TL;DR: Affiliate income depends on a chain with four links: a search happens, someone clicks through to your page, they click your affiliate link, they buy. AI search attacks the second link, and it does so unevenly. Pew Research Center browsing data from March 2025 found that visits where a Google AI summary appeared produced a click on a traditional result 8% of the time, compared with 15% where no summary appeared. Applying that verified delta to the published Amazon Associates rate card produces a uniform 46.7% commission decline across every product category, from 4.5% books to 1% health products. Because the decline is uniform, switching categories does not rescue the model. The real variable is query shape: Pew found AI summaries appeared on 60% of question-format searches but only 8% of one to two word searches. Review content that targets “best X for Y” questions sits in the most exposed position on the board. The strategic response is to move the portfolio toward query shapes and traffic sources that AI summaries do not intercept, and to stop treating affiliate revenue as passive.
For roughly fifteen years, affiliate publishing ran on a stable assumption: if you could rank a comparison page, Google would send you people, and a predictable fraction of them would buy something. Every piece of advice in the category, every keyword tool, every “start a niche site” guide, rested on that assumption holding.
It is no longer holding, and the interesting part is not that traffic fell. The interesting part is where it fell and where it did not, because that distribution tells you exactly what to do next.
This is a reality check built from two public sources: verified click-through data and a published commission rate card. No projections, no vendor forecasts. Just what the numbers say when you put them next to each other.
The one verified number that changes the model
The Pew Research Center published a browsing-behaviour study on 22 July 2025 based on 900 U.S. adults who consented to install a browsing tracker. It covered 68,879 unique Google searches conducted during March 2025, of which 12,593 produced an AI summary.
The findings that matter for affiliate economics:
- When an AI summary appeared, users clicked a traditional search result link in 8% of visits.
- When no AI summary appeared, they clicked a result in 15% of visits.
- Users clicked a link inside the AI summary in 1% of visits.
- With an AI summary present, 26% of visits ended the browsing session entirely, against 16% without one.
- Around 18% of all Google searches in the month produced an AI summary.
That 15% to 8% movement is the whole story. It is a 46.7% relative decline in the probability that a search turns into a visit. And the session-ending figure tells you the traffic is not being redistributed to some other page you might also own. It is leaving.
The 1% in-summary click rate deserves its own moment. A widespread hope in publishing circles was that citation inside an AI answer would replace the lost organic click. At 1% of visits, citation is a branding outcome, not a traffic channel. Plan accordingly.
Original model 1: what the click collapse does to commission revenue
Here is the first thing worth building. The Amazon Associates operating agreement publishes a fixed commission rate card by product category. Combining those published rates with the Pew click figures produces a per-category revenue picture.
The model holds every downstream assumption constant across both scenarios. Conversion rate and average order value are labelled illustrative modelling parameters, not measured statistics, and they are identical in both columns. That means the difference between the columns is driven purely by the verified Pew click data, while the absolute figures are there to give the ratios a physical shape.
Model basis: 10,000 commercial searches. Pre-AI baseline produces 1,500 visits (15%). AI-summary scenario produces 800 visits (8%). Illustrative visit-to-purchase conversion held at 2% in both scenarios.
| Category | Amazon rate (published) | Illustrative AOV | Commission per order | Pre-AI revenue (1,500 visits) | AI-summary revenue (800 visits) | Decline |
|---|---|---|---|---|---|---|
| Physical Books | 4.5% | $22 | $0.99 | $29.70 | $15.84 | 46.7% |
| Kitchen | 4.5% | $85 | $3.83 | $114.75 | $61.20 | 46.7% |
| Apparel | 4.0% | $55 | $2.20 | $66.00 | $35.20 | 46.7% |
| Headphones | 3.0% | $110 | $3.30 | $99.00 | $52.80 | 46.7% |
| PC Components | 2.5% | $180 | $4.50 | $135.00 | $72.00 | 46.7% |
| Televisions | 2.0% | $600 | $12.00 | $360.00 | $192.00 | 46.7% |
| Health & Personal Care | 1.0% | $40 | $0.40 | $12.00 | $6.40 | 46.7% |
CEOtudent editorial framework. Commission rates are from the published Amazon Associates standard rate card. Click-through rates are from Pew Research Center browsing data. Average order values and conversion rate are illustrative modelling parameters held constant across both scenarios.
