The Prompt is Theft
When not to prompt genAI to protect your mind
The skill many are hyping is the better prompt. The skill that matters is knowing when to close the tab.
Recently I ran a genAI’s answer to “Is college still worth it?” through CERIC and watched a confident paragraph fall apart under five questions. A reader wrote back with a fair challenge: I still prompted the machine to get that paragraph. Doesn’t that just prove we need to use the tool better?
It’s a reasonable position, and it’s everywhere. The whole AI-literacy industry is built on it—the workshops, the certifications, “prompt engineering” as a résumé line. Learn the right incantations, the pitch goes, and the tool will serve you. I won’t pretend I’m above it; I’ve been tuning prompts for years, especially as the models turned into decent research assistants. But I want to make the case for the opposite skill, the one almost nobody is teaching: knowing when not to prompt at all.
Start with what a prompt costs. Not money; most chat boxes are free. The cost is hidden in the moment before the answer arrives, that interval when you don’t yet know, when you’re holding a question open and turning it over, feeling mildly uncomfortable and slightly lost. The interval feels like a problem to solve as fast as possible. It is, in fact, where the learning lives. A mind gets sharper and more curious inside the not-knowing. The prompt is a command that ends it early.
This is why the deep-reading parallel is exact. A hard paragraph and an unanswered question are the same kind of object: both ask you to stay inside the difficulty long enough to be changed by it. The genAI summary and the prompt are the same kind of escape: both hand you a clean result while quietly soothing the reason you came. Ask the machine, and you get the conclusion without the journey that was supposed to build the muscle to reach it.
So here is the line. Real AI literacy isn’t better prompting; it’s the discernment to know whether you should be writing the prompt at all, whether this is a task worth outsourcing or a thought worth holding.
The rough rule: outsource the work that has a correct answer and asks nothing of your mind—for example, formatting a citation, organizing a list, hunting down more sources, or expanding a brainstorm you’ll curate anyway. Refuse to outsource the work whose whole value is the doing—such as evaluating an argument, finding your own position, or reading an essay that’s hard precisely because reading it changes you. When the correct, factual answer is the point, prompt away. When the thinking is the point, the prompt is theft, and you are the one being robbed.
Imagine where that theft happens. A student is assigned a page of Kant—it’s dense, recursive, and genuinely hard—and trying to read it the night before an essay is due loads the acute discomfort of not understanding quickly. The reading is a fork. One path: paste the page into the chatbot, get a clean response, feel the relief, move on, and retain nothing. It’s junk food for the mind. The other path: stay, reread, underline the verb, and argue with the sentences until they yield. It’s slow food for the mind. Both paths end the discomfort. But only the second ends it by teaching the reader to think. The machine stands at that fork offering the exit, and the exit always looks like help.
Is this just dressed-up Luddism, refusing a tool because it’s new? I don’t think so, and the test is simple: a Luddite refuses the tool everywhere. I’m asking for refusal in one place, the place where the struggle is the product. Use genAI to book the flight, debug the script, or summarize the meeting you missed. Refuse it where the wrestling was the whole point. That isn’t technophobia; it’s knowing what a tool is for, which is the most pro-technology stance there is, and the most pro-human.
So build the habit nobody taught us, because nobody had to: taking a half-minute pause before the prompt to ask one question. Am I looking something up—what stinging jellyfish live near Hawai‘i’s beaches—or am I trying to skip the deciding: should I take the job? The first is free. The second costs the exact skill no model can value.
My question is unfashionable because what I’m arguing for is slow and difficult, and slow and difficult are what the machine was built to abolish. But the people who actually understand anything five years from now won’t be the ones with the best prompts. They’ll be the ones who can still sit in the discomfort of not knowing, who can tell when the smart move is to close the tab and stay with the hard thing a little longer.
The real thinking work lies in exploring the not-knowing. Don’t let anyone sell you on the idea of pushing a button that skips it.
A question for the comments: where’s your line? What do you happily hand to genAI, and what would you never let it do for you?
Read deeply,
Dr. Genevieve