The 90-Day Crash Nobody Wants to Talk About
The story is the kind of thing that makes content teams sit up a little straighter. Twenty new domains. One hundred AI-generated articles on each. Two thousand articles in total, spread across business, e-commerce, and entertainment. At first, it looked like the dream was working. Indexing climbed fast. Impressions jumped from 120,000 to 520,000 in the first couple of months. Then the floor vanished. By month three, 97 percent of pages reportedly fell out of the top 100 results. The phrase “death spiral” sounds dramatic, but for anyone betting a content strategy on pure volume, it probably feels accurate.
The Problem Wasn’t AI. It Was Empty Scale.
The smartest comments didn’t turn this into a lazy “AI content bad” sermon. They made a sharper point: AI content isn’t the problem. Low-quality scaled content is. That distinction matters. Search doesn’t collapse because a model helped draft an article. It collapses because the article adds nothing. No depth. No first-hand insight. No expert judgment. No reason to exist beyond catching a keyword. It may index. It may even spike. But if the whole thing is built on generic answers wrapped in SEO formatting, the foundation is basically wet cardboard.
One person described the pattern cleanly: mass AI content, low depth, no real expertise, no differentiation, and too much faith in volume. It works for a while, then gets wiped when an update hits. That’s the brutal lesson here. Short-term visibility can trick teams into thinking they’ve found leverage. In reality, they may have only found a delay between publishing and punishment. The graph looks exciting until it starts looking like a cliff.
Original Evidence Is the New Content Moat
The most useful takeaway was simple: ask what value you’re adding that AI doesn’t already know. That question should be taped above every content calendar. If the answer is “nothing, but we rewrote the same topic with a slightly different heading,” the page probably doesn’t deserve to rank for long. The comments kept circling the same fix: original data, customer interviews, first-party findings, tests, examples, expert commentary, and actual editorial judgment. In other words, evidence.
One person called it an “evidence pipeline,” and that phrase gets right to the point. Teams shouldn’t scale output before they have a reliable way to add unique inputs. Every article should include something the model could not have guessed from public information alone: a customer quote, a product benchmark, a real campaign result, a field observation, a support-ticket pattern, a teardown from someone who has done the work. AI can help shape and draft the piece. It should not be the only source of truth.
Fundamentals Still Matter, Annoyingly
There was also a funny, painful comment from someone with a senior developer background who jumped into AI websites and quickly realized they didn’t understand as much SEO as they thought. That may be the quietest warning in the whole thread. Technical confidence is not the same as search competence. No amount of AI helps much if the site has weak fundamentals, poor internal linking, duplicate content issues, thin pages, bad structure, sloppy keyword targeting, or no clear authority signals.
Another commenter listed the usual disaster ingredients: thin content, unnatural language, duplicate issues, weak E-E-A-T, and no human oversight. None of that is new. AI just makes it easier to produce those problems at industrial speed. That’s the real danger. A bad blog strategy used to fail slowly because humans could only write so much. Now a bad strategy can generate hundreds of weak pages before anyone has time to ask whether the site deserves the traffic it’s chasing.
The Skeptics Have a Point Too
Not everyone was convinced the case study proved some massive universal truth. One commenter pushed back hard, basically saying the traffic loss sounded tiny in clicks and didn’t prove much. That’s worth including because marketers love turning every scary chart into doctrine. A single case study, especially one involving new domains and programmatic publishing, shouldn’t be treated like a law of physics. New sites are fragile. Low-authority domains are risky. Search volatility is messy. Sometimes a dramatic percentage drop hides a small absolute impact.
But skepticism doesn’t erase the warning. Even if the case is imperfect, the pattern is familiar enough to matter. Sites can get early impressions from low-competition or long-tail queries, then lose them when quality, authority, and usefulness get re-evaluated. The exact numbers may be debatable. The strategic lesson is not. If your entire SEO plan depends on Google continuing to reward generic pages from low-trust domains, you’re not building a media asset. You’re borrowing traffic from the future and hoping the invoice never arrives.
The Survival Plan Is Slower, But It’s Real
The comments pointed toward a more durable playbook. Use AI, but don’t let it become the strategy. Build around human-led systems. Add original research. Interview customers. Use internal data. Run tests. Merge or prune weak pages every month. Fix technical SEO. Improve topical authority. Publish fewer pieces if that means each one has a reason to exist. That sounds less exciting than “2,000 articles in 90 days,” but it also sounds like something that might survive contact with an algorithm update.
There’s a deeper emotional shift happening here. For a while, AI content felt like a loophole. A way to flood the web, grab impressions, and outrun slow competitors. Now the loophole is starting to look like a trap. The teams that win won’t be the ones publishing the most. They’ll be the ones with the best inputs. Better evidence. Better editorial taste. Better site architecture. Better reasons for people and search engines to trust them.
AI can still be useful. It can speed up outlines, summarize interviews, clean drafts, suggest angles, and help organize research. But it can’t invent authority that a site hasn’t earned. It can’t fake lived experience forever. It can’t turn recycled information into a durable moat. The content farms wanted a shortcut. What they got was a countdown.
The uncomfortable truth is that SEO didn’t suddenly become new. It just became less forgiving. If a page doesn’t add something real, it’s vulnerable. If a site has no authority, it’s exposed. If a team scales before it thinks, the crash may arrive faster than the traffic report. The AI content gold rush isn’t over because AI stopped working. It’s over because empty scale stopped looking clever.