Read the final column again. It is the same number seven times.
This is the finding most affiliate commentary misses. The decline is a property of the traffic layer, not the commission layer. A 4.5% category and a 1% category lose the identical proportion of their revenue, because the loss happens before anyone reaches your page. Chasing higher-commission categories, the standard reflex advice, changes the size of the number in the fourth column and does nothing at all to the seventh.
There is a secondary reading worth noting. Because the loss is proportional rather than absolute, high-ticket categories lose far more revenue in dollar terms from the same click decline. The Televisions row gives up $168 while the Books row gives up $13.86. Sites built on expensive products have more absolute revenue at risk even though their percentage exposure is identical.
Original model 2: exposure is not evenly distributed
If category is the wrong lever, what is the right one? Pew reported how often AI summaries appeared by the shape of the query, and this turns out to be where the real variance lives:
- Question-format searches: summary appeared 60% of the time
- Searches of 10 or more words: 53%
- Full-sentence searches: 36%
- One to two word searches: 8%
- All searches: 18%
Combining trigger rates with the two verified click rates produces a blended expected click-through for each query shape. The arithmetic is straightforward: expected CTR equals the share of searches with no summary times 15%, plus the share with a summary times 8%.
| Query shape | AI summary trigger rate | Blended expected CTR | Retention vs 15% baseline | Typical affiliate content that ranks here |
|---|---|---|---|---|
| Question format (“which X is best for Y”) | 60% | 10.80% | 72.0% | Buying guides, comparison posts, “best of” roundups |
| 10 or more words | 53% | 11.29% | 75.3% | Long-tail problem-solving reviews |
| Full sentence | 36% | 12.48% | 83.2% | Conversational how-to and troubleshooting |
| All searches (blended average) | 18% | 13.74% | 91.6% | Mixed portfolio |
| One to two words | 8% | 14.44% | 96.3% | Brand names, specific model numbers, category terms |
CEOtudent editorial framework. Trigger rates and both click-through rates are from Pew Research Center browsing data; blended CTR and retention are derived.
This table is the actionable one, and it says something specific.
The spread between the most exposed query shape and the least exposed is substantial: 72.0% retention against 96.3%. And the most exposed shape is precisely the one that affiliate review content has been optimised for since the category existed. “What is the best laptop for students” is a question-format search. It is the archetypal affiliate keyword. It is also the query shape most likely to be answered before anyone reaches a result.
Meanwhile, the least exposed shape is short brand and model queries. Someone searching a specific model number has already decided what they are researching and is looking for a particular destination. AI summaries intercept that intent far less often.
That is a portfolio insight, not a traffic-recovery tactic. It says the affiliate content that survives best is the content sitting closest to a purchase decision that has already been made, and the content that suffers most is the content that tries to make the decision for someone.
What the CEO does with this and what the student does
The CEO question is not “how do I get my traffic back.” That question assumes the old channel is recoverable. It mostly is not, and a leader who spends two years trying to restore a structurally impaired channel has made a capital allocation error, not an SEO error.
The CEO question is: given a channel that now retains roughly 72% of its value at the exposed end and 96% at the protected end, what is the correct allocation?
Three implications follow.
Affiliate income stops being passive. The passive framing always understated maintenance cost, and this shift removes the last of the pretence. A channel whose delivery mechanism is being actively restructured by the platform that owns it requires continuous attention. If you are modelling affiliate revenue as a set-and-forget asset, the model is wrong before you start. We have looked at what maintenance actually costs across stream types in the passive income myth, and affiliate sits at the higher-maintenance end of that range now, not the lower.
Single-channel affiliate businesses carry concentration risk that is now visible. One intermediary sits between the business and every dollar it earns, and that intermediary changed the terms unilaterally without a negotiation. That is the definition of a concentrated dependency. The response is the ordinary one for concentration risk: build a direct relationship with the audience that does not route through the intermediary, and add revenue models that do not depend on interception-vulnerable traffic. The expertise monetization matrix maps which models depend on borrowed distribution and which do not.
