AI creates content, AI summarizes and surfaces that content, and AI trains on that content again. This cycle is already in motion.
Everyone senses it intuitively, but you need to see the data to grasp the full extent. Across three axes — readers, writers, and AI models themselves — here’s what’s happening right now.
1. Readers — Clicks Are Disappearing
60% of Searches End Without a Click
SparkToro’s analysis of 2024 Datos/Semrush clickstream data found that out of every 1,000 Google searches in the US, only 360 resulted in a click to the open web. The rest are zero-click searches — users finding their answer on the search results page itself and leaving.
For queries where Google’s AI Overview appears, that rate climbs to 83%.
When AI Overview Appears, CTR Drops 61%
Seer Interactive analyzed 25.1 million organic impressions across 42 organizations and found that for informational queries where AI Overview was displayed, organic CTR fell from 1.76% to 0.61% — a 61% decline. Paid ad CTR also dropped 65-68%.
Google-Sourced Traffic Down by a Third Globally
According to Chartbeat data cited in the Reuters Institute’s 2026 outlook report, Google search referral traffic to over 2,500 publishers worldwide dropped by roughly one-third in a single year. The US was hit hardest at -38%, while smaller publishers saw declines up to -60%. Utility-type content like lifestyle, weather, and TV schedules was particularly affected.
But Getting Cited Changes the Game
The same Seer Interactive study revealed an interesting twist. Brands that were cited as a source in AI Overview actually saw organic CTR that was 35% higher — compared to those that weren’t cited.
That said, this is correlation, not causation. Brands with higher authority are more likely to be cited in the first place. Still, the implication is clear: in a world where AI has become a summarization layer, becoming “a source worth citing” is the new competitive advantage.
2. Writers — AI Has Already Crossed the Majority Line
AI Content Detected in 74% of New Web Pages
Ahrefs analyzed 900,000 new English-language web pages (one per domain) as of April 2025 and found that 74.2% showed detectable AI-generated content.
Context matters, though. Of these, purely AI-generated content (almost entirely machine-written) accounted for just 2.5%. The vast majority was a mix of human and AI content. “AI is writing the articles” is less accurate than “it’s becoming rare to write without AI.”
In English Articles, AI-Written Content Has Overtaken Human-Written
Graphite’s analysis of 65,000 English-language URLs from CommonCrawl found that as of November 2024, AI-generated articles had numerically surpassed human-written ones. Before ChatGPT’s launch, AI detection rates hovered around 4-10% — this shift happened in less than two years.
However, according to Graphite’s analysis, these AI-generated articles get almost no exposure in Google search or ChatGPT results. Volume has exploded, but reach remains limited.
The Synthetic Content Ratio Keeps Climbing
The Europol Innovation Lab cited expert estimates in a report suggesting that “up to 90% of online content could be synthetically generated by 2026.” That figure itself has drawn plenty of criticism as an overstatement, with some pointing out it’s mathematically unrealistic.
But the direction is hard to deny. AI-generated ratios are climbing rapidly across text, images, and video. At this pace, a future where “most content has been touched by AI in some form” is already close at hand.
3. The AI Models Themselves — A Snake Eating Its Own Tail
Model Collapse: AI Trained on AI Data Breaks Down
The phenomenon warned about in a 2024 Nature paper (Shumailov et al.) is Model Collapse. When AI models are repeatedly trained on data generated by other AI, they progressively forget the distribution of the original data with each generation.
It starts with the tails of the distribution disappearing first — rare data, minority perspectives. This is called Early Model Collapse. If it continues, the entire distribution converges into something completely different from the original, reaching Late Model Collapse.
Think of it like photocopying a photocopy — each generation gets blurrier. As AI-generated content fills the web and that content gets scraped for training data, the models gradually lose diversity and converge toward the mean.
Why This Matters for Readers Too
Model Collapse looks like a technical problem, but it ultimately circles back to content consumers.
- As AI loses diverse perspectives, recommendations and search results become homogeneous
- Minority voices, niche genres, and fringe topics disappear from training data first
- A future where “every AI gives similar answers” isn’t a technical limitation — it’s the result of data contamination
There are counterarguments to the Nature paper. Some research shows that training on mixed data (human + AI) can mitigate collapse. But the core problem is that distinguishing and collecting “purely human data” is getting harder by the day.
So, What Should We Do?
Synthesizing these three trends reveals a single structure:
AI produces content, AI mediates content, and AI trains on that content.
If uniquely human experience and perspective get diluted in this cycle, the quality of the entire ecosystem degrades.
| Axis | What’s happening now | Implication |
|---|---|---|
| Readers | Consuming without clicking, declining reach to originals | ”Being a source worth citing” is the new competitive edge |
| Writers | AI-assisted writing is the default | The volume game is over; unique perspective is the differentiator |
| AI Models | Training on synthetic data leads to diversity loss | The value of human-original data is rising |
This leads to a paradoxical conclusion. The more AI floods the content landscape, the more valuable writing becomes that’s rooted in firsthand experience and original thought. What AI can’t replicate is experience, and the perspective that comes from it.
The data points in this direction. The question is whether individual creators and small publishers can survive the traffic cliff in the meantime. That’s a topic for another post.
References:
- SparkToro, “2024 Zero-Click Search Study” (Datos/Semrush clickstream data)
- Seer Interactive, “AIO Impact on Google CTR” (June 2024 - Sept 2025, 42 organizations)
- Reuters Institute / Chartbeat, “Journalism and Technology Trends and Predictions 2026”
- Ahrefs, “What Percentage of New Content Is AI-Generated?” (April 2025, 900K pages)
- Graphite, “More Articles Are Now Created by AI Than Humans” (CommonCrawl 65,000 URLs)
- Europol Innovation Lab, “Facing Reality? Law Enforcement and the Challenge of Deepfakes”
- Shumailov et al., “AI models collapse when trained on recursively generated data”, Nature (2024)