Sixty-eight per cent of Google searches now end without a click, and “best X for Y” is exactly the question an AI answer handles in full. The traffic model that most affiliate income was built on has been taken apart from underneath over about two years, and plenty of people have not worked that out yet because their own decline felt like a personal failure.
The mechanics are unchanged: you recommend something, someone buys through your link, you take a percentage. No product of your own, no inventory, no clients. It remains the most widely used creator income stream and the most widely abused one. What changed is where the buyers come from.
Our position should be on the table before anything else. Top Creators takes no affiliate commission and links to no referral programmes, because affiliate income is the main reason creator advice about tools is unreliable. That does not make it a bad way to earn money. It means the thing it does to your judgement is the actual risk, and nobody writing an affiliate guide has an incentive to tell you that.
What it pays#
There is no useful benchmark and anyone quoting one is guessing. What can be described is the shape.
Commission rates run from about 1% on general retail up to 30% or more on software. Physical goods pay least, because the margin is thin and the retailer keeps it. Digital products and software pay most, and some pay recurring commission for as long as the customer stays, which is a different kind of income entirely: one good recommendation that produces twenty subscribers can still be paying you in three years.
The variable that decides your income is not the rate. It is whether the person arrives ready to buy. A thousand people reading a detailed review of a specific product they were already considering will out-earn a hundred thousand people watching something entertaining, several times over.
So affiliate income has always tracked search traffic and comparison content, not follower count. And why, for a lot of people, it fell off a cliff this year.
The thing that broke#
In the first four months of 2026, about 68% of Google searches ended without a click — the fastest shift in a decade. Referral traffic from Google to publishers fell roughly 38% year on year. Where an AI Overview appears, click-through on the top-ranked result dropped by around 61%. Small publishers lost about 60% of their Google traffic over two years.
“Best X for Y” comparison content is the single most affected category, because that is exactly the question an AI answer handles start to finish. That format exists to answer a question, and an AI answer now handles it completely: the reader gets the comparison, the recommendation and the reasoning without visiting anyone’s page. No page view, no click, no commission.
If you have read an affiliate guide from before 2025 telling you to build a niche site full of product round-ups, that described a real business model that has since been dismantled from underneath.
Sources: SparkToro, Search Engine Land, Search Engine Journal.
What still works#
Three things survived and they have something in common: they cannot be answered in a summary.
Depth on one thing, from use. Not “the ten best microphones” but “I used this microphone for eleven months in an untreated room and here is what went wrong”. An AI answer can list the ten best microphones. It cannot report what happened to you in month seven, and that is the content that converts anyway, because the reader is looking for the thing that stops them regretting the purchase.
Email. A recommendation in a newsletter reaches people without passing through a search engine or a recommendation feed. The traffic is smaller and it converts far better, and it is the reason an owned audience quietly became the whole game this year.
Answering the question after the comparison. The AI answer tells them which three products to consider. It does not tell them which one suits a left-handed person, or which one the manufacturer has quietly stopped supporting. The work that survives is the work a summary cannot finish for you.
What does not work any more: publishing volume, targeting long-tail keywords, and rewriting the same round-up with a different number in the title.
The mechanics people get wrong#
Cookie windows vary enormously and nobody reads them. Some retail programmes attribute a sale only if the person buys within 24 hours of clicking. Software programmes often run 30, 60 or 90 days. That single number changes what your content is worth: a 24-hour window means you are only paid when someone is ready to buy today, which makes considered-purchase content nearly worthless in that programme, however good it is.
Recurring beats one-off, usually by a lot. A 20% one-off commission on a $60 annual subscription is $12. A 20% recurring commission on the same product, for a customer who stays three years, is $36 and it arrives without further work. Where both models exist, the recurring one is almost always the better deal despite looking smaller on the sign-up page.
Programmes cut rates without warning. This has happened repeatedly across the industry and the direction of travel has only ever been one way. Any income stream where the counterparty can unilaterally halve your rate is not a foundation.
