Humans are starting to sound like AI in spoken conversations
AI model collapse was the concern. Now I wonder if human communication will collapse instead.
Ever felt like AI is changing how we write and talk? It's not your imagination, according to a study that concluded LLMs are changing how we speak. It arrived at this conclusion after evaluating 360,000 YouTube videos and 771,000 podcasts.
Influence of LLMs
The paper (pdf) "Empirical evidence of Large Language Model's influence on human spoken communication" was first published in September 2024 and substantially updated in July 2025.
The paper found a "measurable and abrupt increase" in the use of words preferentially generated by ChatGPT after its release.
The report noted: "These findings suggest a scenario where machines, originally trained on human data and subsequently exhibiting their own cultural traits, can, in turn, measurably reshape human culture."
Or put simply, our use of LLMs has resulted in a "closed cultural feedback loop" where cultural traits circulate bidirectionally between humans and machines.
Indeed, v1 of the report noted how a random sample of 50 manually reviewed videos found that the majority who spoke the word "delve" were not looking at a script.
Clear communication
In the same vein, AI use is pervading written content and social media platforms. To me, the issue goes beyond sounding like AI, but is about offering genuine insights and clarity of communication.
Depending on how it's applied, AI has a tendency to muddy or produce averaged-out content that is fluff. And those who are most reliant on it are unlikely to notice because they simply don't read at that level.
This is something B2B firms must pay special attention to. Just because you don't notice the AI slop doesn't mean the experts you're marketing to won't. Some will, and it can harm the brand.
On LinkedIn, I personally use an AI detector I trust to give me an additional signal about whether a creator is worth paying attention to.
Anyway, there were initial concerns about potential AI model collapse. Now, I can't help but wonder if it is human communication that will "collapse" instead.
