I've spent a fair amount of time criticizing AI-generated writing lately. It's often too polished, too structured, and too eager to explain everything. I've even developed an editorial guide to keep LLMs from turning my own writing into something I wouldn't recognize.
But there's an assumption in that criticism I hadn't given much attention.
I can generally get my ideas onto a page without too much trouble. Not everyone can, and the ability to communicate something well isn't necessarily a reflection of how well you understand it.
Consider a researcher who speaks English as a second language. They might have an excellent understanding of their subject, conduct rigorous experiments, and arrive at genuinely interesting findings. But communicating that work in polished academic English presents an entirely separate challenge.
If an LLM helps them organize and express those ideas, have they contributed any less intellectually?
I don't think so.
A 2026 study of graduate students found that generative AI improved writing quality, with particularly strong benefits for non-native English speakers. In that study, AI assistance helped close the measured writing-quality gap between native and non-native speakers.
Research published in the Journal of Experimental Psychology found that people evaluated creative writing less favorably when they believed AI had been involved. The perceived authenticity of the work mattered, even apart from the text itself.
That makes me wonder whether some of the resistance to AI-generated writing is actually about gatekeeping.
We've spent a long time associating strong writing with intelligence, education, and professional credibility. But access to good editing, writing instruction, and even the time required to develop those skills has never been equal.
AI potentially makes some of that assistance available to people who wouldn't otherwise have it.
Of course, there's another problem I've been exploring in my own writing. An LLM can help communicate an idea while also making decisions about how that idea should be expressed. And those decisions aren't necessarily neutral.
A 2025 study presented at CHI found that AI writing suggestions pushed participants in India toward more Western styles of expression, sometimes changing the cultural details of what they wrote.
That's a real concern. Helping someone communicate shouldn't require them to adopt someone else's way of thinking or expressing themselves.
I'm increasingly skeptical of how quickly we dismiss AI-assisted writing as less authentic, though.
Someone using an LLM to articulate an idea they already understand is doing something quite different from asking it to supply the idea, reasoning, and conclusion.
I wonder how much of our criticism actually makes that distinction. And how much comes from discomfort with the fact that polished writing is no longer quite the exclusive signal of expertise it once was.
New notes, occasionally.
RSS