Stephens writes in The New York Times:
Don’t use artificial intelligence to help you write. Never let A.I. do your writing for you.
Don’t use it for school papers, work briefs, letters to your in-laws, speeches at your company gathering or emails (however perfunctory) to your colleagues or friends. Don’t let it organize your notes. Don’t let it suggest an opening sentence, a segue or a closing paragraph. Don’t ask it to write a first draft and pretend that editing that draft somehow makes it your own. It doesn’t.
Do none of these things not because they are unethical. Writing with A.I. is unethical when it’s a deception: when you pass off words, ideas and information as your own when they aren’t. An acknowledgment can largely address the problem. Do none of these things, either, because you might be able to learn to write better than an A.I. can. Pretty soon, if not already, you won’t, just as you can’t outrun a car or outplay a chess app.
The problem with writing with A.I. is that it’s mentally enfeebling — an escalator toward a result when you really need to make a daily habit of taking the stairs. As it becomes ubiquitous, it undermines not only our individual ability to write but also a society’s collective ability to reason, a culture’s inner capacity to create and everyone’s reason to care. We’re already reckoning with the well-documented decline of reading; A.I. is accelerating the decline of writing, ushering us further into what The Atlantic’s Rose Horowitch calls our “postliterate age.”
What, uniquely, does writing do? It compels thought. It compels thinking in ways that silent contemplation or spoken language rarely can. It compels us to subject our thinking to the effort of articulation, the rigor of grammar, the tests of intelligibility and coherence, the inspection of others. In doing so, it also enjoins us to be clear, logical, accurate — and accountable. People may easily forgive a word said in anger but not so easily one written in anger, precisely because the fact that it was written tells us that it was considered.
This column might have been better if he had used AI as an editor.
The column has one checkable number in it.
The study is Dan Sarofian-Butin’s exploratory analysis, and what he examined was 100 EdD dissertations published in 2025 in Educational Leadership and Administration. The column renders this as “100 doctoral dissertations,” which invites the reader to picture doctoral education. The EdD is a practitioner degree. Sarofian-Butin’s own suggestive finding runs against the generalization: lower AI use correlated with R1 or R2 institutions, higher AI use with private non-profits. The column takes the corner of the doctoral world where the effect is largest, drops the qualifier, and presents the result as a fact about doctoral study. A researcher would call that sampling on the dependent variable. An AI asked to verify the citation returns the abstract in seconds and the writer sees the gap himself.
The reliability question sits underneath. Detection tools disagree with each other. One controlled comparison found intraclass correlations ranging from 0.57 to 0.95 across three open-access detectors, which raised concerns about the reliability of the tools. The column’s entire empirical foundation is a single exploratory paper using contested instruments. It bears the weight of the paragraph about the death of academic integrity and the paragraph about democratic self-governance after that.
Second, the strongest objection to the argument goes unmentioned. Socrates makes this case against writing in the Phaedrus. The new technology will supply the result and the faculty will wither, memory in his version, thought in this one. The structure is identical. Anyone who has read Plato hears the echo in the first paragraph, and the column has to explain why the parallel fails. Maybe it does fail. Writing externalized memory and produced philosophy, so the trade was good, and perhaps this trade is not. That argument is available and the column does not make it. An AI prompted with “what is the best case against this piece” produces the Phaedrus, the calculator, GPS and spatial memory, and the transactive memory research of Sparrow, Liu, and Wegner on how people offload to search engines. The writer then has four objections to answer and a stronger column.
Third, there is an empirical claim that the column asserts and never defends: “We become better writers by the constant effort that mundane writing demands.” Anders Ericsson (1947-2020) spent a career arguing the opposite. Repetition without feedback produces a plateau. Most office memo writing is repetition without feedback. Nobody grades your calendar-invite prose. If the mundane writing were building the muscle, the average corporate email would be better than it is. The column needs the claim to be true because the wedding toast and the routine memo have to be the same activity for the argument to hold.
Fourth, the piece bans note organization and segue suggestion and never draws a line. Spellcheck, outlining software, a thesaurus, a research assistant, and an editor all supply what the writer did not generate. Newspaper columnists work with editors who rewrite ledes and cut closing paragraphs, and nobody says the desk enfeebled them. So the rule cannot be that assistance corrupts. It has to be something narrower about generation, and the column never says what. Asked “state your rule as a test a reader could apply,” an AI exposes that there is no rule yet, only a mood.
Fifth, the ethics paragraph tangles. The first sentence says do none of these things and not because they are unethical. The second says AI writing is unethical when it deceives. The third says acknowledgment mostly fixes that. So the ethical objection is raised, conceded, and resolved in three sentences, and the reader is left unsure why the paragraph exists. It exists to clear the ground for the enfeeblement argument, which is the real one. Cut it to a clause and start the piece a paragraph earlier.
Sixth, the ending. If the thesis is that everyone’s capacity to reason is eroding, the last line should not sort the audience by party. “Or a vote against Trump” converts a claim about cognition into a coalition signal and tells half the readership that the argument was never addressed to them. The writer may want that. But he should want it knowing the cost, and a reader who asks “who does this sentence lose” makes the cost visible.
What evidence would change his mind? If a study found that students who drafted with AI and then revised produced better arguments than students who drafted alone, would the thesis survive?
So the improvements are all upstream of the prose: check the number, find the counterargument, defend the causal claim, state the rule, cut the tangled paragraph, count the cost of the ending. Which is the irony worth sitting with. The most useful thing AI does for a writer is adversarial rather than generative. It is a fast, tireless, unembarrassed reader who says your best evidence is thinner than you think and here is the objection you skipped. The writer still has to decide whether the objection lands. That decision is the thinking the column wants to protect, and nothing in the process removes it.
Robert Wright (b. 1957) and Paul Bloom (b. 1963) spend the free hour on the Bret Stephens (b. 1973) column against writing with AI and the last third on frontier model containment. The two halves belong to different shows.
At 5:10 Bloom gives the professor’s position. He has students who cheated, he cannot describe the cases, and he says he has to change his courses so they no longer have take-home essays or take-home reading responses. He calls this a minor inconvenience for him and a real loss for the students.
At 6:33 Wright puts the strongest challenge on the table and does not press it: “If we’re moving into an age where what you need to be able to do to flourish economically is use AI, and in fact the people who delegate the most cognitive tasks to AI may do the best.” Is the college obliged to build the skills the market will not pay for? Bloom answers that it need not be either-or, that students should graduate able to write and able to use the tools, and that the trouble is the tool’s second function as a cheating device.
At 8:52 comes the best sociology in the episode. Bloom sympathizes with the cheaters because grading is zero-sum: “It takes a lot of character to say I’m going to spend five, ten hours producing something that will be nowhere near as good as what my fellow students produce.”
At 9:38 they read the tangled Stephens paragraph aloud, the one with the stacked negations. Bloom’s verdict: “These are awful sentences. Claude did not write these sentences.”
At 12:06 Bloom lands the argument the column never anticipates. “If AI makes things better for the reader, your refusal to use AI is a choice to privilege your own needs over those of the reader.” He extends it to fact-checking. Decline the tool and your work carries more errors, and you get to feel holier.
That’s the argument I made Aug. 1.
A writer who puts the reader first will have less of a problem using a machine’s sentence rather than his own if that better serves the reader…
So when you read something on my blog since May 2025 that sits in the median, that’s likely AI-influenced. And when you read something off the median, that’s me.
The reader-first frame is not something I generated on August 1. I ran it across four newspapers on June 1: the Times, the Post, the Financial Times, the Los Angeles Times. I’ve used the frame for decades (what would a newspaper look like if it put the reader first and why I find it strange that nobody does this) and applied it to the AI question when the AI question came up. Bloom got to the same place through his fact-checking practice (20:42).
