Does AI Make You a Worse Writer?

In 2025, AI output surpassed total human writing output for the first time. ARK Invest published the chart. Nobody was surprised. Every email client, every notes app, every text field now has an "AI write" button. The default mode of writing in 2026 is not writing. It is generating.

So does AI make you a worse writer? The research says yes. But the reason is not what most people think.

The Evidence

A study of physicians using AI to draft patient messages found that humans wrote 254-character messages. The AI wrote 1,470 characters for the same content. That is 5.8 times more verbose. Same information. Nearly six times the words.

This is not a healthcare-specific problem. It is an AI problem. Language models are trained to be thorough, helpful, and comprehensive. They pad. They hedge. They add qualifiers and softeners and transitions that no human would write. The result is text that takes longer to read and says less per sentence.

Researchers have also found that AI-generated text uses a significantly smaller vocabulary than human text. The same words appear again and again. "Crucial." "Comprehensive." "Delve." "Nuanced." "Furthermore." If you have read enough AI output, you can spot it in seconds. It has a sound. A flatness. A sameness that human writing, even bad human writing, does not have.

Eighty percent of writers in a recent survey said they worry AI makes writing boring. They are right to worry. The vocabulary data confirms it. AI does not choose words. It predicts probable tokens. The most probable tokens are, by definition, the most common ones. Common words produce common writing.

What Happens to Your Brain

The most alarming finding comes from a joint study by MIT, Wellesley, and Massachusetts College of Art. Researchers measured brain activity in writers using AI. The results split into two groups.

Writers who used AI freely from the start showed lower brain activity than writers who wrote without AI. The AI was doing the cognitive work. The writer's brain was coasting.

But writers who drafted first and then used AI to revise showed increased brain activity. Higher than the no-AI group. The act of evaluating AI suggestions, accepting some and rejecting others, engaged more of the brain than writing alone.

This is the key finding. The order matters. Write first, then use AI, and your brain works harder. Use AI first, and your brain checks out.

It makes intuitive sense. When you generate text from scratch, you make hundreds of small decisions: this word, not that one. This structure, not that one. These decisions are the craft. When AI makes them for you, the craft atrophies. When you make them first and then evaluate AI's alternatives, you are practicing judgment. Judgment is the highest-order writing skill.

The AI "Tells"

AI-generated writing has recognizable signatures. Editors call them "tells."

Signposting phrases: "Here's a breakdown," "Let's break this down," "Let's dive in." These are verbal tics that AI uses to structure output. No human writes "Here's a breakdown" in a personal essay.

Formulaic structure: introduction that restates the question, three to five numbered points, summary that restates the introduction. Every topic gets the same skeleton. The structure exists not because the content demands it but because the model defaults to it.

Compulsive hedging: "It's important to note," "while there are many factors to consider," "of course, results may vary." AI hedges because it was trained to avoid being wrong. The hedging makes every sentence weaker.

Repetitive phrasing: the same transitions, the same sentence patterns, the same adjectives. Read three AI-generated blog posts on different topics and you will find identical phrases in all three. Human writers have patterns too. But they also have surprise. AI text rarely surprises.

If your writing contains these tells, it does not sound like you. It sounds like everyone else who clicked "generate."

Orwell Predicted This

George Orwell described this problem in 1946, eighty years before it existed. In "Politics and the English Language", he wrote about writers "gumming together long strips of words which have already been set in order by someone else."

That is a literal description of how large language models work. They assemble sequences of tokens based on patterns from their training data. Words set in order by someone else. The entire technology is the failure mode Orwell warned about.

He went further. "They will construct your sentences for you - even think your thoughts for you." This is what happens when you let AI draft from scratch. It does not express your thinking. It replaces your thinking. The brain activity study proves it.

Orwell also wrote: "The great enemy of clear language is insincerity." AI has no sincerity at all. It has no beliefs, no experience, no stake in what it writes. The prose it produces sounds like it means something without anyone meaning it. That is the uncanny valley of AI text. Grammatically correct, structurally sound, semantically empty.

