Affiliate marketing in the AI era

The argument about whether AI killed affiliate marketing is not useful, because it collapses two different things. One layer of the web did lose its audience, permanently. Another layer became more valuable, because far fewer people can produce it. Almost all affiliate content sat in the first layer, which is why the panic feels justified from the inside.

This guide sets out what changed, what a page has to carry now to be worth publishing, and what an affiliate operation has to change once it accepts that. No doom, no hype, and no claim that we have solved it.

What actually changed: the definitional layer moved into the answer

Search results now routinely answer the question on the results page. Google's own documentation on AI features in Search describes AI Overviews and AI Mode using a "query fan-out" technique — issuing multiple related searches and assembling an answer from them — and states that to appear as a supporting link, a page must be indexed. It also states plainly that there are no additional requirements to appear there: no special file, no rewriting for models, no separate optimisation track. The best practices are the ordinary ones.

What that does to the traffic distribution is straightforward. Questions with a settled, factual answer get answered in place. Questions that require judgement, comparison against a reader's own situation, or information that exists nowhere else still send the click. The definitional layer — "what is NGR", "how does revenue share work" — is exactly the layer most affiliate sites built their content plan on.

Two things follow, and they point in opposite directions. Zero-click behaviour reduces the volume of easy traffic. But Google also reports that clicks from pages with AI Overviews tend to be higher quality, in the sense that users are more likely to stay on the page they choose. Fewer visits, better visits. For an affiliate paid on depositors rather than pageviews, that is not automatically a bad trade — it is only bad if your pages were built for the volume.

The page types that lost their audience, ranked by how much they lost

Why the 800-word "what is X" explainer is finished as a traffic asset

Not because it is bad, and not because it is short. Because the reader's need is fully met before the click, and no amount of formatting changes that. The right response is not to make the explainer longer — length is not the variable — but to demote it. On this site, definitions live in a single glossary where they support other pages instead of competing for traffic on their own. One glossary entry that a dozen guides link to is a structural asset. Twelve separate definition pages are twelve pages that will never be clicked.

What still earns a click: four properties, with examples

1. A number you produced. Not a statistic quoted from a study everyone quotes — a measurement from your own operation, with the date you took it. "We measured X on this date and here is the method" cannot be synthesised from other pages.
2. A document you read. Affiliate contracts are not public and not indexed. A page that says what clause 7.3 of a specific program actually does to your earnings has an input the model does not have. Every entry in our affiliate program directory is written from the terms for this reason.
3. A calculation the reader can run on their own inputs. A summary can tell you what negative carryover is. It cannot tell you what it does to your next six months at your traffic level, which is what our commission calculator exists to answer.
4. A judgement with the reasoning shown. "We would not take this deal, and here is the arithmetic that decided it." Models hedge by construction; a stated position with its working attached is a different product.

First-hand experience is the only input a model cannot borrow

Google's helpful content guidance asks whether content "clearly demonstrate[s] first-hand expertise and a depth of knowledge". For most affiliate pages the honest answer is no — the writer has never opened an account, never read the contract, never been paid by the program they are describing. That is not a moral point. It is a supply point: the input is scarce, so the pages that carry it are scarce.

The practical version for a beginner with no portfolio: your first-hand asset is your own process. Sign up to two programs, document what the application asked for, what was refused and what was approved, how long the reply took, what the tracking dashboard actually shows. That is a page nobody else can write, on day one, without any traffic at all. It also happens to be the page a program manager reads before approving your next application — which is why we put our own operating history on our about page rather than hiding it behind a persona.

Helpful content is a system, not an update: what it measures

The "helpful content update" language has been retired; the assessment is now part of Google's core ranking systems and runs continuously. This matters operationally in two ways. First, there is no event to recover from and no date to point at — a site is assessed on what it currently is. Second, the assessment is site-wide in effect: a large volume of unhelpful pages can weigh on the pages that are good.

The self-assessment questions in the documentation are worth answering in writing, per page, before publishing. The three that catch the most affiliate content:

• Would someone reading this leave feeling they had a satisfying experience, or feeling they need to search again?
• Does the page demonstrate first-hand expertise, or does it summarise what other pages say?
• Was this made for people, or primarily to rank? If you would not publish it with no search engines in the world, that is your answer.

