When Readers Prefer The Machine: What New AI Writing Research Means for Tech Writers

Here is the part many writers would prefer to discuss only in low light, after a drink, and with access to a fainting couch: readers may not value human writing nearly as much as writers value the idea of human writing.

That is not a small emotional inconvenience. It is a professional one.

If readers consistently prefer clear, frictionless AI prose, then a lot of sentence-level writing starts to look less like a rare craft and more like a commodity. That should get the attention of anyone whose job involves producing words for a living.

So yes, technical writers should care.

The Quiz That Made Writers Flinch

In a New York Times blind quiz discussed by Reid Hoffman, readers were asked to judge short passages without being told whether a human or a machine wrote them. Hoffman, for anyone mercifully untouched by years of Silicon Valley self-congratulation, is the co-founder of LinkedIn, a PayPal alum, a venture capitalist at Greylock Partners, and one of the most visible public cheerleaders for AI. That makes his interest in the results especially worth noting. He’s not some neutral birdwatcher peering at the AI phenomenon through binoculars; he’s standing in the aviary feeding the things.

By the public summaries of the quiz, more than 86,000 participants took part, and readers slightly preferred the AI-written passages overall.

That result may sting, but it is not as shocking as writers might hope.

I saw a smaller, thoroughly unscientific version of the same pattern with my own Facebook audience. When I shared short passages and asked people to guess whether a human or AI wrote them, they often could not tell the difference. More awkwardly, they tended to prefer the AI-created version. No grant funding. No journal publication. No one in a lab coat. Just enough evidence to make a person who has spent years trying to write well stare at the ceiling like it personally betrayed him.

Why The Machine Often Wins

Research outside that quiz points in a similar direction. A 2024 study in Scientific Reports found that readers could not reliably distinguish AI-generated poetry from human poetry and often rated the AI poems more favorably. The likely reason was not that the chatbot had become a secret poet. It was that the AI poems were easier to process, easier to understand, and less demanding.

Readers often reward prose that glides.

That fits a broader body of research on cognitive ease. A 2024 Science Advances study found that readers were more likely to choose simply written headlines over more complex ones. This is because most people are busy, distracted, mentally overbooked, and trying to absorb information while life bangs pots together in the background. The easier language path usually wins because it asks less from them.

This is where the illusory truth effect matters. The illusory truth effect is the tendency for repeated or easy-to-process statements to feel more believable, even when they are false. In plain language, if something sounds smooth, familiar, and well packaged, the brain may start treating it like truth’s respectable cousin.

Learn more: Encyclopaedia Britannica provides a solid overview in its discussion of misinformation and disinformation.

AI is exceptionally good at producing fluent prose. It likes symmetry. It likes polished transitions. It likes sentences so tidy they could be folded into thirds and filed under “Looks Correct.”

It is the beige cardigan of writing: soft, dependable, broadly acceptable, and just interesting enough to get invited back.

Human prose, by contrast, often has sharp edges. It swerves. It startles. It sometimes arrives wearing a leopard-print scarf and saying the one thing everyone else was too polite to say. AI tends to iron that out. For many readers, THAT ironing reads as quality.

The Risks Hiding In Plain Sight

If this trend continues, the biggest risks are not simply that AI will write more. The deeper risk is that it may quietly reshape how people think, judge, and communicate.

One problem is style flattening. If AI keeps producing the kind of prose readers reward, organizations may start converging on one dominant style: calm, polished, generic, and emotionally pre-sanded. Over time, readers may begin to equate “familiar machine smoothness” with “good writing.” That narrows the range of what sounds credible and leaves less room for originality, tension, or voice.

Researchers have already found evidence that AI-assisted writing can reduce diversity across outputs, making different authors sound more alike and nudging writing toward more standardized styles.

Related: The homogenizing effect of large language models on human expression and thought

Anothe