Will AI Replace Writers? What the Research Actually Says

In 2025, AI-generated text surpassed total human writing output for the first time. ARK Invest published the chart. Every writing tool added a "generate" button. Every email client added a "draft for me" feature. The volume of machine-written text now exceeds everything humans have ever written.

So will AI replace writers? The honest answer is: it depends on what kind of writer you are.

What the Research Shows

Three findings matter here. Each points in the same direction.

Finding one: AI users show lower brain activity. A joint study by MIT, Wellesley, and Massachusetts College of Art measured brain activity in writers using AI. Writers who used AI to generate text from scratch showed lower brain activity than writers who worked without AI. Their brains were coasting. The machine was doing the cognitive work.

But writers who drafted first and then used AI to revise showed increased brain activity. Higher than the no-AI group. Evaluating AI suggestions, deciding what to keep and what to reject, engaged more of the brain than writing alone. The order matters. Write first, then use AI, and your brain works harder. Use AI first, and your brain checks out. (I wrote more about this in Does AI Make You a Worse Writer? )

Finding two: AI can evaluate but not create at human level. Tuhin Chakrabarty and colleagues at Stony Brook University tested this directly. They built a 14-criteria rubric for evaluating New Yorker-style short stories and compared LLM-generated stories against human writers. The AI stories scored worse across every dimension. Technically competent. Structurally acceptable. But flat. No surprise. No risk. No voice.

Here is the interesting part. When the same models were used to evaluate stories instead of write them, they performed well. AI is a better critic than creator. It can tell you what is wrong with a paragraph. It cannot write one that makes you feel something. (More on this in Why AI is a Better Editor Than Writer )

Finding three: even the people building AI admit the limits. Sam Altman, CEO of OpenAI, said in an October 2025 interview with Tyler Cowen that even GPT-7 might only produce "a real poet's okay poem." Not great. Not good. Okay. Patrick Collison and Tyler Cowen's "New Aesthetics" grant program went looking for great AI-created art. Their conclusion: "we haven't seen much great work that only uses AI."

Volume has exploded. Quality has not kept up.

What AI Will Replace

AI will replace writing that was already replaceable. Template content. Generic marketing copy. Boilerplate documentation. SEO filler articles that exist only to rank on Google. Product descriptions generated from a spec sheet. Emails that follow a formula.

This writing never required original thinking. It required following patterns. AI follows patterns better than humans and does it faster. If your job is to produce text that could have been written by anyone, AI will produce it cheaper.

This is not a prediction. It is already happening. Content mills are replacing writers with AI generators. Marketing teams are using AI for first drafts of landing pages. Customer support teams are using AI for template responses. The writing that existed because it was cheap to produce will be replaced by writing that is cheaper to produce.

What AI Cannot Replace

Voice. Lived experience. Specific observation. Original thinking. The things that make writing worth reading.

Verlyn Klinkenborg, in Several Short Sentences About Writing, argues that the core skill of writing is noticing. Not generating sentences. Noticing. Seeing what is actually there, not what you expected to see. The sentence comes after the noticing. And the noticing is the hard part.

AI cannot notice. It has no experience to draw from. It has no body that has walked through a particular place at a particular time. It cannot tell you what the light looked like at 6am in a specific kitchen on a specific morning. It can generate a description that sounds like noticing. But it is pastiche. Patterns assembled from other people's noticing.

Joan Didion wrote: "I write entirely to find out what I'm thinking." This is writing as discovery. You sit down not knowing your argument and you build it sentence by sentence. Each sentence changes the next one. The final piece says something you did not know before you started writing. AI cannot do this. It does not discover. It predicts.

Orwell understood this decades ago. In "Politics and the English Language," he showed how vague language lets writers avoid committing to a specific claim. You can write a paragraph that sounds like it says something without actually saying anything. AI generates exactly this kind of writing. Fluent, confident, and empty. (I explored Orwell's rules applied to AI in Orwell's 6 Rules for AI Writing )

Here is a simple test. Can the writing only have been written by this specific person? Does it contain observations, arguments, or phrasings that could not have come from anyone else? If yes, AI cannot replace it. If no, AI probably will.

