I checked the viral chart about AI content and Google. Half of it holds up.
That chart where a site dumps out tens of thousands of AI articles, shoots up and then falls off a cliff keeps doing the rounds. The phenomenon is well documented. The specific cases used to illustrate it usually aren't — and Google says something rather different from what gets quoted.
I came across a reel on Instagram with a chart that keeps circulating. A brand dumps out tens of thousands of AI articles, organic traffic climbs for a few months, holds for a while, then falls and never stops falling. The shape looks like a mountain. The punchline is usually "Google worked it out", followed immediately by a link to a course.
I wanted to know how much of that actually holds up, so I checked it against primary sources. The conclusion splits awkwardly down the middle: the phenomenon is real and decently documented. The specific story used to illustrate it in that reel, I cannot verify at all.
What Google actually says
I'll start here, because this gets misquoted almost every time.
The relevant policy is called scaled content abuse and it landed in March 2024. The current wording on Google Search Central:
"Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users. This abusive practice is typically focused on creating large amounts of unoriginal content that provides little to no value to users, no matter how it's created."
That last phrase — no matter how it's created — is precisely the bit that gets dropped in the quoting. The policy is not about AI. It is about unoriginal, valueless pages mass-produced for rankings, and it applies exactly the same way to a content farm staffed by underpaid humans.
Google also has a separate piece on AI content from February 2023, and it is similarly unambiguous:
"Appropriate use of AI or automation is not against our guidelines. This means that it is not used to generate content primarily to manipulate search rankings, which is against our spam policies."
And, bluntly:
"Using AI doesn't give content any special gains. It's just content. If it is useful, helpful, original, and satisfies aspects of E-E-A-T, it might do well in Search. If it doesn't, it might not."
So the honest reading is narrower than either camp would like: AI content is not penalised for being AI. What gets penalised is producing masses of worthless pages for rankings — whoever wrote them.
The forty per cent nobody measured
One more number that keeps getting repeated wrongly. In March 2024, Google wrote that "the combination of this update and our previous efforts will collectively reduce low-quality, unoriginal content in search results by 40%". That is a forward-looking expectation for a bundle of measures, not a measured outcome of a single update. Google later revised it to 45%. Every time someone writes "the March update removed 40% of AI content", they are quoting a prediction as a result — and changing its scope while they're at it.
The measured numbers came from third parties. Originality.ai tracked 49,345 sites and 837 of them vanished from the index entirely; across a wider sample of roughly 79,000 sites it found 1,446 manual penalties. In the deindexed sample, every site showed signs of AI content and roughly half was 90–100% AI-written. Footnote: Originality.ai sells an AI detector, so it is not a disinterested source — but unlike most claims in this area, it has published its methodology and sample size.
The data that genuinely supports the phenomenon
The best evidence for the mountain shape isn't the viral chart. It's Lily Ray's analysis from May 2026, which looked at 220+ sites that AI content platforms themselves published as their success stories. In other words, the cases the vendors picked as the best they had.
| Outcome | Share of sites |
|---|---|
| Lost 30% or more from their organic peak | 54% |
| Lost 50% or more | 39% |
| Lost 75% or more | 22% |
The typical arc: pages pile up for 6–12 months, traffic peaks three to six months after content production peaks, and over the following year most of the gain is wiped out — often below the starting point. Most of the declines came after the case studies were published. A number of those sites later deleted or redirected the very pages they had been praised for.
A related analysis by Malte Landwehr and Tomek Rudzki found that among one AI content tool's flagship clients, 57% declined and 22% showed the classic spike-and-crash; for a second tool, 75% of the showcased clients lost visibility.
One methodological detail is worth noting: both analyses deliberately withhold the names of the sites and the vendors. Ray's summary was that the pattern is the story, not the individual players. That is a stricter approach than it first appears, and it is exactly why this data is more trustworthy than named anecdotes off social media.
Where the reel fell down
The specific example used to illustrate it: the brand Momcozy supposedly went from around 700 pages to 62,000, spiked, then crashed.
I could not verify this anywhere. No write-up from the SEO industry, no visibility chart from Ahrefs or Semrush, no report — and nothing on the website of the person presenting it either. The one substantial public SEO analysis of that brand is positive about it and mentions neither a page explosion nor a crash.
That doesn't mean it's false. It means it is an unsourced anecdote delivered in the visual language of data — and it shouldn't be passed around as fact, especially when properly conducted research across hundreds of sites reaches a similar conclusion, better.
The counter-argument that belongs in the same piece
In fairness. Ahrefs studied 331,000 pages and concluded that Google doesn't punish AI content, it punishes bad content. That fits both the policy quoted above and Ray's data: what correlates with the decline is thin, unoriginal publishing at volume, not the involvement of a language model.
It is also true that the first half — the climb — is real, and can be substantial. Published cases show sites going from hundreds of thousands to millions of organic visits in two months on the back of AI content. The question isn't whether it works. The question is how much of it is still there in eighteen months — and on the vendors' own hand-picked examples, roughly half of it isn't.
What I take from it
The risk factor is volume, not the tool. When a plan starts with a page count instead of a list of questions customers actually have, it's the plan that's wrong. Publishing 60,000 pages fails on its own terms before Google even enters the picture — nobody can keep 60,000 pages current.
Using AI to write is entirely fine. Having a model draft, restructure, translate or edit text that a person who knows the subject then finishes and puts their name to has nothing to do with any of this. Banning it outright is an overreaction.
The test is whether the page would exist without search. If the only reason it exists is a number from a keyword tool, it falls into the category the policy is aimed at — regardless of who wrote it.
Expect your measurement to lag. The most uncomfortable finding in this data is that the traffic peak arrives three to six months after the production peak. Anyone evaluating an AI content programme in month four is reading the part of the curve that always looks good.
Design the exit up front. The sites that recovered did it by deleting or redirecting pages en masse. If you're already doing something of this kind, keep it structurally separate from the rest of the site, so that removing it later is a decision rather than a rebuild.
Why I'm writing this
Because this type of content is treacherous precisely because it is fundamentally right. The phenomenon exists, the data supports it, and the advice "don't dump thousands of AI pages out there" is sound. It just comes bundled with an unverified number and a policy quote that says something other than what's claimed — and the result is a correct conclusion resting on bad reasons.
When I forward something on, I'd like to know which half is which.
Verified 12 August 2026. Policy wording quoted from Google Search spam policies and Google Search's guidance about AI-generated content; the March announcement from Google Search Central and blog.google. Data from Lily Ray, "It Works Until It Doesn't"; deindexing figures from Originality.ai via Search Engine Journal; the counter-argument from Ahrefs. Some figures come from interested parties, which is flagged in the text; treat them as indicative.