ou hand back a stack of essays and one of them stops you. It’s clean. Too clean. The transitions are smooth, the vocabulary is a notch above where this student usually writes, and something about it feels off — but you can’t say exactly what.
That uneasy feeling has a name, and it isn’t “cheating.” It’s the blurry line between AI that supports a student’s thinking and AI that does the thinking for them. Learn to see that line clearly and most of your hardest AI questions get easier. Here’s the one sentence that draws it.
Table of Contents
The line: “After the AI helps, what thinking is left for the student?”
That’s it. That’s the whole test.
Ask it about any AI-assisted task, and the answer sorts instantly into one of two buckets:
- Scaffolding supports the student while they do the thinking. The cognitive work the assignment was designed to build still happens — it’s just more accessible.
- Substitution does the thinking instead of the student. The work looks finished, but the learning it was supposed to produce never happened.
Same tool. Same student. Opposite outcomes. The difference is never the technology — it’s whether real thinking is left on the student’s side of the desk.
What the line looks like in practice
The two can be almost identical on the surface, which is exactly why teachers get caught out. Look at the pairs:
A sentence starter is a scaffold. A finished sentence is a substitution. One gives the student a way in; the other gives them something to paste.
A word bank for a struggling reader is a scaffold. A rewritten paragraph is a substitution. The first widens access to the same thinking; the second removes the thinking entirely.
Three thesis options the student has to choose between and defend is a scaffold. One thesis, handed over ready to go, is a substitution. Choosing and justifying is the intellectual work — so keeping that step protects the learning.
Feedback on a draft (“your second paragraph makes a claim with no evidence”) is a scaffold. A rewritten draft is a substitution. The student still has to decide what to do with the feedback.
Notice the pattern: scaffolding hands the student a tool and makes them use it. Substitution hands them a result.
The trap: substitution usually looks like help
Here’s why this is hard. Substitution is faster, it’s cleaner, and the student walks away happy — with a better-looking product and a worse understanding. Nobody feels like they cheated. The AI was “just helping.” And the cost doesn’t show up until the test, the next assignment, or the moment the student has to do it without the tool.
That’s the quiet danger. Substitution doesn’t announce itself. It disguises itself as support. Your job — the part no tool can do — is to design AI use so the thinking stays with the student, even when it would be faster not to.
How to redesign almost any AI task toward scaffolding
The good news: you rarely have to ban a tool. You just have to add a constraint that puts the thinking back. Most of these live in the Role part of a prompt — what you tell the AI to do, and what to refuse to do:
- “Ask me three questions before you give an answer.” Turns the AI from an answer-machine into a guide.
- “Give feedback only. Do not rewrite. Quote no more than ten words.” Keeps revision in the student’s hands.
- “Give me a hint and the next step — not the final answer.” The single most useful move for homework help.
- “Provide sentence starters only, never full sentences.” Access without ghostwriting.
- “Use only the sources I provide. If it’s not there, ask me.” Keeps the work evidence-based.
Each one is small. Together they’re the difference between a tool that builds thinkers and one that quietly replaces thinking.
It shifts by grade — but the principle doesn’t
What counts as a scaffold changes with age. Upper-elementary students are learning to ask for help without outsourcing — “explain it in three steps,” not “write it for me.” Middle schoolers can add context and request hints instead of answers. High schoolers can prompt for rubric-aligned critique without letting the tool write the paper. The guardrails move; the line — does the student still do the thinking? — stays exactly where it is.
Make the thinking visible
One last move makes this whole thing manageable without turning you into a detective: collect a little proof of process alongside the final work. A three-line note on what AI was used for. An outline. One revision the student can point to and explain. When a student can show you where they changed their mind, you don’t need forensic software — the thinking is right there on the page.
That’s the shift underneath all of this. You’re not policing whether AI was used. You’re designing so that when it’s used, the student is still the one who learned.
The one question, one more time
Before any AI-assisted task reaches your students, ask: after the AI helps, what thinking is left for them to do? If the honest answer is “not much,” redesign it. If the thinking is still theirs, you’re using AI exactly the way it should be used.
New here? Start with the full guide: How to Use AI in the Classroom Without Replacing Student Thinking.
Free: The Teacher AI Emergency Kit
Want prompts that are already built to scaffold instead of substitute? Grab The Teacher AI Emergency Kit — 25 copy-paste prompts, free.
Go deeper
- The Human–AI–Human Classroom — the full framework for building a classroom where the teacher leads, the student thinks, and AI earns its place. ($14.99)
Use the tool where it helps. Keep the thinking where it belongs.
— Blake W. Ellis