The Real Shift: From Generator to Editor

AI does not eliminate the need for writers. It changes what writers do. The role shifts from generator to editor. From producing words to choosing them.

This is not a demotion. Editing is harder than generating. It requires taste. Knowing what to keep. Knowing what to cut. Knowing when a sentence is doing real work and when it is filling space. These are judgments AI cannot make reliably because they depend on intent, context, and audience in ways that statistical prediction cannot capture.

The MIT brain study confirms this. Evaluating and editing AI output engages more cognitive effort than writing from scratch. The writer who uses AI well is working harder, not less. The skill is different. But it is still skill.

Think about what happened to photography. The camera did not kill painters. It killed portrait painters who offered nothing beyond accurate likeness. Painters who brought vision, interpretation, and style became more valued, not less. The camera freed painting from the obligation to reproduce reality. AI will free writing from the obligation to produce volume.

The Taste Gap

The writer's most valuable skill in 2026 is taste. Not the ability to generate sentences. The ability to judge them.

Taste means reading an AI-suggested paragraph and knowing it is wrong before you can articulate why. It means cutting a sentence that is technically correct but tonally off. It means choosing the unexpected word over the predicted one. It means recognizing when AI output sounds like AI output, even when it is grammatically flawless.

You develop taste by reading widely and writing regularly. There is no shortcut. AI cannot give you taste. It can only give you more material to exercise taste on.

This is why the question "will AI replace writers" misses the point. The real question is: will you develop the judgment to use AI without losing your voice? The writers who answer yes will thrive. The writers who hand over their thinking to AI will produce text that sounds like everyone else's text. And that text is, by definition, replaceable.

What Bad Writing Reveals

There is a deeper problem. Bad writing is a sign of bad thinking. When you let AI generate your writing, you skip the thinking step. The words appear on the page without passing through your mind first. They look like ideas. They are not ideas. They are statistically probable arrangements of tokens.

This matters beyond writing. Clear writing forces clear thinking. The discipline of choosing the right word, building the right sentence, constructing the right argument sharpens your mind in ways that delegating to AI does not. If you stop writing, you stop thinking clearly. Not immediately. Gradually. The way a muscle atrophies when you stop using it.

The executives who write their own memos think more clearly than the ones who have AI draft them. The students who write their own essays understand the material better than the ones who generate them. The researchers who write their own papers develop deeper insight than the ones who outsource the prose. Writing is not a communication tool. It is a thinking tool.

A Practical Framework

So how should writers respond? Three principles.

First, always write the first draft yourself. The MIT brain study is clear on this. If AI generates and you edit, your brain disengages. If you generate and AI helps edit, your brain engages more. The first draft is where thinking happens. Do not outsource it.

Second, use AI as an editor, not a writer. Let AI suggest cuts, flag weak sentences, identify structural problems. Then decide yourself what to accept. The decision is the skill. A tool like Athens shows you diffs of what AI changed so you can evaluate each suggestion individually. This is the right workflow. You see exactly what the AI proposes. You accept or reject each change. You stay in control.

Third, develop your voice deliberately. Read authors whose style you admire. Write regularly, even when you do not have to. Pay attention to the specific details of your experience. The things AI cannot know because it has not lived your life. Your voice is your competitive advantage. It is the one thing that cannot be replicated by a model trained on everyone else's words.

The Bottom Line

AI will replace writers who only generate. It will not replace writers who think. The volume of AI-generated text will keep growing. Most of it will be mediocre. The demand for writing that is specific, surprising, and true will not decrease. It will increase. Because when everything sounds the same, the writing that sounds different becomes more valuable.

The question is not whether AI will replace writers. The question is whether you will remain a writer who thinks or become a writer who delegates thinking to a machine. The research is clear about which path leads where.