What AI Overviews did to affiliate traffic — and the part that is genuinely good news.
The headline numbers are bad and widely quoted. The second-order finding is not, and it changes what you should do: the traffic that survives converts far better than the traffic that left.
The short version
- Zero-click searches reached about 68% of US queries in early 2026, up from roughly 60% two years earlier.
- AI Overviews appear on over 20% of searches, and where they do, click-through has been reported to fall by close to 60%.
- Informational queries lost the most. Comparison and purchase-intent queries lost far less, because a summary cannot settle a $400 decision.
- The surviving traffic converts better — sometimes dramatically. Fewer visitors, further down the decision, is not the same as less revenue.
- Being cited without being clicked still has value, and it is now something you can deliberately optimise for.
Every affiliate publisher has now seen a chart of their own traffic bending downward and been told it is AI Overviews. Usually that is at least partly true. What is less often explained is which traffic left, why the rest stayed, and what the honest response is beyond “write better content”.
What actually happened, in numbers
On these numbers specifically. Google does not publish click-through data, so every figure here comes from third-party analyses of clickstream and rank-tracking data. Different studies use different samples and get different values. Treat the direction as well established and the precise percentages as one methodology’s answer — and be sceptical of anyone quoting them to two decimal places. Checked August 2026.
Which queries lost, and which did not
This is the part that matters for planning, and averages hide it completely. The loss is not spread evenly — it is concentrated in queries an AI Overview can finish.
| Query type | Exposure | Why |
|---|---|---|
| Simple factual “what psi for espresso” | Severe | Three sentences settle it completely. There is no reason left to click. |
| Definitional and how-does-it-work “what is a pid controller” | Severe | Same. These were the traffic engine of most informational content and they are largely gone. |
| Simple how-to “how to descale a kettle” | High | A numbered list in the box is a complete answer. |
| Comparison “gaggia classic vs breville bambino” | Moderate | A summary can list differences but cannot tell you which to buy, and readers know that. |
| Purchase intent “best espresso machine under $500” | Lower | People spending real money want images, a verdict, and someone who has held the thing. |
| Experiential “is the gaggia classic worth it after a year” | Lowest | The one thing a model genuinely cannot supply: someone who lived with it. |
Read that table as a content brief. The bottom three rows are where affiliate content already made its money. If your site is built on roundups, comparisons and hands-on reviews, your exposure is far lower than the headline figures suggest — and the pages you lost were mostly the ones earning least anyway. If your traffic was built on definitional explainers, the damage is real and the fix is a change of content mix rather than a change of technique.
The finding almost nobody reports
Here is the second-order effect, and it is the reason this article is not simply bad news.
The visitors who still click are further down the decision than the ones who stopped. An AI Overview absorbs the person who wanted a fact and was never going to buy anything. The person who reads the summary, remains unconvinced, and clicks through anyway is doing so because they want to be persuaded — which is the most valuable state a reader can arrive in.
Published case work points the same way for referral traffic from AI assistants: one analysis found that although only around one visitor in fifty arrived via AI referral, that slice drove a disproportionate multiple of sales. A small share of sessions can carry far more revenue than its size suggests.
What this means practically. Judge the damage in revenue, not sessions. Plenty of affiliate sites lost 30% of traffic in 2026 and lost a much smaller share of income, because what left was the informational tail that never converted. If your analytics show a traffic cliff, look at earnings over the same window before concluding anything — the two lines often diverge sharply, and that divergence is the most reassuring chart you will see this year.
What to actually do
1. Shift the content mix toward what cannot be summarised
Not “write better” — write different things. Every hour spent on a definitional explainer is now worth less than an hour spent on a hands-on comparison, a long-term verdict, or an original test.
Practically: fewer “what is” articles, more “which one and why”, more first-hand detail, more photographs. → product photography · roundups that differ
2. Check for an overview before you write
This takes thirty seconds and should be part of choosing every topic. Search the query, see whether a box appears, and ask whether it ends the search. If it does, the ranking may not be worth having. → the ten-minute keyword check
3. Optimise for being cited, not only for being clicked
Appearing inside an overview is a brand impression from a source the reader trusts, and some of that converts later even without a click. Nobody can guarantee inclusion, but the patterns are unsurprising:
- Answer the question directly and early, in a self-contained paragraph a model can lift.
- Use plain declarative sentences rather than building to a conclusion over four paragraphs.