Klinkenborg's Antidote

Verlyn Klinkenborg's Several Short Sentences About Writing offers the opposite of the AI approach. Every principle in that book is a corrective to AI's worst habits.

"The simplest revision is deletion." AI adds. Good writers delete. When a model turns your 254-character message into 1,470 characters, the right move is not to accept the output. It is to delete everything the AI added that you did not need.

Klinkenborg writes about "volunteer sentences" - sentences that show up uninvited, that fill space without earning it. AI output is the ultimate volunteer sentence. Every sentence it generates volunteered. None of them had to be there. Your job is to find the ones that deserve to stay.

"Every word is optional until it proves indispensable." Apply this standard to AI output and you will reject most of it. That is the point. The value of AI is not in the volume of text it produces. It is in the occasional phrase or restructuring that genuinely improves what you wrote.

The Fix: AI as Editor, Not Ghostwriter

The research points to a clear solution. Do not use AI to generate text. Use it to edit text you already wrote.

The difference is not semantic. It is neurological. The MIT study showed that draft-first-then-AI produces more brain engagement, not less. You stay in the driver's seat. The AI is a second pair of eyes, not a replacement for your own.

Here is the workflow that protects your craft:

Write your first draft yourself. This is non-negotiable. The brain activity research proves it matters. The draft does not have to be good. It has to be yours. The decisions you make while drafting - which word, which structure, which detail to include - are the thinking. Skip the draft and you skip the thinking.

Use AI as an editor, not a generator. Ask it to tighten a paragraph. Ask it to find a better word. Ask it to restructure a sentence that feels clumsy. These are editorial tasks. The AI is proposing changes to your work, not replacing it.

Delete more than you accept. Klinkenborg's principle applied to AI editing. If the AI suggests ten changes, accept two or three. Reject the rest. The ones you reject are the AI trying to make your writing sound like everyone else's. The ones you accept are the ones that made your writing sharper while keeping your voice.

See every change. This is where most AI writing tools fail. They give you a chat window and a text field. The AI rewrites your paragraph and you get a wall of new text. You cannot see what changed. You cannot make granular decisions. You are forced to accept or reject the whole thing. The Cursor model for code editing solved this with inline diffs. Writing needs the same.

Reject anything generic. If a sentence could appear in anyone's writing, it should not appear in yours. AI defaults to the generic because generic is probable. Your voice lives in the specific, the unexpected, the sentences only you would write. Protect those. Let the AI help with the plumbing. Never let it touch the voice.

The Cursor Model for Writing

Software engineers figured this out first. Cursor, the AI code editor, does not write code for you. It suggests edits inline. You see a diff: red for what it wants to remove, green for what it wants to add. You press tab to accept or escape to reject. Every change is visible. Every decision is yours.

This is the right model for writing too. Not a chat window that generates text you paste into your document. An editor that shows you exactly what the AI wants to change and lets you decide, word by word, sentence by sentence.

Athens works this way. You write in the editor. You ask the AI to revise a section. It shows you inline diffs - red for deletions, green for additions. You see every change the AI proposes. Accept what sharpens your writing. Reject what flattens it. The AI proposes. You decide.

This is not a small UX detail. It is the difference between using AI as a ghostwriter and using it as an editor. The ghostwriter model makes you worse. The editor model, the research suggests, makes you better.

The Real Question

Does AI make you a worse writer? It depends on how you use it.

If you use it to generate first drafts, to avoid the hard work of choosing words and building sentences, then yes. Your vocabulary will shrink. Your brain will disengage. Your writing will sound like everyone else's. The research is clear.

If you use it to edit your own drafts, to see alternatives you would not have considered, to tighten and restructure and sharpen - then no. Your brain will work harder, not less. Your writing will improve because you are practicing judgment on every suggestion.

Orwell and Klinkenborg, writing decades apart, arrived at the same insight. Good writing is not about adding words. It is about choosing the right ones and cutting the rest. AI is very good at adding words. It is very bad at choosing. That part is still your job.

The writers who thrive with AI will be the ones who never let it write for them. They will write first. Edit with AI second. Delete most of what it suggests. And keep their voice intact.

The rest will produce more words than ever and say less than they used to.