The gistA Graphite study, run with the Copyleaks detection tool and published May 19, 2026, finds that roughly 50% of new web articles are AI-written. But those articles rarely rank in Google's top results and are nearly absent from ChatGPT's answers. The parity is about production volume, not visibility. For a business, writing more with AI is pointless if the content stays generic.
On May 19, 2026, analytics firm Graphite published a study, conducted with the Copyleaks detection tool, putting the share of new web articles written by AI at roughly 50%. The finding was picked up the next day by Search Engine Land in its weekly coverage. It is the first time an independent measurement places machine content at parity with human content across the web.
The method: Graphite analyzed a sample of more than 55,000 pages published between 2020 and March 2026, keeping only articles of at least 100 words with verifiable publication dates. Copyleaks served as the independent detection platform, classifying each text as "human," "AI-generated" or "AI-assisted."
The curve: fast climb, then a plateau
AI content did not grow in a straight line. It surged right after ChatGPT shipped in November 2022, then leveled off.
Since early 2025, the share has hovered around 50% without crossing it. Machine content, the study says, concentrates in news updates, how-to guides, lifestyle posts, product reviews and explainers. In other words: generic informational content, the kind that aims to inform without an angle of its own.
The real signal: this content doesn't rank
This is where the study gets interesting for anyone doing SEO. The 50/50 parity describes production volume, not visibility. AI-generated articles rarely appear in Google's top results and are largely absent from ChatGPT's browsing summaries, as Stan Ventures' analysis of the same study points out.
The reason is mechanical. When a tool generates a "how to do X" article, it produces a page statistically very close to a thousand other pages on the same topic. Google has no reason to favor the thousandth copy. That is exactly the pattern seen in the study of 220 sites that lost traffic after publishing AI content at scale, and in the March 2026 core update that cut traffic 71% for sites packed with thin AI content.
Bottom line: producing at half the cost has never been easier. Getting visible has never been harder. The competitive edge is no longer quantity, it is what AI cannot manufacture on its own.
What this means for a business
If your content strategy rests on "publish more articles than competitors," this study makes it obsolete. Your competitors are publishing on the same conveyor belt, with the same tools, and everyone falls into the same filter. The web just absorbed millions of interchangeable pages that search engines have learned to ignore.
The problem is not AI as a tool. The problem is content with no signal: no data of your own, no field experience, no angle. Google now evaluates those signals closely, as shown by the evolution of E-E-A-T criteria applied to AI content.
What to do now
- Audit your existing content. Spot the generic articles that generate neither traffic nor conversions. Rewriting 10 pages with a sharp angle beats publishing 50 more.
- Inject what AI cannot invent. Your client numbers, field feedback, concrete cases, a firm opinion. That is the difference between a page that gets cited and a page that gets ignored.
- Use AI as an accelerator, not an author. Structuring, first drafts, research: yes. Raw publishing with no expert review: that is the direct path to the filter.
- Think GEO as much as SEO. AI engines cite sources that carry authority and are cleanly structured. See our analysis of GEO adoption and strategy for 2026.
What this study does not say
A few limits to keep in mind. AI detection tools, Copyleaks included, are not infallible: a heavily formatted human text can be misclassified, and a heavily rewritten AI text can pass for human. The study measures the presence of AI content, not its quality or its performance page by page. Finally, the sample covers the general web: the split can vary widely by industry. The finding is solid on the trend, less so on the exact decimals.
Our take
At Cicéro, this result does not surprise us, it reassures us. For two years, the fear was that AI would drown good content. The opposite is happening: AI is drowning mediocre content, making good content rarer and therefore more visible. Half the web just became noise. The other half, the half with an angle, evidence and a voice, has never been worth more.
Frequently asked questions
According to the Graphite study published May 19, 2026 and run with Copyleaks, about 50% of new web articles are AI-written. The share reached 35.9% one year after ChatGPT's launch, around 48% after two years, then plateaued near 50% since early 2025.
No. The study shows AI-generated articles rarely appear in Google's top results and are largely absent from ChatGPT's summaries. The parity between human and AI content is about production volume, not search visibility.
No, but AI alone is no longer enough to stand out. Purely generated content joins an ocean of near-identical pages Google ignores. What ranks is content that adds proprietary data, verifiable expertise and an editorial angle AI cannot manufacture on its own.
Sources
- → Copyleaks press release (GlobeNewswire), May 19, 2026 — announcement and methodology of the Graphite study
- → Stan Ventures — analysis of why AI content rarely ranks on Google
- → Search Engine Land — SEO coverage of the study, May 20, 2026
Growth and SEO & GEO content strategy specialist, I founded Cicéro to help businesses capture lasting organic visibility — on Google and in AI-generated answers. Every piece of content we produce is designed to convert, not just exist.
LinkedIn