Bloom stops at the benefit. I price it: a model trained on the median of published English pulls any writer toward the median, and the writer’s value comes from where he sits off it. Neither Bloom now Wright on the podcast makes this point. Bloom gestures at it when he says AI prose is smooth and corporate and good but not very good, and he treats it as an aesthetic complaint rather than an economic one. I treat it as the price of the trade.
I also have a rule. I tell the reader how to sort my prose, median is likely the machine, off-median is likely me. It’s rough, it’s unverifiable, and it is still more of a disclosure protocol than Stephens offers or the podcast reaches. Wright says he’ll disclose when he starts. Bloom doesn’t consider his fact-checking disclosable. Neither says where the line falls. I did.
At 13:05 the Air Canada story. A cancelled flight, a hundred-dollar coupon, a chatbot telling him he was owed more and then drafting the demand letter. “I do not need to exercise my muscle for legalese.”
At 16:36 Wright describes what he uses Claude for. Subtle questions of usage, a sophisticated thesaurus, a fine point of semantics. He connects it to the era when newspapers employed brilliant copy editors, many of them women whose other career paths were closed. “I’ve never encountered a human version of this that was better than Claude at this.” Then: “I feel a kind of emotional connection with Claude when it’s doing that. It’s weird.”
At 19:24 Bloom draws his line. “For my writing it’s all me. I take pride in my writing.” He offloads administrative prose, uses AI heavily for research questions, and then hands finished drafts to Fable and asks what the arguments miss and what he should be reading. The fact-checking catches errors no human editor would catch, he says, including a place where a secondary source misrendered a primary one. “Less of what I write will be false.”
At 22:09 the analogy run. GPS and map reading, memorized poetry, the slide rule Wright learned as a sophomore before calculators arrived. Bloom voices the dismissal: “It’s nostalgia.” Wright pushes back a little, noting that doing arithmetic in your head might carry over to other analytical work, then lets it go.
At 23:42 Wright says that writing an argument down exposes what you have not thought through. “Putting it on the page forces you to confront it as if it was a different person.”
At 25:33 Wright explains that his edge with op-ed editors used to be that he writes better than think tank fellows. Now, he says, a bunch of people at think tanks are having Claude go over their pieces, “and so my stuff reads much more like the stuff from everyone else.” Bloom, kindly: your comparative advantage used to be that you were the better writer, and now the gap does not help you.
At 29:02 Bloom names the tells. AI submissions arrive error-free, smooth, easy to digest, and something close to LinkedIn. “It’s not X, it’s Y, and everything’s a list of threes. It’s good, but it’s not very good.”
From 30:04 to 41:00 they move to the July containment failures and Wright’s argument that autonomy is what the market demands and that the race between labs and between countries rewards recklessness.
Bloom’s reader-standpoint objection at 12:06 is their contribution, and neither man builds on it. Stephens argues from the writer’s soul. Bloom answers from the reader’s interest, and the two are not commensurable. The reply available to Stephens is that a byline is a warranty. The reader wants accurate, readable prose, and he also wants to know whose judgment stands behind the sentences, because that judgment is what he is deciding whether to trust next month. Smoothness he can get anywhere now. Warranted judgment is the scarce good. That answer sits there unused for the rest of the hour.
The second thing they concede and then walk past is the asymmetry between the professional and the student. Both men have already built the faculty. Bloom writes his own prose and hands the machine the search committee summaries. Wright asks about adjectives. For them the tool is a supplement to a skill that exists. For the nineteen-year-old it substitutes for building one. Bloom half-sees this in the cheating discussion and never joins it to the general argument, so the episode ends up saying that Stephens goes too far without saying for whom.
The analogy run at 22:09 is weaker than it plays. A calculator replaces a procedure you have already learned to specify. GPS replaces route memory. Neither touches the formulation of the problem. Writing is where the problem gets formulated, which is Wright’s own point at 23:42, made ten minutes later without either man noticing that it kills the analogy. If writing is how you discover what you have not thought through, then offloading it is not the slide rule going away. Wright brushes the edge of this when he wonders whether mental arithmetic carries over, and drops it.
