Someone Actually Built the AI Affiliate Sites Everyone Warns About. Here is What Happened

There are two kinds of people writing about AI-generated affiliate content right now. The first group explains in theory why it is a bad idea. The second group builds the sites anyway to find out for themselves.

Tim Kraft, a growth and SEO operator writing for Search Engine Land, falls into the second category. His experiment with AI affiliate sites is worth reading in full, not because the results were surprising, but because the data is unusually clean, and the conclusion he draws from it goes further than most people are willing to go.

The setup was straightforward. Kraft bought three cheap partial-match domains targeting “best welding schools,” “best plumbing schools,” and “best electrical schools.” He used AI to build the sites, fetched publicly available data through a Python API call, and used ChatGPT to template the subheadings and paragraph text that typically appears across ranking pages in those categories. Within a few hours, he had published thousands of bottom-funnel pages across three sites. Every trust signal was missing: no brand, no authorship, no original research, no backlinks earned through anything other than the domain itself. Aggressive internal linking was built for crawl coverage, not user intent.

The sites worked. For a while.

What Google's Spam Systems Actually Did

Indexation was fast. The pages surfaced for long-tail queries and impressions climbed quickly. Within the first couple of months, all three sites were generating around 200 in-market clicks each. Then, during the first December spam update after launch, clicks dropped to zero. They never recovered.

Kraft tried data updates and performance plugins. Neither moved the needle.

The failure itself is not the news. Anyone paying attention to how Google's spam crackdown has reshaped affiliate content already knew this outcome was likely. What is more interesting is his observation about what Google was actually doing during those first few months. He writes that Google “tolerated them just long enough to learn from them.”

That framing matters. The sites were not immediately penalised. They were indexed, sampled, assessed, and then removed from relevance once their pattern was understood. Google's updated spam definitions now cover mass content production with little originality and content aggregated from public data with no added value. These three sites were a clean example of both. There was no single tactic that killed them. The combination of zero trust signals, templated AI content, and no defensible original value is what made them disposable.

The Actual Argument Kraft Is Making

Kraft is careful not to declare affiliate content dead. His position is more precise: affiliate content marketing works as a monetisation layer, but it does not work as a growth engine. That distinction is one more affiliate marketers should sit with.

A site where affiliate revenue is earned on the back of genuine, useful content built around real expertise and community trust, which is the model he describes running elsewhere before this experiment, can still perform. The trap is treating affiliate links as the point, and content as the delivery mechanism for getting in front of search queries. Google's own guidance on AI content has said as much directly: if the reason your content exists is to attract search engine visits, that is not what their systems are built to reward.

What is worth pulling out separately is the observation about what replaces the old model. Kraft points toward verticalised research, benchmarking, and content that sparks real community conversation rather than ranking for individual queries. He cites Stripe's “Developer Coefficient” and HubSpot's “State of Marketing” as examples of the direction things are heading: content that earns distribution through owned channels, paid media, and partnerships rather than relying on organic search as its primary vehicle.

This is a live concern for affiliate marketers, not an abstract one. The great affiliate bypass is already happening. AI is citing affiliate content sources while skipping the click. Over 58% of Google searches now end without a click to an external site. The argument for building content around community, first-party data, and owned audience relationships is not theoretical anymore.

What This Means For Program Managers And Publishers

The experiment confirms what Google's crackdown on mass-produced SEO content has been pointing toward for some time. Programmatic, low-trust content at scale is not a distribution shortcut. It is a liability with a delayed fuse.

For program managers evaluating publisher partners, this is a useful filter. A publisher whose traffic is built on templated AI content and public data aggregation with no original research, no authorship signals, and no brand is a publisher whose traffic can be zeroed out by a single spam update. That is not a stable affiliate partnership. It is a countdown.

SEO predictions for 2026 consistently point in one direction: the publishers who survive will operate more like media brands than search arbitrage plays. That means audience relationships, email lists, community, video, and content that earns trust rather than briefly exploiting a gap in how a search system processes new pages.

Kraft's experiment is a useful proof point because it removes the theoretical. He built exactly the kind of sites that a lot of people are still quietly building or commissioning. He documented what happened. The sites ranked briefly, then stopped. Nothing he tried brought them back.

The cleaner lesson is buried in his closing line: search marketing now is about building things that cannot be easily copied with AI. That is a harder brief than publishing thousands of templated pages. It is also, finally, the only brief worth working to.