Self-referral and coupon stuffing get accounts closed. So does buying search ads on the brand’s own name, in most programmes. Read the terms once, properly.
Disclosure#
Affiliate links are advertising and they have to be identifiable as such. In the UK that is an Advertising Standards Authority requirement, in the US the Federal Trade Commission, and most jurisdictions have an equivalent. Platforms require it as policy on top of the law.
Do it in the content, near the recommendation, in plain words, not in a footer, not in a page-bottom disclaimer nobody scrolls to. “If you buy through this link I get a commission” costs you nothing with an audience that already assumes it, and it is the difference between a legal problem and no problem.
Check your own regulator’s current guidance rather than relying on a summary, including this one.
The slow drift in what you recommend#
Not the disclosure and not the traffic. The slow drift.
It does not usually happen as a decision. Nobody sits down and chooses the worse product for the better rate. What happens is that over eighteen months you make slightly more content about the categories that pay well and slightly less about what your audience actually needs; you mention the tool with the recurring commission a little more often; you notice the thing with no affiliate programme did not make the list and you do not examine why.
The audience notices before you do. What you lose in the process is the thing that made your recommendations worth reading, which was also the thing making the commissions possible.
The only workable defence is a rule you set in advance and do not negotiate with: you recommend what you would recommend unpaid, in the order you would recommend it unpaid, and the commission rate never determines what goes first. That is not a moral position. It is what keeps this earning anything past year one.
AI shopping agents: the interesting half#
As assistants start completing purchases rather than just answering questions, affiliate turns out to be the commission model that fits them best. An agent recommending a product earns by picking something suitable rather than by taking the highest bid, which aligns better than advertising does.
The infrastructure is not ready. Consumer appetite is real: around 39% adoption, and agent-driven traffic up 805%. But conversion through agents runs about 86% worse than through conventional affiliate channels, because merchant systems were not built for it. OpenAI’s Instant Checkout retreated from native checkout in early 2026 after only around thirty merchants went live.
Meanwhile Google is rebuilding search, YouTube and shopping around agent-driven buying, and Meta moved product tags directly into Reels while declaring that “the era of link in bio is over”.
What that adds up to: the old route through search traffic is shrinking, and a new one through platform-native commerce and agent recommendation is being built. Neither is a reason to make this your main income this year, and the second is worth watching rather than betting on.
Sources: MetaRouter, eMarketer, Search Engine Land, Tubefilter.
Who this suits#
People whose content is already about choosing between things, tools, gear, software, equipment, where a recommendation is what the audience came for.
People in subjects with expensive purchases, where a handful of conversions is worth having and a 5% commission is a real number.
People with an email list, which is now considerably more reliable than search traffic for this.
People who can hold a line on what they will recommend, and who have other income so that holding it is not expensive.
Who it does not#
Anyone planning to build a comparison site and live off search traffic. That model has been substantially undermined and the trend is still moving in the same direction.
Anyone who will find it hard to refuse a higher commission on the worse product. This has already happened at scale. It is why the whole genre of tool recommendations now gets read with suspicion, and why we stay out of it.
Anyone whose audience is not in a buying frame of mind, and anyone who would rather not disclose.
If you do it#
- Only recommend things you use and would recommend unpaid. Write the rule down; you will need it when a good rate arrives on a mediocre product.
- Disclose clearly, in the content, every time.
- Build on email or a community rather than search traffic. That is the structural change of 2026 and it is not reversing.
- Prefer recurring commissions on software you rely on over one-off percentages on retail.
- Check the cookie window before you write anything. A 24-hour window makes considered-purchase content nearly worthless in that programme.
- Write depth from use rather than round-ups. It is the only format an AI answer cannot replace.
- Never let the rate decide the order.
- Treat it as one stream among several, and do not plan on it being stable.
If affiliate income was the plan and the traffic has gone, getting found covers what replaced search. Digital products is where most former affiliate publishers have moved, because it uses the same audience and the same expertise without the conflict of interest sitting in the middle of it.