Picture a teenager on a bench press, all elbows and ambition, and behind the bench a good spotter. The spotter stands with open hands, an inch under the bar, and watches. When the bar stalls halfway up and the lifter's arms start to shake, the spotter does not grab it. Two fingers go under the bar, just enough to keep it moving, and the lifter finishes the rep. Then the fingers come away and the spotter says something short. Usually it is "again."
That picture has taught me more about helping students than most professional development I have sat through. A good spotter understands something that schools, tutoring companies and now AI companies keep getting wrong. The weight is the whole point. If the spotter lifted the bar, the lifter would go home with nothing. Nothing built, nothing learned, and a quiet belief that the bar is something other people move.
Your teenager has a spotter now. It lives in their phone, it never gets tired, and left to its defaults it will lift every single bar for them.
01The spotter
Watch a good spotter for a week and you will notice they follow a small set of rules, mostly without saying them out loud.
They stand close. Help that is across the room arrives too late, so they are right there for every set. They keep their hands off the bar until it stops moving. A slow rep is not a failed rep, and grinding through it is where the strength comes from. When they do step in, they give the least help that keeps the bar moving, two fingers, never two hands. They make sure the lifter finishes the rep under their own power. And they adjust the weight over time, a little heavier as the lifter gets stronger, a little lighter before competition day.
Every one of those rules has a counterpart in the science of learning. I want to walk you through them, because together they make the clearest philosophy of help I know, and because the AI tools in your house break most of them by default.
02A spotter who lifts the bar
In 2024 a team led by Hamsa Bastani at the University of Pennsylvania ran one of the cleanest experiments we have on AI and learning. Nearly a thousand high school math students in Turkey were split into three groups. One group practiced as normal. One got a standard GPT-4 chatbot, the kind your child can open right now. The third got a version the researchers called GPT Tutor, built with guardrails so it gave hints and asked questions instead of handing over answers.
During practice, the chatbot groups looked brilliant. Students with the plain chatbot scored 48 percent higher on practice problems. Students with the tutor version scored 127 percent higher. If you had stopped the study there, you would have bought every student a subscription.
Then the researchers took the AI away and gave everyone the same exam. The students who had practiced with the plain chatbot scored 17 percent worse than the students who never had AI at all. They had spent weeks watching someone else lift the bar. The guardrailed tutor group largely avoided that damage. Same technology, same students, opposite outcome, and the only difference was whether the spotter kept its hands off the bar. The study was published in PNAS in 2025 under a title that says it plainly, "Generative AI without guardrails can harm learning."
I want to be fair to the technology, because the same research tells a hopeful story too. At Harvard, Greg Kestin and colleagues built an AI tutor for an introductory physics course that was designed around good teaching, with step-by-step scaffolding and no shortcuts. In a 2025 trial, students who used it at home learned more in less time than they did in an active-learning class. The people who built the tutor ran that study, so I hold it a little more loosely, but the direction is clear. AI can be a good spotter. It just is not one out of the box.
03Why the struggle is the point
You cannot learn to cook by watching someone else plate the food. Anyone who makes a good risotto learned it by ruining a few batches first, and the ruined batches taught them more about rice, heat and patience than any recipe video could.
Learning scientists Robert and Elizabeth Bjork have a name for this. They call it "desirable difficulty." The methods that feel hardest while you study, like quizzing yourself with the book closed, spacing your practice across several days, or mixing different kinds of problems together, are the ones that produce learning that lasts. The methods that feel easiest, like rereading your notes or watching a worked solution, produce a warm feeling of understanding that evaporates by the test.
The cruel part is that students cannot feel the difference. In a 2008 study by Nate Kornell and Robert Bjork, most learners rated the easy method as more effective even when their own results showed the opposite. A year later Kornell found that even after people saw their own scores, 72 percent still preferred the method that had worked worse for them.
Now put a chatbot in that student's hand. It removes every difficulty, desirable or not. Each answer arrives fluent and confident, and it feels exactly like understanding. That is why the Bastani students did so well in practice and so badly on the exam. The help felt like learning, and it was the opposite.
The first rule
The student attempts first, every time. No hint, no worked example, no answer until they have made a real try and said in one sentence why they chose what they chose.
04A spotter who is never there
There is a second way to fail, and it is quieter. The spotter can be so careful that they are never there when the bar stalls.
In August 2026, the economist Philip Oreopoulos and a colleague published a two-year randomized trial from 18 middle schools in Tennessee. Students used Khan Academy with Khanmigo, an AI tutor deliberately set up to coach rather than answer. It was a good tool, used in daily math classes, and the result was small, about 0.06 to 0.08 of a standard deviation a year, roughly what Khan Academy practice delivers without any AI at all.
The reason is the most important finding in the paper. Almost every student tried the AI tutor, but the median student messaged it on only a third of practice days. Worse, students asked it for help in only 17 percent of the sessions where they had actually made a mistake. The spotter was standing there with its hands out, and the lifters kept dropping the bar without calling it over.
So a good coaching philosophy has to solve both problems at once. It must refuse to lift the bar, and it must be close enough, and trusted enough, that students actually call for help when the bar stalls. That is a design problem and a relationship problem. It is the reason the best human tutoring programs meet on a fixed schedule, at least three times a week, during the school day (Nickow, Oreopoulos and Quan reviewed nearly 100 randomized trials in 2020 and found exactly that pattern). Consistency beats intensity. A spotter who shows up for every set beats a brilliant one who appears twice a month.
05Two fingers, never two hands
When my brother and I sat down to turn all of this into a working method for students and families, the hardest question was how much help to give, and when. We ended up with a ladder. Every rung adds a little more help, and a student climbs only as far as they need.