Scaled content abuse, and where the line actually sits

Google's spam policies define scaled content abuse as "when many pages are generated for the primary purpose of manipulating search rankings", including "creating large amounts of unoriginal content that provides little to no value to users, no matter how it's created". The last clause is the one people misread. The policy is not about the tool. It is about volume without value.

Two neighbouring policies matter just as much to affiliates. Thin affiliation covers pages with affiliate links "where the product descriptions and reviews are copied directly from the original merchant without any original content or added value" — the default output of a template fed with operator copy. Doorway abuse covers near-identical pages built to catch query variants, which is precisely what a program-by-country page matrix is.

Using AI to draft versus generating pages at scale — the distinction that matters

Workable floors: at least 1,000 words for a post and 1,200 for a guide, at least three substantive H2 sections, and at least 40% net-new substance measured against any source we drew on. Those numbers are not magic; they are the level below which a page tends to say only what a reader could get elsewhere.

Writing for a reader who has already read the AI summary

Assume the reader arrives knowing the definition. That single assumption rewrites the article structure:

• Delete the throat-clearing. No "in today's competitive landscape" opening, no restating the question. Start at the first thing the summary could not tell them.
• Lead with the specific. A number, a clause, a decision. If your first paragraph could open any article on the topic, it is not an opening.
• Make the thing that cannot be summarised the centre of the page, not an aside at the bottom — the table of real terms, the calculator, the account of what happened.
• Say what you do not know. An explicit "we have not tested this" is credibility, and its absence is the most common tell of generated content.
• Answer the next question. The reader who clicked through has already had the easy one answered. Their real question is what to do about it.

What has to change, and what it costs

Four changes follow, and none of them is free. Stop publishing definitional articles and consolidate the definitions into a single glossary. Stop counting published pages as progress. Run a fail-closed uniqueness check that blocks any article overlapping another site you publish, because near-duplicates across your own domains are the same problem as near-duplicates across the web. Put your operating history on the about page, including the parts that do not flatter you.

The cost is real: the publishing rate falls, and pages already paid for go into the archive. Illustrative arithmetic: if a guide takes six hours and you value your time at EUR 40, that is EUR 240 a piece; cutting a plan of forty thin pages down to twelve substantive ones does not save money, it moves EUR 2,880 of effort from pages that would not have been indexed to pages that might be. That is the trade, stated honestly. Our content operations guide covers how to run the reduced cadence without the queue drying up.

The measurement problem: how you would even know this is working

This is the part most articles on the subject skip. If you publish better pages and traffic does not move for six months, you cannot tell whether the strategy is wrong or the site is simply too young. Our answer is to measure the things that respond quickly and are not confounded by ranking:

Two lessons from our own data belong here. Most of what looks like early traffic on a new affiliate site turns out to be query spam and automated clicks rather than readers — which is why bot and human have to be separated before any chart is read. And the structured-data-as-an-AI-lever theory does not hold up in this category: extensive markup on affiliate pages rarely produces a single rich result, because most page types here are not eligible for one. Keep markup that describes the page honestly; do not budget for it as a traffic strategy. The SEO guide covers the measurement setup, and we track the wider policy picture in our running notes on Google updates for affiliates.

One constraint does not move with the search results: gambling content is for adults only, 18+ or the legal age in your market. Whatever you publish about operators or offers carries the licensing frame and responsible-gambling messaging with a route to help, and no amount of first-hand detail substitutes for it.

What we would do this week

1. Audit your last ten published pages against one question: what is on this page that a model could not have assembled from other pages? Pages with no answer get merged, rewritten with a real input, or removed — decide this week, not eventually.
2. Create the first-hand asset you do not have. Apply to two affiliate programs and document the whole process: what was asked, what was refused, how long it took, what the dashboard shows. That is a page nobody else can publish.
3. Set your floors and write them down — minimum length, minimum sections, minimum share of net-new substance, and a rule that nothing publishes without one number or clause you personally verified. Then hold the cadence you can actually verify, and no faster.