AI Overviews Cite AI-Generated Content More Than Human Writing

AI Overviews are AI-generated content, which means they can contain hallucinations.

Google uses “grounding” to improve their accuracy, but according to our research, AI Overviews are more likely to cite AI-generated content than human-written content.

Here’s what we found:

Keywords Explorer and extracted the top three cited links. 1.9 million URLs in total, of which we had 500,000 in our database.

We ran each URL through our own AI content detector, which is part of Page Inspect in Site Explorer.

Running my blog post through AI Content Detector

Here’s what our content detector found:

3.6% of pages cited in AI Overviews were categorized as “pure AI.”
8.6% were categorized as “pure human.”
87.8% were categorized as a mix of two.

Of the ones that were a mix of both human and AI:

11.2% showed minimal AI use (1-10% of the page content was categorized as AI)
44% showed moderate AI use (11-40%)
24.7% showed substantial AI use (41%-70%)
7.9% showed dominant AI use (71%-99%)

How many pages in AI Overviews are created/assisted with AI

These findings become even more striking when compared to our previous research on AI content across the web. In our analysis of 900,000 new pages, we found that:

2.5% of pages were categorized as “pure AI.”
25.8% were categorized as “pure human.”
71.7% were categorized as a mix of the two.

Even though this research looked only at new pages (and not all cited URLs in AI Overviews will be new), this suggests that Google’s AI Overviews might show a bias toward citing AI-generated or AI-assisted content compared to the general distribution of content on the web.

Sidenote.

No AI content detector is perfect. Like LLMs, AI detectors are statistical models. They deal in probabilities, not certainty. They can be incredibly accurate, but they always carry the risk of false positives. You can learn more about how AI detectors work, and why they’re useful, in these articles: