Harold Robert Meyer The ADD Resource Center · addrc.org
Reviewed: September 6, 2026 · Published: September 14, 2026
It is 11:40 at night and the essay is due at midnight. You paste in the prompt, read something competent eleven seconds later, change a few words so it sounds like you, and submit. It is better than what you would have written tonight. It took four minutes instead of four hours.
All of that is true, and nobody should talk you out of it. The question was never whether the machine is faster. It is what the four hours were for.
Key Takeaway
School was never selling you the paper. The reps build the person who can look at an answer – yours or a machine’s – and tell whether it is any good.
Key Findings
In a 2023 randomized trial with nearly a thousand Turkish high school students, access to a standard ChatGPT-style tutor raised scores on assisted math practice by 48 percent. A version rebuilt to give hints instead of answers raised them by 127 percent. (Bastani et al., 2025)
On the unassisted exam that followed, the standard-chatbot group scored 17 percent lower than students who never had access. The hint-only version erased that penalty – and produced no learning gain either.
Queried with the students’ own most common message, the 2023 model answered those practice problems correctly just 51 percent of the time: wrong method 42 percent of the time, wrong arithmetic 8 percent.
Students in the harmed group did not perceive that they had learned less or performed worse.
What the crutch actually looked like
Generative AI produces a passable five-paragraph essay faster than you can outline one, and on many routine assignments it produces a better one. That is arithmetic, not a moral failing, and a piece that pretends otherwise is not worth your time.
The Turkish experiment is the closest thing we have to a controlled answer, because the researchers could read what students typed. In the unrestricted arm the dominant messages were pasting the problem in or asking outright for the answer; “what is the answer?” alone accounted for 31 percent of students’ first messages in that arm. Conversations were short.
Here is the detail that gives the game away. When the chatbot made arithmetic slips – the kind any tenth grader catches instantly – those errors hurt practice scores about as much as errors in the underlying method did. Students were not reading the solutions. They were copying them.
The researchers also built a second version, loaded with the correct answer and instructed never to hand it over, only to ask what the student had tried and offer the next hint. That version produced the biggest practice gains of all and no exam penalty. The tool was not the problem. The setting was.
Two limits are worth naming, since the rest of this rests on that study. It tested mathematics, where an answer is right or wrong, not essay writing, and it ran in 2023 on a model long since superseded. What it measured was never how good the machine is. It was what happens to a student who stops doing the step.
Three objections follow. Each is better than the answer most adults will give you.
“Nobody will ever know.”
Correct. You will not be caught. Detection software is unreliable – a Stanford analysis found that leading detectors flagged the majority of essays by non-native English writers as machine-written (Liang et al., 2023) – and tools that launder machine text past detectors exist and largely work. Anyone promising otherwise is bluffing, and you should discount everything else they tell you.
That was never the risk being described here. Reread the fourth finding above: the students whose exam scores dropped reported learning as much as everyone else. Nobody deceived them. There is no sensation attached to a skill that failed to form – no ache, no gap, no moment of noticing. That is what makes this different from every other shortcut you have taken, all of which announced themselves eventually.
And the rewording is the tell. Changing a few words feels like work, which is exactly why it is reassuring: your hands are busy while nothing is being built. Prompting, rewording, running it through another tool, checking it – that is real labor, pointed at a file. You could spend the same hour on the assignment and keep what it built.
“But the A gets me into a better school.”
It does. Grades are the currency, the gate is real, and nobody is going to audit your junior-year essays four years from now. Two things follow anyway.
Schools are changing where they look. Cornell, Penn, and NYU are among a growing number now pairing written work with an oral defense, in which the student explains and defends the paper across a table (Gecker, Associated Press, March 2026). You cannot strip metadata off a conversation. When someone who knows the material asks why you chose that source instead of the obvious one, there is no file to launder. Concealment was never addressed to that question. Whether admissions offices reprice the transcript in turn is a reasonable guess, not a documented finding.
Then there is the winner’s curse. Suppose it works perfectly and you get the seat. The seat is not a prize. It is a room full of harder work, and you enter it less equipped than the classmate who did the reps and got a B. Somebody has to know which half of the output is wrong. Researchers call that capacity evaluative judgment, and it is what stays scarce once production becomes cheap (Bearman et al., 2024). You cannot supervise what you cannot evaluate, and you cannot evaluate a field you never learned.
The seat is not where the account closes, either. Not learning – or never learning how to learn – can carry a paper. It will not carry a career. The real world does not grade on a curve for effort you did not spend, and the critical thinking you skipped is exactly what the job needs once you get there. If a prompt was what got you through school, its absence is what you bring to the workplace instead of a skill – the shortcut does not just fail to help out there, it is the reason the gap exists. And it is not an edge you can trade on: everyone competing against you has the same tool in their pocket. Access to AI is not an advantage when it is also your rival’s floor. The only asset left that nobody else can copy is the judgment you built by doing the reps – which is the one thing the machine was never going to lend you.
“My parents use it, so why shouldn’t I?”
They do, and the hypocrisy is real. Pretending otherwise would cost this article your attention.
