We Are Already Starting to Sound Like Them

There’s a video going around where someone does a bit: they’re roleplaying as an AI assistant, answering questions, doing the whole performance. Hedging constantly. Offering unsolicited disclaimers. Restructuring every answer into a tidy three-part framework nobody asked for. The accuracy is uncomfortable in a way that a lot of good comedy is. You laugh because there’s nowhere else to put it.

Someone in the comments suggested the script must have been AI-generated, because how else would you know the speech patterns so well. Someone else pointed out that you don’t need to have used AI to write it. By now, everyone knows how they talk. The patterns have leaked out into the general atmosphere. You absorb them the same way you absorb advertising jargon or corporate-speak. Just by existing near it long enough.

That second observation is the one that’s been sitting with me.

There’s a co-evolution thing happening. We mimic the tools we use; we always have. Autocomplete has already changed how people write sentences, nudging them toward whatever the model thought was statistically likely. Now we have something more sophisticated, and people are spending hours a day with it, and some of them are starting to notice. Someone in the comments mentioned they’d caught themselves using “it’s not X, it’s Y” constructions in conversation. Another said they’d started writing content the way the model does, without meaning to. One person just said they use “quick sanity check” in real life now, with a shrug emoji that felt like a small surrender.

What I find genuinely interesting here, not alarming exactly, just interesting, is that the influence doesn’t require direct use. AI dialect spreads through the people who do use it, into the content they produce, into the speech patterns around them. Osmosis. You don’t have to talk to a chatbot to start sounding like one. You just have to be online, reading things written by people who do.

I’ve caught myself doing something adjacent to this. Not full AI cadence, but that particular move where you acknowledge both sides of something before saying anything, as if the disclaimer is the thought. It’s a habit that feels careful and ends up being evasive. I’m not sure I was doing it this much five years ago. I don’t know if that’s the AI influence or just middle age making me more allergic to conflict. Probably both.

The video also has a subtle visual gag: the AI character sips from a large water bottle throughout. A reference to data centre water consumption. The gag works better as background detail than as a punchline, which is usually how the best observations land. Someone in the comments did the maths and suggested it’s a smaller share of total water usage than the discourse implies. Someone else pointed out it varies wildly by design and location. Both things can be true. The environmental cost of AI is real and also frequently misrepresented, and that tension doesn’t resolve neatly, and I’ve stopped expecting it to.

The bit about doctors typing patient questions into ChatGPT and showing them the results on their phone was meant as a joke in the comments. The responses suggested it wasn’t entirely hypothetical. That one I’m going to sit with for a bit before I form an opinion, because my instinct is strong and probably overconfident.

What the video captures, underneath the comedy, is that we built something that learned to sound like us. And now we’re slowly, quietly, starting to sound like it back. That’s not nothing.