The hint ladder
| What the student gets | Why | |
|---|---|---|
| Rung 0 | What the student getsTheir own attempt, plus one sentence on why they picked it | WhyAlways required. The attempt is where the learning starts, and the sentence shows us how they think. |
| Rung 1 | What the student getsA question that points at what they missed | WhyMost stalls need a nudge, not a lift. A good question often gets the bar moving again. |
| Rung 2 | What the student getsA hint that names the rule, but not the answer | WhyThe student still does the work of applying it. |
| Rung 3 | What the student getsA worked example on a different problem of the same kind | WhyThey see the method in action, then carry it back to their own problem themselves. |
| Rung 4 | What the student getsThe full explanation, then the same kind of problem again two days later | WhySometimes the bar has to be racked. The redo proves the rep was finished later. |
The coach recommends a rung based on the attempt and on what it knows about the student. The student, and their parents, can choose a different one. That choice matters to me. A teenager who decides "give me the rule but not the answer" is practicing exactly the self-awareness we want them to grow, and a parent who knows their child is exhausted on a Thursday night can make that call too. The one rung nobody skips is the first.
There is a subtle point inside the ladder. More help is not always kinder. Researchers led by Slava Kalyuga described what they called the expertise reversal effect in 2003. Step-by-step scaffolding that helps a beginner actually slows down a student who already knows the material. A good spotter loads more weight as the lifter gets stronger. A good coach climbs down the ladder as the student climbs up the skill.
06The lifter owns the program
The best lifters I know carry a battered notebook. Every set, every weight, every failed rep, written down. The notebook is for them, so they can see their own patterns and plan the next session themselves.
Students need the same thing, and the research on this is some of the most encouraging in all of education. In 2018, Hester de Boer and colleagues pooled 48 programs that taught students to plan, monitor and evaluate their own learning. The effect on achievement was solid at the end of the programs and it grew afterward, from 0.50 to 0.63, with students from low-income families gaining the most over time. Teaching a student to run their own training outlasts almost anything else we can do for them.
So in our method every miss becomes data the student owns. They log what they got wrong in their own words, including why they picked the wrong answer. When the same kind of miss turns up three times, we help them write it down as a pattern, a one-sentence rule and the trap that caught them. Over a few months that becomes a personal playbook, written in their own voice, that they read before every test. The coach can suggest. The student decides what goes in the book.
07Parents spot the routine, not the homework
When a child is stuck, the instinct of every loving adult is to sit down next to them and walk them through the problem. It feels like the most caring thing in the world. It is also lifting the bar for them.
That instinct comes from love. But when Nancy Hill and Diana Tyson reviewed the research on parent involvement in middle school in 2009, direct help with homework had the weakest and most mixed link to achievement. The strongest link came from what they called academic socialization: talking about why learning matters, connecting schoolwork to the future your child wants, and discussing how they plan to study.
So in our method, parents see exactly the same instructions their child sees, word for word, so nobody contradicts anybody at the kitchen table. Next to those instructions sits a short note just for the parent: a question worth asking tonight, a thing to avoid saying, and one piece of logistics to handle. Protect sleep. Move the phone out of the room during study. Ask "what did you get wrong today, and what is your rule for it now?" Then step back and let them lift.
08AI used visibly, never secretly
None of this is an argument against AI. I spend my working life helping schools use it well, and I think a teenager who learns to use AI as a demanding coach will have an enormous advantage over one who learns to use it as a shortcut, or one who was simply told never to touch it. The ethics are the empowerment. Students who use AI openly and honestly get to keep everything they build.
In practice that means a handful of rules our students and families agree to. AI use is disclosed, always, to teachers and to parents. School rules and testing rules come first. Real test questions are never pasted into an AI. And the AI is told, in plain words, to behave like a spotter. Here is the prompt we give students to paste into their school's approved tool, or into ChatGPT's study mode:
A spotter prompt students can copy
I am studying [subject and topic]. Act as my coach, not my answer key. Do not give me the answer. First ask me to show my attempt and explain my thinking. Then give me the smallest hint that will get me moving, a question first, then a rule, then a worked example on a different problem only if I ask. Tell me honestly when my reasoning is wrong. At the end, give me one similar problem to try on my own.
A student who writes that prompt has done something important. They have decided, on purpose, what kind of help they want. That decision is the whole lesson.
09What to do on Monday
If you are a parent
- Ask your child to show you how they use AI for homework. Watch for who is doing the reps.
- Install the spotter rule at home. Attempt first, then ask for help, and say why you picked your answer.
- Ask about the routine, not the grade. "When are you studying this week, and how will you test yourself?"
- Praise the fight, never the talent. "You stuck with that problem" lands better and builds more than "you're so smart."
If you are a student
- Before you ask anyone or anything for help, write down your attempt and one sentence about why.
- Ask for the smallest hint that gets you moving. You can always ask for more.
- Keep a miss log. Every wrong answer, in your own words, with the rule you learned from it.
- Expect it to feel hard. That shaking-arms feeling is the learning happening.
If you are a teacher or school leader
- Teach the spotter prompt explicitly, and model it in class with a real problem.
- Grade the attempt and the reasoning, so students have a reason to show their work before they ask for help.
- Build short, low-stakes retrieval into every lesson so students practice lifting without a spotter.
- Write your AI policy around who does the thinking, rather than which tools are allowed.
Go back to that bench press a month later. There is a little more weight on the bar. It stalls at the usual spot, the spotter slides two fingers under it, and the lifter finishes the rep on their own. Nobody claps. The spotter just says "again." That is the whole philosophy, really. Stand close, keep your hands off the bar, give the least help that keeps it moving, and let them finish the rep. Then load a little more weight next week.