What did the experiment reveal about AI-generated content and Google rankings?

The experiment showed that mass-produced AI content, without human edits or EEAT signals, initially gets indexed and some visibility but experiences a significant drop in rankings after about three months, indicating Google's lack of long-term trust.

What is the 'three-month cliff' for AI content?

The 'three-month cliff' refers to the point in the experiment where AI-generated content's visibility collapsed, with pages ranking in the top 100 falling from 28% to just 3%, effectively disappearing from meaningful search positions.

Why are EEAT signals crucial for AI content to maintain rankings?

EEAT (Experience, Expertise, Authoritativeness, Trustworthiness) signals are crucial because Google appears to evaluate whether content shows enough of these qualities to sustain visibility. Without them, even indexed AI content will likely fade from search results over time.

Does Google directly penalize AI content?

The experiment suggests it's not a blunt 'AI penalty' but rather a 'trust test.' Google seems willing to crawl and index AI content initially, but if it fails to build trust through EEAT signals, its rankings are demoted rather than being directly penalized for being AI-generated.

Experiment Reveals Google Demotes AI Content After 3 Months Without EEAT Signals

If there was any doubt left about whether mass-produced AI content can sustain organic rankings on its own, this experiment answers it: not for long.

A large-scale study led by SE Ranking CMO Bogdan Babiak tracked what happened when 20 brand-new websites were built almost entirely from raw AI output. Across those domains, the team published 2,000 fully AI-generated articles in multiple niches, including business, travel, health, finance, and lifestyle. The setup was intentionally bare. There were no human edits, no author bios, no internal links, no backlinks, no images, and no obvious EEAT signals—just AI content published at scale and submitted to Google.

At first, the strategy looked like it might work.

The Early Surge Looked Promising

Within the first 36 days, Google had indexed nearly 71% of the pages. The sites generated more than 122,000 impressions and 244 clicks. Some categories—especially Hobbies & Interests, Business & Services, and Travel & Tourism—picked up traction quickly. Around 80% of the sites ranked for more than 100 keywords, and roughly 28% of ranking URLs appeared in the top 100.

That early movement helps explain why so many marketers still believe AI content can be scaled cheaply and perform in search. On the surface, those first numbers looked encouraging.

But the visibility didn’t last.

The Three-Month Cliff

At around the three-month mark, the experiment hit a wall. Visibility collapsed.

Pages ranking in the top 100 fell from 28% to just 3%. In practical terms, the content disappeared from meaningful search positions. What looked like momentum was really a short testing period. Google seemed willing to crawl the pages, index them, and even surface them for lower-competition queries—but not trust them over time.

That’s the real takeaway. This does not look like a blunt “AI penalty.” It looks more like a trust test. Google appears to evaluate whether content shows enough experience, expertise, authoritativeness, and trustworthiness to keep earning visibility. When those signals are missing, rankings fade.

Why EEAT Was the Missing Piece

The value of this study is that it isolated one variable unusually well: scale without substance.

These were not authority sites. They had no brand equity, no expert contributors, no reputation, and no real human enrichment layered into the content. That makes the outcome especially useful for anyone still hoping that publishing hundreds of AI-written articles on fresh domains is a viable long-term SEO strategy.

It isn’t.

Google may crawl and index AI content quickly, especially when it targets long-tail, low-competition queries. But indexing is not endorsement, and temporary rankings are not durable rankings. If pages fail to build trust over time, they appear to be pushed down.

SEO observers already have a nickname for this pattern: “Mount AI.” You get an early rise, then a steep drop, followed by long-term stagnation.

The Long-Term Results Were Even Worse

The full 16-month dataset makes the failure hard to ignore.

Across 2,000 articles, the experiment produced about 1,092,079 impressions and only 1,381 clicks in total. That works out to less than one click per article over more than a year. Even six months in, growth had already stalled, with roughly 70% to 75% of all activity happening in the first two and a half months.

In other words, the content got its chance. It just couldn’t keep it.

Even a temporary lift during a later Google Spam Update didn’t lead to a meaningful recovery. Visibility spiked briefly, then flattened again. Once those sites lost trust, they didn’t truly bounce back.

YMYL Niches Were Hit Hardest

One of the clearest findings was how poorly the experiment performed in sensitive categories.

Finance and Health—classic YMYL (“Your Money or Your Life”) niches—showed especially weak indexing from the start. That should not be surprising. Google has stronger reasons to demand credibility when content could affect a person’s health, finances, legal standing, or safety.

If raw AI content struggles to hold rankings even in lighter informational niches, it has even less chance in areas where expertise and trust matter most.

What Marketers Should Take From This

This experiment should put to rest the idea that AI can replace a real content strategy. It can speed up production, help with drafting, outlining, summarizing, and scaling workflows. But by itself, it does not create the signals Google needs to trust a page.

The winning model is not AI-only. It’s AI plus human value.

If you want content to last, you need:

Real editorial input
Original examples and insights
Better structure and readability
Internal linking
Experienced authors
Visible trust markers
Site-level authority built over time

AI can help produce the first version, but it cannot manufacture credibility on its own.

The study also points to a more practical role for AI: supporting established websites that already have trust and authority. In that setting, AI can amplify output instead of exposing weaknesses.

The Bigger Shift in Search

All of this is happening while search quality standards are getting tighter, not looser. Between Google’s spam systems, its emphasis on helpful content, and the growing value of citation-worthy information in AI-driven search experiences, generic content is becoming easier to detect—and easier to ignore.

The web does not need more average pages repeating what already exists. It needs content that reflects real experience and adds something distinct.

That’s where many AI-first publishing models break down. They create volume, but not differentiation.

FAQ

Does Google automatically penalize AI-generated content?

Not necessarily. This experiment suggests the issue is not AI alone, but the lack of EEAT signals and trust-building elements around the content.

Why did the pages rank at first?

Google often tests new pages in search results, especially for long-tail and lower-competition queries. Early visibility does not mean long-term trust.

What is “Mount AI” in SEO?

It describes a pattern where AI-generated content rises quickly in rankings, then drops sharply and stays flat over time.

Can AI still be useful for SEO?

Yes. AI can be highly effective for drafting, outlining, research support, and workflow efficiency—especially when human editors and subject matter experts improve the final content.

What matters most for long-term rankings?

Strong EEAT signals, editorial oversight, original value, internal linking, clear authorship, and broader site authority all play a role in sustaining visibility.

Conclusion

The lesson here is simple: AI content itself is not the problem. Unverified, unenhanced, trust-poor AI content is.

If you want to use AI without falling into the same trap this experiment exposed, you need a workflow built around human oversight, authority signals, and long-term search performance. That’s why tools that support smarter, more strategic content creation matter more than ever, and AIuthority is worth a look if you want AI-assisted content built to perform beyond the first three months.