- Structure with clear headings that match how the question is asked.
- Include specifics — numbers, dates, named sources. Extractable facts are what get quoted.
- Be the primary source where you can. Your own test result cannot be summarised from someone else’s page.
4. Reduce your dependence on search entirely
The uncomfortable one. Tighter rules from Amazon and a search-behaviour shift from Google, in a single year, is a clear signal about how much control you have over your own traffic.
An email list is the only audience nobody can algorithm away, and it has never been more obviously worth building. → building an email list on an affiliate site
Measure your own exposure, not the industry’s
Sitewide averages tell you nothing about your site. Twenty minutes in Search Console tells you a great deal.
- Compare impressions against clicks over the last 16 months. Flat or rising impressions with falling clicks is the AI Overview signature — you still rank, fewer people click. Both falling together is a ranking problem instead, and has a different fix.
- Segment by query type. Filter queries containing “what is”, “how to” and similar, then compare against those containing “best”, “vs” and “review”. The gap between those two trends is your real exposure.
- Check your top twenty queries by hand. Search each one and note whether an overview appears and whether it finishes the job. Twenty minutes, and far more useful than any published average.
- Put revenue on the same chart. If earnings held while sessions fell, you lost the tail rather than the business.
What not to panic about
Three conclusions people are drawing that the data does not support.
“SEO is dead”
360 clicks per 1,000 searches is a smaller number than before and an enormous amount of traffic. Ranking still matters — being cited in an overview generally requires ranking well in the first place.
“Affiliate sites are finished”
What was damaged was thin informational content. Sites built on genuine testing and real recommendations came through comparatively well, and have less competition than before.
“I should chase AI referral traffic”
It converts well and it is small. Worth being visible for, not worth restructuring a business around. Build for readers; citation follows.
“I need to block the crawlers”
Blocking removes you from citations without restoring the clicks, since the queries were answerable regardless. A defensible choice for some publishers, but rarely for an affiliate site that needs discovery.
Part of The Amazon Affiliate Site Playbook — the whole job in six stages, from choosing a niche to keeping a mature site earning.
See which pages still earn, not just which still get traffic
Ama Affiliate Pro tracks clicks and revenue per page first-party, so you can put sessions and earnings on the same chart — and find out whether you lost the business or only the tail that never converted.
Start a 14-day trialCommon questions
How much traffic do AI Overviews take?
Published 2026 analyses put zero-click searches at around 68% of US queries, with roughly 360 clicks to the open web per 1,000 searches. Where an AI Overview appears, click-through to a traditional result has been reported at about 8% versus 15% without one. Google publishes none of this, so treat the direction as solid and exact figures as methodology-dependent.
Which content lost the most traffic?
Simple factual, definitional and basic how-to queries — anything a three-sentence summary finishes. Comparison, purchase-intent and experiential content lost far less, because a summary cannot settle a $400 decision or tell you what a product is like after a year.
Are affiliate sites still viable?
Yes, if they are built on things a summary cannot replace: genuine testing, real photographs, a verdict. What was damaged was thin informational content assembled from other people’s pages. The bar rose, and a lot of low-effort competition left with it.
My traffic fell but my revenue did not. Why?
Because what you lost was the informational tail that never converted. AI Overviews absorb the reader who wanted a fact and was not going to buy. The ones who still click are further down the decision, so revenue per visitor rises even as visitors fall. Put both lines on the same chart before concluding anything.
How do I get cited in an AI Overview?
Nobody can guarantee it, but the patterns are consistent: answer the question directly in a self-contained early paragraph, use plain declarative sentences, structure with headings that match how the question is asked, include specifics like numbers and dates, and be the primary source where possible. Ranking well is generally a prerequisite.
Should I block AI crawlers?
Rarely, for an affiliate site. Blocking removes you from citations without restoring the clicks, because those queries were answerable from other sources anyway. It is a defensible choice for large publishers with licensing leverage; it mostly costs discovery for everyone else.
How do I tell whether AI Overviews are hurting my site?
In Search Console, compare impressions against clicks over 16 months. Flat or rising impressions with falling clicks is the signature — you still rank, fewer people click. Both falling together is a ranking problem with a different fix. Then check your top twenty queries by hand.
Is it worth writing informational content at all now?
Yes, but choose differently. Skip the definitional pieces a summary finishes. Write informational content that requires experience, judgement or original data — and check for an overview before committing an afternoon to any topic.