I was reading a Substack post recently that had some great original insights. I didn’t get written-by-AI vibes…until I got to the phrase “quietly reshaping.”
“Quietly reshaping” is one of those terms that ChatGPT loves, but I rarely encountered it in writing before the AI era.
Once I noticed this, I started seeing other phrases that are beloved by AI, like “what matters most” and “it’s important to note,” in articles that otherwise seemed to be written by actual people.
Hey, maybe it’s a phrase they’ve always liked and used. Maybe they used AI to help them write.
Or maybe they’ve just seen these phrases so many times in AI-generated writing that they’ve started reaching for them naturally in their own work.
Is AI quietly reshaping the way we write?
How it works
AI tools are trained with gigantic data sets that include language patterns developed by humans. During training and subsequent tuning, AI “learns” that some patterns consistently produce responses users view as helpful, clear, or satisfying. As a result, those patterns tend to appear more often in AI-generated text than they do in typical human writing.
Whether we realize it or not, most of us now encounter AI-generated or AI-assisted writing every day: emails, articles, blog posts, reports, LinkedIn posts…it’s everywhere. Repeated exposure can lead us to unconsciously absorb elements of AI style. AI phrases like “quietly reshaping” start to sound normal.
When enough individuals make these same small shifts, the language itself begins to change. None of this requires people to consciously imitate AI—we see it, so we use it. Linguists call this accommodation: People naturally converge on the language they encounter frequently.
As people begin adopting familiar patterns in their own writing, phrases that once stood out as AI “tells” start to show up in ordinary human-created content. AI isn’t just reflecting human language anymore—it’s evolving it.
The cycle doesn’t end there. As more AI-influenced writing is published online, those same patterns are increasingly reflected in the language future AI models learn from. Future AI may conclude that those phrases are even MORE representative of human writing than before, making them more likely to appear in the next generation of AI output.
What’s the problem?
Language changes all the time. That’s normal and expected. We pick up expressions from many sources, including coworkers, family, and social media. AI is becoming another source of linguistic influence—however, it’s one that’s unusually widespread and happening at warp speed.
AI style isn’t just specific words or phrases. It’s a tendency toward polished, balanced, repetitive, statistically common language. In her 2025 perspective article “The Feedback Loop: Is AI Teaching Us to Talk Like Machines?” Brenda K. Wiederhold points out some of the possible problems that could arise as AI-generated language influences human writing:
- Individual voices may become less distinctive. As people adopt similar AI-generated phrasing, writing may begin to lose the personal style that makes it recognizable and memorable.
- Language may become more homogeneous. If AI favors certain words, sentence structures, and rhetorical patterns, writing across different people and settings may start to sound increasingly alike.
- Linguistic diversity could decline. Regional expressions, idioms, humor, metaphor, and other culturally rich forms of language may become less common if AI consistently favors more standardized phrasing.
- Creativity may be constrained. Predictive models naturally reinforce what’s statistically common, making unexpected or unconventional expression less likely to emerge.
- AI could subtly shape how we think. If people increasingly accept AI’s suggested wording instead of exploring alternatives, language may encourage more predictable patterns of thought and problem-solving.
Wiederhold concludes, “The challenge is to embrace the efficiency and reach of LLMs without surrendering the creativity, nuance, and individuality that define human communication.”
Now what?
As usual, I don’t have the answers—but I do have more questions to consider:
- At what point does an AI “tell” stop being an AI tell because humans have adopted it? Is the concept of an AI tell going to seem quaint soon?
- Is AI style inherently inferior to human style? Should we try to avoid adopting it, and if so, how?
- What are the unique elements of human writing, and should we be more conscious about cultivating and preserving them? Which of those qualities become more valuable because they’re becoming rarer?
- Right now, there’s a lot of focus on writing “sounding like AI.” In the future, will AI-generated (or assisted) writing be assumed? Will instead talk about writing that “sounds human”?
- If we intentionally avoid AI-sounding language today, are we preserving something uniquely human or just resisting the next stage of language evolution?
Part of the issue is that this all feels terrible and inevitable, like AI language is taking over and there’s nothing we can do about it. Defining the real problem and asking thoughtful questions can help us determine what to do next…Actually, that approach works for more or less everything, doesn’t it?