An adventure in three levels
Policy decides for the student
Policy decides for the student. That's the usual story, anyway. A syllabus clause permits AI here and forbids it there, and your part is to follow along. This book starts somewhere else, with the decisions you make about AI in your own work, one assignment at a time.
The French theorist Michel de Certeau had a useful way to put this. Institutions set strategies, and people improvise tactics inside them. A policy is a strategy. What you do with a tool late at night, draft due in the morning and a syllabus with nothing to say on the matter, is a tactic, and it belongs to you. The anthropologist Brian Street noticed something close to this about literacy. Older accounts treated it as a skill a person has or lacks. Street saw it as a practice, something people do in particular places, for particular reasons, with something at stake.
So this book asks for practice, which is a different thing from fluency. Fluency is ease with a tool. Operative literacy, the framework's name for what you'll exercise here, is judgment you're willing to stand behind, about when, how, and whether to use the tool, with your reasons attached. You'll move through three levels in order. First you describe what the tool does. Then you look at how it's built and who it leaves at a disadvantage. Last, you decide what this situation calls for: use, limits, or refusal. At the end you'll have a statement of your decision, in your own words, to keep or to share with your instructor. When a new tool or a new assignment comes along, you start again at the first level.
First time through? Read the book, about ten minutes. Its passages are numbered by level: 1 is Describe, 2 is Analyze, 3 is Decide, and 4 is your statement. Back with a new tool or a new assignment? The one-page version asks the same questions at all three levels, in about four minutes. Your answers last only while this tab stays open, so download your statement before you close it.
The one-page version
Three questions, in a fixed order. First you say what you asked the tool to do. Then you work out whether the tool's answer does the thinking your assignment exists to build. Last you decide what to do about it, with your reasons. Each question links to the part of the book which explains it. First time here? The book takes about ten minutes and tells the whole story. .
Tuesday, 11:40 p.m.
You're on the third floor of D. H. Hill, and your history essay is due at nine tomorrow morning. Two tabs are open: your draft, and a chat window you haven't typed in yet. The syllabus, which you've checked twice, says nothing about AI. Maybe the silence means go ahead. Maybe your instructor hasn't gotten to the question yet. There's no telling which, and you're far from the only one in the pack wondering.
Student decision
Whether, and how, to use the tool on your history essay.
What the systems are, and how they produce text.
Key question: Whose knowledge is encoded?
Checkpoint one: name the use
Whatever you ask for, the answer will be built from writing other people did, a great deal of it, sorted by what was most common. Keep that in mind as you type.
For example: make this paragraph sound more academic. If no sentence comes yet, choose the second answer, and the next passage will help you find one.
The tab is closed
You close the chat window and finish the history essay yourself. It takes until 1:15. That's a decision, and this book treats it as one, so it asks the same three questions it asks everyone, briefly.
Would the tool have been supplying the judgment your history essay is built to develop?
Closing the tab was the decision. The next passage asks for your reasons, since the person deciding owes them, and a refusal nobody hears about asks nothing of the person who set the assignment.
What the systems are, and how they produce text.
No sentence yet
You're in good company. Plenty of the pack hits this spot, and it's a good place to stop for a minute. It usually happens in one of three ways.
Which of these is you?
Untraced changes
You asked for something broad, help with this paragraph, make it sound better, and the tool did several things at once. A word swapped here, a sentence turned around there, and by the end it's hard to say where its hand stopped and yours began. The framework has a name for this: treating AI as one thing. You asked for one thing, help, and the tool answered with several.
Go back to the draft and ask where the thinking happened. Pick one sentence you're unsure about. What did the tool do to this sentence? Did it choose the word? The order? The claim? A use too blurry to name across a whole draft usually comes into focus at the size of one sentence, and one sentence is enough to answer the checkpoint.
No use yet
What you have is closer to a feeling than a plan, something like I need to get through this faster. Start with the slow part. Which piece of the history essay drags for you? Writing an opening line, sorting out a source, turning notes into sentences? Name the piece. Wanting speed tells you nothing about what the tool would do to your thinking. Naming the task tells you what the tool would do, and what it would cost.
Avoiding the name
Lots of people feel this, especially when the syllabus says nothing. Putting the sentence into words feels like the first step toward telling someone. Take the reluctance seriously. It's information. Often it means you're unsure you'd defend the use, and better to learn so now, before the sentence goes into the draft.
None of these counts against you. Refusal theory, the work this book draws on, treats declining to act, or declining to answer yet, as a position with standing of its own. Working out why the use won't name itself is the practice, at this scale.
When you've figured out which of these fits, go back to the checkpoint and try the sentence again. If it comes, carry on. If it still won't come, you've learned something worth bringing to your instructor. A use without a name gives you nothing yet to defend, to yourself or to anyone else.
How they are built, and whom they disadvantage.
Key question: Who controls this system?
Whose standard is the sentence meeting?
“”
Your request, in your words. Now read what came back, then reread your own version of what you asked about. Out loud, if the third floor is quiet enough.
Your request, in your words. Picture what would come back: the most common version of what you asked for, written in a confident, finished-sounding voice. Then reread your own version of what you asked about.
Does the new version still sound like you?Would the new version still sound like you?
Here's what's going on underneath. The tool learned from an enormous pile of writing, and what it hands back reflects what was most common in the pile. Generative output leans toward the most statistically likely phrasing, and the most likely phrasing comes overwhelmingly from dominant, standardized English. Lean on the tool to draft or fix a sentence, and the sentence drifts toward its standard, away from your voice, your dialect, your way of getting to a point. If English isn't the language you think in first, the drift costs you more than it costs a classmate who already writes in the tool's default.
One more thing about where your words went. The moment you pressed enter, your draft was on a company's server.The moment you press enter, your draft will be on a company's server. Do you know which one, or what the terms say about keeping it? That question, who controls the system your writing passes through, is the structural question, and most of us have never read the answer.
Look closer: what researchers found
Moon and colleagues compared 372,793 college admissions essays written before and after ChatGPT came out (2025, a preprint). The later essays had bigger vocabularies and fewer ideas. The steepest losses in ideas fell on international applicants, Black applicants, and students from homes where English isn't the main language. Liang and colleagues found the other half of the pattern (2023). AI-writing detectors flag essays by writers whose first language isn't English as machine-written. The same writers get flattened by the generators and then suspected by the detectors. One structural condition produces both.
There's a second cost, quieter than the first. The judgment behind a sentence, why this word, why this claim here and not three lines down, is usually what your history essay is there to build. Hand the sentence over, and the judgment goes with it. The tool has no ear for what sounds like you. You do, as long as your judgment stays in the sentence.
Carry one question to the next checkpoint. If the judgment behind what you asked about disappeared, would your history essay still be doing its job?
How they are built, and whom they disadvantage.
Checkpoint two: the judgment test
Your request:
Is the judgment the tool supplies, in the use you named, the same judgment your history essay is built to develop?
Three answers wait below. Yes and no are both findings, and either one takes you to the deciding level with something useful in hand. Uncertain is a fair answer too. It keeps you here a little longer, until a finding turns up.
How they are built, and whom they disadvantage.
Uncertain: stay here a little longer
Uncertain usually means the test is hard to run, and it's an honest place to be. It tends to happen in one of three ways.
You might not know what this history essay is built to develop. Courses rarely say so outright. The syllabus lists a topic, a length, a due date, a citation style, and leaves the skill underneath unnamed, whether it's building an argument, interpreting a text, or noticing what a source didn't already tell you. Without a name for the skill, the test has nothing to measure against. Look for the assignment's own account of itself: the prompt beyond the logistics, a line in the rubric, something your instructor said in class about what the paper is for.
You might know the skill and still be unsure whether the tool crossed into it. Fixing grammar and reorganizing a paragraph around a claim you didn't choose are different acts, yet from inside the sentence the line between them is hard to see. Take your one sentence and sort what was corrected from what was decided. A misplaced comma is a correction. Putting a claim here instead of there is a decision.
Or you might suspect the honest answer is yes, and uncertain feels easier to choose. Ask yourself plainly whether you already know the answer.
With the skill in mind, go back to the checkpoint and answer again. If you can't name it tonight, that's a finding of its own, and there's a posture for it: defer, and ask your instructor.
How they are built, and whom they disadvantage.
Your finding: yes
A yes is nothing to hide. It means the tool would be doing the very thinking this history essay asks of you, which is one of the most useful things to know before you decide. Name it here, so your statement carries it.
How they are built, and whom they disadvantage.
Your finding: no. Test it once.
So the function you named leaves the real work of the history essay with you. A citation formatted, a comma checked on a paper graded for its argument, a reference list put in order. Each of these changes the surface, and your claims, your order, and your reading of the sources stay where you put them.