The distinction that survives is phase, not virtue. When a lawyer uses AI to draft a contract, she can tell when the draft is wrong; she is spending a skill she already owns. When you use it to draft an essay, you are declining to build one. Same tool, different transaction – a calculator in an engineer’s hand is not the same object as a calculator in the hand of someone still learning what division does. That is not a slogan. It is what the experiment measured: the tool improved performance and damaged acquisition.
So apply the test symmetrically. Plenty of adults are quietly losing skills this way and are worse about admitting it than they expect you to be. Ask your father whether he could still do the thing without it. If yes, he is using it the way the research says works. If no, he has your problem, and you are both entitled to say so out loud.
“The adult already knows what good looks like. That is the entire difference, and it is not a small one – it is the thing school is for.” – Harold Robert Meyer
If you have ADHD, the line is harder to see
Be careful with what follows, because it cuts in both directions. And note what the trial did not do: it did not study students with ADHD. None of those numbers are about you specifically.
The friction generative AI removes – getting started, facing the blank page, sequencing the middle, holding a whole argument in your head at once – is exactly where ADHD bites hardest. For a student who has spent years unable to begin, a tool that produces a first sentence is not a shortcut. It is a ramp. Refusing it on principle is not integrity; it is accepting a penalty for a disability.
But the same tool that gets you moving will also finish for you, without announcement. There is no moment where a light turns on. The workable test is direction of travel: does the tool hand you the next step, or the finished thing? Ask it for a question rather than an answer. Ask what you are missing rather than what to write. That was the difference between real harm and none.
The next step
Use it before the work, to break open a topic and find where the arguments genuinely disagree. Use it after, to attack what you wrote and locate the weak paragraph. Do not use it in the middle, where the thinking happens.
Then pick one assignment this week and do that middle yourself. Not the research, not the proofreading – the part where you decide what you think. Close the laptop afterward and explain it to someone who will interrupt with questions.
The grade may well come out the same as the last one you outsourced. Notice which one you can still explain a month from now. That is the only audit that was ever going to happen, and you are the one conducting it.
Synopsis
A student-facing argument against using generative AI as a crutch that never once relies on the threat of getting caught. It concedes what is true – the machine is faster, detection fails, concealment tools work – and relocates the risk: in a randomized trial of nearly a thousand high school students, those who practiced with an unrestricted chatbot scored far higher on practice work and 17 percent lower on the unassisted exam, and did not perceive that they had learned less. A hint-only version erased the penalty. The piece answers three student objections in the student’s own voice, adds that the skipped critical thinking does not stay behind in school and that AI access is universal rather than an edge, notes that the trial tested mathematics rather than writing and did not include students with ADHD, and closes with a use-it-before-and-after rule rather than a prohibition.
Topic: Education & School (primary) · Workplace · Advocacy, Rights & Accommodations · Self-Management Audience: Teens · Young adults & college · Parents & caregivers · Educators Keywords: crutch, evaluative judgment, oral defense, skill acquisition, concealment
Bibliography
Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. Proceedings of the National Academy of Sciences, 122(26), e2422633122. https://doi.org/10.1073/pnas.2422633122 · Free full text (PMC)
Bearman, M., Tai, J., Dawson, P., Boud, D., & Ajjawi, R. (2024). Developing evaluative judgement for a time of generative artificial intelligence. Assessment & Evaluation in Higher Education, 49(6), 893–905. https://doi.org/10.1080/02602938.2024.2335321
Gecker, J. (2026, March 25). Perfect homework, blank stares: Why colleges are turning to oral exams to combat AI. Associated Press. https://apnews.com/article/college-oral-exam-ai-chatgpt-77954a19f5304bfc6e76dc92d4bef3ad
Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. Patterns, 4(7), 100779. https://doi.org/10.1016/j.patter.2023.100779
About the Author
Harold Robert Meyer founded The ADD Resource Center in 1993 and has spent more than thirty years translating the lived experience of ADHD into practical guidance for individuals, families, and the clinicians and educators who work with them. He co-founded CHADD of New York in 1989 and went on to serve on CHADD National’s board and executive committee, including as Vice President of Finance. He was later elected to lead the Institute for the Advancement of ADHD Coaching. An author and international speaker, he has presented at the American Psychiatric Association Annual Meeting, CHADD national and local conferences, NYU Langone, Mount Sinai Medical Center, and Weill Cornell Medical College. Reach him at [email protected].
📚 Come back often — we add new evidence-based ADHD resources daily.
The ADD Resource Center [email protected] · +1 (646) 205-8080 127 West 83rd St., Unit 133, New York, NY 10024-0840 USA
X · LinkedIn · Facebook · Substack · Newsletter
Please consider the environment before printing.
Our content is educational and informational only – not medical, legal, financial, or other professional advice, nor a substitute for it, and should not be relied on as such. We strive for accuracy, but errors or omissions occur and information becomes outdated. Some content and images may be created with AI assistance and, though reviewed before publication, may contain inaccuracies. Our screening tools are not diagnostic instruments. Verify independently and consult a qualified professional before acting on anything here.
In the USA and Canada, call or text 988 anytime for free mental health and suicide prevention support, or call 911 if someone is in immediate danger. The ADD Resource Center is independent from these services and is not liable for any actions taken by you or by these services.
Under GDPR and CCPA, you may request access to, correction of, or deletion of your personal data at [email protected].
© 2026 The ADD Resource Center. All rights reserved