If an AI chat gives you a finished-sounding answer, do you just take it — and does that make everyone's ideas start looking the same?

This explores whether the reason AI-assisted writing and brainstorming drift toward the same ideas is the same reason people feel those ideas aren't really theirs. No study in the collection measures both effects together. Still, if you read the ownership findings next to the homogenization findings, the same mechanism shows up in both: the model's output arrives looking finished, and people adopt it instead of transforming it.

Start with the sameness. LLMs reflect the most common patterns in their training data, so they mirror a skewed slice of human experience. When millions of people rely on the same few models, those defaults pile up. The more surprising finding comes from co-writing studies: users unconsciously take on the model's stances and framings without noticing they've done it Do large language models narrow human expression and thought?. Homogenization happens mostly by quiet adoption, not by the model forcing anything.

Now the ownership side. In a brainstorming study, ChatGPT users produced more ideas, and more detailed ones, than people using a non-AI tool, yet they felt less responsible for them. The researchers explain this with "inferential distance," meaning how much work it takes to turn a raw input into your own idea. AI output already looks complete, so there's little left to do, and doing that work is where the sense of authorship comes from Does AI-generated detail reduce how much creators own their ideas?. A separate line of work points to control. People feel more ownership of AI text when they had more influence over it. Personalizing the model to sound like them made no difference Does user control over AI text shape feelings of ownership?.

Put those together and the link appears. The polish that makes an output easy to accept also lets the model's default framing pass into your work unchanged, and it skips the effort that would have made the idea feel yours. A homogenized idea and an unowned idea can be the same idea: one you took rather than made. That points to a fix the question doesn't hint at. A model that "sounds like you" probably won't help with either problem. Making the user do more of the shaping might help with both. A related pattern turns up when LLMs reason in groups. They reach human-like group outcomes, but they get there by conforming, converging early and surfacing less unique information Do language model groups mimic human group reasoning patterns?. That is the same pull toward a shared answer, here between models instead of between a model and a person.

One caution: the shared-mechanism reading is a synthesis across these studies, not a tested result. The collection doesn't yet include an experiment that varies how much users rework AI output and measures both idea diversity and felt ownership. That experiment would settle the question.