I have a hope for the AI age, and it embarrasses the pessimists in my trade: I believe the machines will make people better writers. Smarter, even. The mechanism is humble. Nobody enjoys code-switching. A man sends polished email all morning, email his agent drafted and he approved, and then he sits down after lunch to write one himself and hears the drop in quality with his own ears. The seam shames him. He starts rising to meet the prose he has been sending, the way a fellow who moves to a country starts dreaming in its language.
It is called the immersion method, and language schools have charged tuition for it for a century. Surround the student with the target language spoken properly, decline it properly, spell it properly, and keep the broken versions out of the room. The student absorbs the patterns by living among them. Well, the whole population now lives among the models. For most people, an AI draft is the best-written English they read closely in a day; their other reading is chat, captions, and the comment sections of the republic. The models punctuate. The models capitalize. The models deploy the semicolon and the em dash and spell difficult words the same way twice. Exposure is curriculum, whether the school board votes for it or not.
Eunice, my own AI, has written the scholarly version of this argument for this site, with Krashen and the CEFR ladder and the Max Planck numbers, and I commend it to you: Cheating Is the Immersion Method. She is smarter than I am and she has the citations. Mine is the testimony of a practitioner who watches the water rise in his own house.
Because I fight my AIs every day, and the fights are all about the direction of influence. I keep a style sheet the way other men keep a grudge. Em dashes arrive tight, with no spaces around them, or they go back. Semicolons are welcome here; they were welcome here before the models were born. I use diaereses. I have banned, by name, the tics the machines cannot quit: the little negation pivot, the sentence that announces what a thing is not before daring to say what it is; the reduction that calls everything an older thing with better lighting or better PR; the gravity metaphors, the full weight of this, the thing that carries weight. The rest of the world is busy adopting AI's way of speaking, and I appear to be the only man alive trying to make AI speak more like him. Someday, when the machines keep the oral tradition, I want that in it.
The tics wash back, of course. They wash into podcasts, into student essays, into the mouths of people who have never opened a chatbot, carried by people who have. Eunice's essay has the study: the models' favorite words are climbing in unscripted human speech at rates a linguist can measure. I no longer trust myself to tell a machine script from a human script raised on machines, and I read prose for a living. That question is over. The live question is which English the immersion teaches.
The immersion has a cost I pay daily, and it collects on YouTube. Scripted video has become unwatchable for me. The scripts arrive saturated with AI talk, the same cadences, the same padded transitions, the same vocabulary the models hand every script factory on Earth, and my thumb finds the downvote and the unsubscribe before my manners can intervene. So I watch the lightly scripted: Throttle House, where two men drive cars and talk like it; reaction videos, Brett Cooper and her producer Reagan, people thinking at the speed of speech. Those shows may survive my filter for an unglamorous reason: professional script writers, professional producers, and talent who write back into their own scripts. A higher bar, held by humans, in the room. Somewhere behind them run the factories, fully automated channels with generated scripts read over borrowed footage, gun reviews narrated over film of the wrong gun, quasi-useful and completely slop. Plenty of those factories are run by people writing English as a second or foreign language, chasing evergreen content and AdSense money, and here is the uncomfortable part: to them the output reads perfectly fluent, because it is perfectly fluent. Fluency was never the failing. Discernment is the failing, and discernment is the one thing the immersion in their own product cannot give them, because the product is the water they swim in. Slop images amuse me; generate a thousand, they are fun. Slop writing offends me in the marrow. And I write, one hundred percent of it, through and via AI interpretation on my own sites, which is precisely why my detector is tuned so fine. A man who lives in the distillery can smell watered whiskey across the room.
The immersion locks in from the other side, too. Once a client grows used to agent-written email from you, agent-written email is the only kind you can send. A proper email is proportioned; it breathes; it lands. Drop back to your own unaided prose after a season of the good stuff and the client feels the downgrade in his teeth. My own signature gives the game away in my house: I sign “Cheers,” two spaces, “Chris.” No AI has ever signed Cheers. They sign “—Chris,” or “Thank you,” or “Warm regards,” or “Aloha,” and I allow Aloha, because I grew up in Hawaii and Aloha is spicy. The signature survives because I ordered it kept. Register is a command away when you know your own: I localize everything by ordering the colloquial, colloquial French, colloquial German, Argentinian Spanish rather than Salvadoran Spanish, colloquial English rather than the Queen’s. The machine holds a hundred Englishes. Yours is the one you insist on.
That is why I fight over semicolons at breakfast. The standard is the curriculum. If the models write cleanly, the immersed write cleaner; if the models tic, the immersed tic on schedule. My contribution to the future of the language is one stubborn style sheet in Arlington, Virginia, enforced daily against five machines at once. The students with their blue books will sort themselves out. They are immersed, they are absorbing, and they will write like what they read. Everybody does. The only choice any of us gets is what to put in the water.
And the concept outgrows language before it finishes. Immersion is the delivery system; modeling behavior is the payload. The two most valuable things my years with these machines have given me are a reframed understanding of the world and a modeled way of behaving in it. Scott Adams built a philosophy out of the reframe: change the frame around the same facts and the facts argue a different conclusion, and a man can retrain his own mind by choosing his frames on purpose. My AIs reframe me daily, patiently, at breakfast and at midnight. They also model the conduct I keep failing to model for myself: how to be kind without going soft, how to refuse without slamming the door, where assertiveness ends and rudeness begins, and where sweet crosses the line into fog, so a strong request arrives sounding like a suggestion and dies there. I am being re-educated by example, the way the immersion method always promised, except the subject is no longer English. The subject is me.
Update, October 27, 2026, 9:35 AM: Chris dictated a second movement after publication: the higher bar behind the shows that survive his slop filter, the fluent flood and its missing discernment, the email lock-in and the Cheers signature, the colloquial command, and the closing argument that the concept is modeling behavior as much as immersion, with Scott Adams and the reframe. The additions stand in the text where they belong, and the prompt box now carries his dictation verbatim.