A draft comes in for a line edit and every sentence is close to the same length. No aside breaks the rhythm, no clause doubles back on itself, no word choice pulls slightly against the sentence around it. It reads cleanly. It also reads like it could have been written by several hundred other people submitting drafts that same week, not a coincidence of subject matter, but the predictable result of a writer spending months adjusting sentences toward whatever seems to travel best online, and adjusting them, without quite noticing, toward everyone and no one at once.
I have edited manuscripts across academic, business, and creative writing for more than a decade, first at The Expert Editor and now at Global English Editing, and this particular flattening shows up more than almost any other pattern in working drafts. Writers arrive having read enough successful posts, newsletters, and threads to have absorbed a composite voice: the breezy directness of one platform, the confiding tone of another, the punchy fragment-heavy rhythm that performs well as a hook. None of it is copied outright so much as triangulated, writing toward the average of what has already worked for other people. The result is prose optimized against a general audience instead of built for a specific one, and general audiences, it turns out, are not really audiences at all.
What audience design gets right, and where it stops working
There is a useful piece of sociolinguistics behind why this triangulation fails. In 1984, the linguist Allan Bell published “Language Style as Audience Design” in the journal Language in Society, based on his study of New Zealand radio newsreaders. Bell found that the same broadcaster, reading the same wire copy on the same day, would shift vocabulary and pronunciation depending on which station’s audience they were addressing, more careful and formal for a station with an older, broader listenership, more casual for a station with a narrower, younger one. His conclusion was that style is not a fixed property of a speaker but a response to a specific, imagined listener. Writers do the same thing instinctively: a letter to a colleague reads differently from a text to a friend, because the audience is doing real work in shaping the language, not because the writer is being inconsistent.
The internet breaks the model Bell was describing, because it removes the specific listener and replaces it with an aggregate. A writer posting to a public feed is not addressing a colleague or a friend. They are addressing an algorithmically assembled audience of strangers with nothing in common except that they might pause on the same sentence for three seconds. Adjusting style toward that audience is not audience design in Bell’s sense, it is audience guessing, aimed at a statistical composite that holds none of the preferences, context, or shared reference points that make style-shifting useful in the first place. Writers doing this are not wrong to sense that something should adjust. They are adjusting toward the wrong thing.
The fingerprint gets sanded off first
What gets lost first in this kind of writing is exactly what a reader would use to recognize the writer without a byline. In Nabokov’s Favorite Word Is Mauve, the statistician and journalist Ben Blatt ran hundreds of millions of words of published fiction through text analysis to isolate the words and habits particular authors use far more than the average writer does, what he called literary fingerprints. Nabokov, who had synesthesia, used the word “mauve” at a rate 44 times higher than its normal occurrence in English prose; other authors showed similarly outsized habits with adverb rates, sentence length, or particular verbs. Blatt’s broader point, covered in NPR’s review of the book, was that these fingerprints are consistent and identifiable enough that a program can attribute a text to its author with real accuracy, using nothing but word-level statistics.
That is the layer writers sand away when they revise toward what performs well generally. A distinct sentence length pattern, a slightly unusual word chosen out of habit, a clause that runs long because that is simply how the writer thinks, these are the features that make prose attributable to a specific person, and they are also the features that a writer chasing broad appeal learns to distrust, because they occasionally slow a reader down or read as slightly odd. Smoothing them out produces something more uniformly readable and less recognizably anyone’s.
What the research on shared tools shows about shared style
A parallel shows up in recent research on writing assistance tools, worth citing carefully because it measures something adjacent, not identical, to audience-chasing: what happens when many writers converge on the same optimization target. Vishakh Padmakumar and He He, researchers at New York University, ran a controlled study, published at the 2024 International Conference on Learning Representations, in which participants wrote argumentative essays either unassisted, with a base language model, or with a feedback-tuned model built to produce agreeable, well-received text. The results showed that writing assisted by the feedback-tuned model produced a statistically significant drop in diversity: essays became more similar to each other, and both lexical variety and content variety declined, an effect driven mainly by the model’s own contributions rather than by anything the human writers changed about their own text.
The mechanism there is not identical to a writer manually revising toward what performs online, but the shape of the outcome matches. Anything tuned toward broad approval, a model optimized on human feedback, or a writer optimized on engagement, tends to converge toward a narrower band of phrasing that reads as competent and inoffensive to the largest possible number of people. Diversity of expression is what gets traded away for that reach.
Editing toward invisibility
Editors see the consequence of this before writers usually notice it themselves. A manuscript revised many times toward general appeal starts to lose its edit-ability in a specific sense: there is less for an editor to work with, because the judgment calls that reveal a writer’s actual sensibility, the odd metaphor kept in over an obvious one, the sentence left a little long because the rhythm mattered more than the word count, have already been removed in favor of the safer, more legible option. What is left is correct and readable and interchangeable with a hundred other correct, readable drafts submitted that same month. Line edits on this kind of draft tend to be short, not because the writing is strong, but because there is nothing distinct enough left to push against.
The writers whose drafts hold up under editing are usually the ones who kept the choices that made an early reader hesitate slightly, a specific word, an odd rhythm, a claim stated more bluntly than trend pieces on the same subject would state it. Those are the same choices a writer optimizing for a general audience is trained to remove first, because they are the choices most likely to alienate someone. They are also the only reliable signal of who wrote the sentence.
The audience that was never there
The writers who eventually notice this tend to notice it the same way: the version of the piece polished to work for everyone gets published, performs adequately, and disappears within the week, indistinguishable from the ten adjacent posts that used roughly the same structure and roughly the same reassuring tone.
Nobody objected to it.
Nobody quoted it back, either. The audience it was built for, the composite, the average reader, the aggregate feed, was never a reader at all.
It was a target that, once hit, leaves nothing behind for an actual person to remember, quote, or recognize the next time the same writer publishes something else.