At 25:33, Wright says his advantage over credentialed experts was prose quality and that the advantage is gone, and he says it about himself rather than dressing it as principle. That explains a great deal of the anti-AI writing appearing in prestige outlets, including some of the Stephens column. A profession whose rent came from a scarce skill is watching the skill become cheap. The argument may still be right. The interest behind it should be visible.
Bryan A. Garner writes in National Review Aug. 20, 2026:
Next spring, the University of Chicago Press will publish my book How to Write Well: 50 Pointers on Process, Usage, and Style. It’s the product of a lifetime spent studying, teaching, and practicing the craft. One lesson may surprise some readers: If you’re going to use AI, use it well. Write careful prompts. Keep editorial control. Above all, don’t let the machine do your thinking….
But Stephens mistakes the danger. The problem isn’t AI. It’s intellectual surrender. There’s a world of difference between using AI as an editorial assistant and using it as a substitute for judgment. Stephens largely ignores that important distinction…
Writing isn’t one activity. It’s many. It involves discovering ideas, organizing them, testing them, arranging them, revising them, and polishing them. AI isn’t equally useful at every stage. It can generate ideas, but its greatest value often comes from interrogating them. Ask it to challenge your reasoning, tighten your organization, identify your weakest paragraph, or argue the opposite side, and it can become an inexhaustible editor.
That’s why the best writers often get the most from AI. Good prompting isn’t easy. It requires precision, organization, audience awareness, and analytical judgment. A vague prompt produces a vague response. A thoughtful prompt reflects thoughtful writing before the AI generates a single word. The quality of the output usually reflects the quality of the thinking behind the request.
Nor is revision somehow less intellectual than drafting. Experienced writers know that the real thinking often happens after the first draft. We discover weak logic, hidden assumptions, awkward transitions, and better ways of expressing an idea. AI can accelerate the process by generating alternatives that invite comparison. You must still decide which version is accurate, persuasive, and true to your voice. Faster revision doesn’t mean less thinking. It means more opportunities to think — and more revisions, which lead to a better product in the end.
Consider the novice drafting a memo, a brief, an email, or an op-ed. The hardest part usually isn’t having an idea. It’s expressing that idea clearly. AI can offer a rough structure, suggest several organizational approaches, or expose weaknesses before anyone else reads the piece. The writer still bears responsibility for every word. In that role, AI resembles a good editor asking uncomfortable questions: What exactly do you mean? What’s your evidence? Why this order? Why this tone?
Stephens’s strongest objection is really about apprenticeship. If students let AI write everything, they’ll never develop the habits that good writing demands. That’s true. Beginners shouldn’t outsource first drafts any more than beginner pianists should let software practice their scales. But that’s an argument for disciplined use, not blanket prohibition. We don’t teach arithmetic by pretending that calculators don’t exist. We teach the fundamentals first and then show students how to use powerful tools intelligently.
History points in the same direction. Nearly every important intellectual technology has inspired predictions of decline. Dictionaries, word processors, online research, and now AI have all been accused of making us lazy. Yet none of those tools diminished the value of careful writers. They increased it. They rewarded the people who knew how to use them well.
Not long ago I asked a Pulitzer Prize–winning journalist whether he’d encountered accomplished writers who remained fiercely resistant to AI. “Yes,” he said. “They remind me of the writers in the 1990s who prided themselves on not using the internet.” That comparison captures a recurring pattern. Every generation confuses familiarity with virtue. Eventually the tool becomes ordinary, and the real distinction — between those who use it well and those who use it poorly — is restored.