Before you move on, give the no one quick test. A narrow name at checkpoint one makes a no easy to reach, so compare what you asked for with what came back. If you asked for a grammar check and a sentence returned making a different claim, the tool did more than you named, and the answer has changed. Head back and answer again.
Before you move on, give the no one quick test. A narrow name at checkpoint one makes a no easy to reach, so picture what would come back. Tools asked for a grammar check often return a sentence making a slightly different claim. If that's what you'd expect, the tool would do more than you named, and the answer has changed. Head back and answer again.
If the no holds up, you have a finding, and it would be the same for any student using this function on this assignment. It tells you one thing about what comes next. The strongest reason to decline, a tool doing the thinking the assignment exists to build, doesn't apply here. Plenty is still open. The syllabus is as quiet as it was. You still get to choose a limit for the tool and decide what you'll disclose.
When to use one, how, and when to refuse, with reasons tied to the task.
Key question: What does this situation warrant: use, limits, or refusal?
What will you decide, and on what grounds?
Your request: Your finding at checkpoint two: .
You've named the use and worked out what it does to the judgment your history essay is built to develop. One question is left, and it's a different one from whether you're allowed to use the tool. It asks what you'll decide tonight, and why. Your finding at checkpoint two would be the same for any student in your seat. The answer here belongs to you, and it exists only once you decide.
Most students land on one of three postures. None of them is the answer the book is hoping for. Each one holds up, as long as you have your reasons, and you'll weigh it fresh whenever the conditions change.
What do you do?
When to use one, how, and when to refuse, with reasons tied to the task.
Your decision: decline
Position taken
You've decided not to use the tool on your history essay. No rule made the call. You made it, because the judgment this assignment is built to develop is the very judgment the tool would supply in your place, whether it's a sentence-level choice, the shape of an argument, or your own reading of a text.
This decision belongs to this task. Name what you're protecting, so the decision travels with its reasons. The person deciding owes their reasons, and here the person deciding is you, so let your instructor know. A refusal nobody hears about protects nothing, and it asks nothing of the person who set the assignment.
This is refusal in the sense Simpson, Campt, and Benjamin describe: an accountable rejection of conditions which would otherwise reproduce harm, grounded in analysis instead of preference. Declining here, for this reason, on this task, is an answer you arrived at honestly.
Three questions, wherever you landed
What would reopen this decision
Your decision holds for the conditions you made it in. A draft list is waiting in the box. Change it to fit your case.
When to use one, how, and when to refuse, with reasons tied to the task.
Your decision: use with conditions
Position taken
You've decided to use the tool on your history essay, inside limits you set and state before you start. Limits written in advance are yours. Limits written afterward to match what already happened are an explanation attached after the fact.
Three things make a conditional use conditional. Name the specific function, outlining, checking grammar, generating alternatives, in terms narrow enough to mean something. Set a limit on what the tool is allowed to touch (a first pass, say, and leave the final draft to you) and hold to it. Note what you used and how, with the care you'd give a citation, so a reader knows where it came from.
The limit is what makes the choice yours. Set it before you open the tool.
Three questions, wherever you landed
What would reopen this decision
Your decision holds for the conditions you made it in. A draft list is waiting in the box. Change it to fit your case.
When to use one, how, and when to refuse, with reasons tied to the task.
Your decision: defer
Position taken
Something the decision depends on is still unknown, the policy, what the history essay is for, or what the tool would do to it. You've decided to find out before you write another sentence.
A silent policy leaves the question open, and the deadline pressing on you is no reason to answer it by default. The framework has a name for deciding by default: a decision made by the clock in place of the person. So ask what the assignment is trying to develop, and if the syllabus doesn't say, ask your instructor directly.
A deadline changes how soon you need the answer. The question still deserves one. Deferring is a choice, and it keeps a due date from making the choice for you.
Three questions, wherever you landed
What would reopen this decision
Your decision holds for the conditions you made it in. A draft list is waiting in the box. Change it to fit your case.
Your AI practice statement
Here's your decision, in one place. Add your name, course, and date if you plan to share it with your instructor, then download or print it. Nothing on this page is saved or sent.
Your copy of the book
The end
For this tool and this history essay, anyway. A new tool or a new assignment opens the book again at the first level.
How did the book work for you?
Thanks for trying the beta. These questions are about the book itself. The form sends only what you type here, none of your answers from the path, and your statement stays with you.