AI leaves its mark: One in three web texts is now generated by chatbots

The data is not surprising. Not in the least. One-third of the web pages published after the emergence of ChatGPT show some indication that they have been written or edited (tweaked, one might say) by artificial intelligence, a fact that demonstrates the unstoppable trend of this technology on the internet. Or the laziness of those who constantly seek shortcuts.

The appearance of ChatGPT in November 2022 represented a paradise for internet users in the form of content-generating tools. With fewer instructions than an appliance made in China, it can emulate, with a good dose of realism, what is written by a human.

Four years after its arrival in the world, AI as a facilitator is increasingly popular. ChatGPT is accompanied by Gemini or Claude. Furthermore, artificial intelligence has surpassed, for the first time, web traffic from humans on a global level, according to data from Cloudflare.

Pew Research Center (PRC) has verified the percentage of online content written by AI. The result, from last July, is enlightening: around 35% of web pages published after the arrival of ChatGPT show significant indications of having been written or edited by this technology.

PRC used the Common Crawl web archive for its research, with which it collected almost half a million web pages written in English over the past five years, and used the AI detection tool Open Pangram to determine how many of them were likely written or substantially edited by artificial intelligence.

The analysis also shows that it is more likely to find web pages written or edited by artificial intelligence in .com domains compared to web pages with .org, .edu, and .gov domains.

It is also noted that AI-generated texts have certain common characteristics that somewhat give them away, such as the use of the em dash, the Oxford comma, or certain words, including 'delve,' 'interplay,' or 'testament.'

The use of the em dash has doubled compared to 2023, the study indicates, while the Oxford comma has increased by 63%, and negative parallelism (the structure: not X, but Y) has tripled.

PRC officials clarify that the types of punctuation, words, and linguistic peculiarities that appear much more frequently in AI-generated content are not exclusive to this technology, as humans also use them, albeit to a lesser extent.

The reason for their greater use is that AI models have been trained with an overrepresentation of texts that include them, which increases the probability that AI will use them in the content it generates.