The legal profession is already living through the transition. Clients increasingly refuse to pay for work that competent software can perform in seconds. Some firms now expect young lawyers to devote a meaningful share of their billable work to AI-assisted research, drafting, and analysis — not because AI is replacing legal judgment but because intelligent lawyers using AI can often deliver better work more efficiently. The profession won’t be divided between humans and machines. It’ll be divided between lawyers who know how to use AI intelligently and lawyers who don’t…
One of AI’s greatest virtues is immediate feedback. It lets writers test a claim, experiment with tone, ask for skeptical objections, or compare several ways of organizing an argument. Used thoughtfully, the process exposes blind spots quickly. Of course AI also hallucinates, flatters, and generates oceans of lifeless prose. That’s exactly why human judgment remains indispensable. AI can produce possibilities. Only the writer can recognize which possibilities deserve to survive…
Good writers have never worked alone. They’ve relied on editors, colleagues, dictionaries, style manuals, teachers, and trusted readers. These didn’t diminish authorship but refined it. AI belongs in that tradition. It’s another instrument for criticism, revision, and improvement…
Writers who hand their thinking to a machine will almost certainly become weaker. Writers who use AI to challenge their assumptions, strengthen their prose, and sharpen their judgment will become stronger than they would have been without it. The decisive variable isn’t the software. It’s the mind directing it. Good writing has always depended on disciplined thinking. And AI has made disciplined thinking more important and valuable than ever.
The prestige tier finds the formulation that permits the switch without loss of face, and Garner has found it: the writer as editor, in command, using the machine as an instrument of judgment. Stephens holds the line. Garner moves it and keeps the guild.
Garner breaks writing into stages. Discovering, organizing, testing, arranging, revising, polishing. That single move dissolves the equivocation that ran through the Wright and Bloom hour, where an undifferentiated verb let both men agree with Stephens and disagree with him in the same breath. And he identifies interrogation as the highest-value use, which matches what Bloom described at 20:24 and what I described on August 1. Three men in three formats converging on the same finding, which is that the machine is more useful as an adversary.
The central empirical claim is that the best writers get the most from AI. That is the claim the whole essay rests on, and it is the one claim in the piece with directly relevant evidence, and the evidence runs against him. Noy and Zhang, published in Science in 2023, ran mid-level professionals on writing tasks and found the gains concentrated among the weakest writers, with the spread between top and bottom compressing. Brynjolfsson, Li and Raymond found the same distribution in customer support, largest effects for the least experienced. Dell’Acqua and colleagues at BCG found the same shape in consulting. The pattern across the literature is a raised floor and a lowered ceiling. Garner asserts the opposite with no citation.
Garner supplies a test nobody can fail: remain in command, keep editorial control, don’t let the machine do your thinking. Every writer who has ever pasted a generated paragraph believes he remained in command. The criterion is interior, unobservable in the artifact, and self-certifying. Which is precisely the structure of the claim he is criticizing. Stephens says the value lies in unauditable labor by the writer. Garner says it lies in unauditable judgment by the writer. Same unfalsifiable location, opposite conclusion. He beat Stephens on the taxonomy and then reproduced his method.
Clients refuse to pay for work software does in seconds. That is substitution and price collapse at the bottom of the skill range, which is the leveling story.
No reader appears anywhere in the essay. The entire case runs on writer development, exactly as Stephens does, which means Bloom’s best objection has still not been answered by anyone in print. No disclosure question. No pricing of the cost, no acknowledgment that a model trained on the median pulls prose toward the median, no version of my line about where the writer’s value sits. The essay has no defeat conditions. Ask what would change Garner’s mind and there is no answer in the text.
The man who wrote Garner’s Modern English Usage has produced a column dense in the construction Bloom named at 29:02 as the machine’s signature. The problem isn’t AI, it’s intellectual surrender. It isn’t. The problem isn’t that a tool can be misused. The real question isn’t whether, it’s whether. I count nine in a short piece. I am not making a charge, and I would not make one, since the construction predates the machine and Garner has always written in short declaratives. It is observable, and any reader who has spent an hour with a chatbot will observe it, and that is a hazard for anyone publishing on this subject now.
This is the guild’s second line of defense, and it is a better one. Rather than sacralizing the skill that is depreciating, redefine the scarce asset as the judgment that directs the tool. That may even be true. It is also the belief most convenient for a man with a book on writing well coming from Chicago next spring, and the belief he would have to hold whatever the evidence said.