The remaining click is worth more, so treat it that way. If the volume of people arriving is structurally lower, the value of each arrival rises. That argues for depth over breadth: fewer pages, better ones, with genuine testing and comparison that an AI summary cannot compress without losing the substance. A page that exists to restate specifications is exactly what gets summarised. A page containing information that is not available anywhere else cannot be.
The student side of this is simpler and harder. The mechanics of this channel changed inside eighteen months. They will change again. Anyone who learned affiliate marketing as a fixed set of tactics is now holding depreciated knowledge; anyone who learned it as a system with identifiable links in a chain can see precisely which link broke and reason about the next one. That difference is not about effort. It is about how the learning was structured in the first place.
What is still working
Three things are visibly less exposed, and they follow directly from the exposure model.
Traffic that does not originate in a search box. Email lists, communities, video platforms and direct visits do not pass through an AI summary layer. This is the clearest structural implication of the Pew data, and it points the same direction as the minimum viable audience question: owned distribution is worth more when borrowed distribution becomes unreliable.
Content aimed at decisions already made. The 96.3% retention on short model and brand queries is the most protected position on the board. Content that serves someone who knows what they want and needs specific detail is structurally safer than content that tries to help someone choose.
Genuine testing and original comparison. An AI summary works by compressing consensus across sources. It struggles to compress information that appears in only one place. Original measurement is the one form of content that resists summarisation, which is the same logic that makes it worth doing regardless of the channel.
None of this restores 2019 economics. It gets you a channel that works at a realistic scale, with a realistic understanding of what it costs to maintain. That is a worse business than affiliate marketing was and a considerably better one than pretending nothing changed.
One operational note that has nothing to do with algorithms: the U.S. Federal Trade Commission’s Endorsement Guides require that material connections be disclosed clearly and conspicuously, close to the relevant recommendation, every time it appears. This obligation is unchanged by how the traffic arrives, and it applies to the recommendation itself rather than to the traffic source.
Frequently asked questions
Is affiliate marketing dead in 2026?
No, but its unit economics changed materially. The verified data shows a 46.7% relative decline in click-through when an AI summary appears, and that summaries appear on about 18% of searches overall but 60% of question-format searches. That is a serious impairment concentrated in a specific place, not an extinction. A business built on the exposed query shapes has a real problem; one built on short-query, owned-audience or original-testing traffic has a manageable one.
Which product category should I switch to?
Category switching does not address the problem. The revenue model above shows an identical 46.7% decline across every commission tier from 4.5% to 1%, because the loss occurs at the click layer, before the commission rate is ever applied. Higher-commission categories produce more revenue per sale; they do not retain more traffic. Query shape and traffic source are the variables that actually move retention.
Does getting cited inside an AI summary replace the lost traffic?
Not on current data. Pew found users clicked a link inside the summary in 1% of visits, against 8% clicking a traditional result on those same summary-present searches. Citation has brand value and may have downstream effects that click data does not capture, but treating it as a traffic channel is not supported by the numbers.
Why did short one and two word searches hold up so much better?
Pew found AI summaries appeared on only 8% of one to two word searches, against 60% of question-format ones. Short queries are often navigational or highly specific, where a summary adds little. That produces a blended expected click-through of 14.44% against the 15% baseline, roughly 96% retention.
What should I do first if most of my traffic is question-format?
Two things in parallel. Move a meaningful share of effort toward capturing the audience directly, so that future revenue does not depend on the search click at all. Then shift new content toward the protected end of the exposure table, meaning specific-model and post-decision content, and toward original testing that cannot be summarised from other sources. Attempting to restore the previous traffic level on the same query shapes is the option least supported by the data.
Sources
- Pew Research Center, Google users are less likely to click on links when an AI summary appears in the results, 22 July 2025
- Amazon Associates Programme, Operating Agreement, standard commission income rate schedule
- U.S. Federal Trade Commission, Guides Concerning the Use of Endorsements and Testimonials in Advertising
- U.S. Federal Trade Commission, The FTC’s Endorsement Guides: What People Are Asking
- Organisation for Economic Co-operation and Development, research on digital platform intermediation and market concentration
- Peter Drucker, Management: Tasks, Responsibilities, Practices, Harper and Row
This content was compiled with the support of AI following in-depth research, then written and prepared for publication by the CEOtudent editorial team.
