped area and category frontmatter fields were corrected to match the program convention. The reader takeaway is concrete: sort publishers by funnel position, read the click-to-sale timing, test the families you doubt, and price by contribution rather than final click.
Comments