What Counts as ‘Substantial AI Assistance’ Under Most Honor Codes

The working definition of substantial AI assistance in most academic honor codes is AI that produced or shaped the ideas, structure, or wording you’re being graded on — not the mechanical help you’d get from a spell-checker. Codes rarely spell out an exact number of words or a percentage. Instead they lean on a judgment: after the AI touched your work, is the intellectual effort still recognizably yours? That’s the whole game, and once you see it, most of the gray zones get clearer.

The catch is that “substantial” is deliberately vague, which leaves students guessing and, sometimes, guessing wrong. Below is how graders actually apply the term, where the real line sits, and how to stay comfortably on the right side of it even when your policy is fuzzy.

Key takeaways

• “Substantial” usually means AI shaped the ideas, structure, or wording you’re graded on — not just grammar.
• The core test: could you reproduce and defend the work yourself, and are the words and ideas still yours?
• Grammar and light phrasing fixes are normally fine; wholesale rewriting or generated passages usually aren’t.
• Brainstorming sits in a gray zone that depends on your policy and what you did with the ideas.
• The real rule lives in your syllabus, not the campus-wide code — when unsure, ask in writing.

Why the word “substantial” exists

Honor codes use soft words like “substantial,” “unauthorized,” and “material” on purpose. AI tools change every few months, and “assistance” covers an enormous range — from a spell-checker that’s been permitted for decades to a chatbot that will write your entire essay on request. A rigid, quantified rule (“no more than 10% AI”) would be both unenforceable and outdated by the next semester. So codes stay flexible and push the specifics down to instructors.

That’s why the single most useful thing you can do is stop looking for a universal answer and read your actual syllabus. The campus-wide code sets the outer boundary — don’t submit work that isn’t yours — while the course policy sets the rule that will actually be applied to your assignment. When people say “my school allows AI” or “my school bans it,” they almost always mean a specific class.

The definition of ‘substantial AI assistance’ in academic honor codes

Strip away the legalese and integrity officers tend to apply two questions:

Could you reproduce and defend this without the AI? If you were asked to walk through your argument, explain your sources, or write a similar paragraph on the spot, could you? If the answer is no — if the tool did the thinking and you’d be lost without it — that reads as substantial, because the graded skill isn’t yours.

Are the words and ideas in the final product yours or the model’s? If your submission contains sentences the AI generated and you kept, or a thesis and structure it produced, the output isn’t clearly your work anymore. If the AI’s contribution washed out because you rethought and rewrote everything in your own terms, it’s more like advice than authorship.

Notice that neither question is about detection software. A detector flag is a probability, not a verdict, and it can be wrong in both directions. The honor-code question is about the process — what you did — not about a score. For the bigger picture on this, see our guide to what counts as misconduct in the ChatGPT era.

A rough ladder from “fine” to “over the line”

Most policies, read charitably, sort AI use into something like this:

• Almost always fine: spell-check, grammar fixes, formatting a citation, having AI quiz you on material you’re studying.
• Usually fine, sometimes disclosed: light phrasing suggestions on sentences you wrote, asking for an explanation of a concept you then apply yourself, brainstorming that you develop independently.
• Gray zone — check the policy: AI-generated outlines you build on, paragraph-level rewrites, translation of your own writing, using AI to summarize sources you then cite.
• Usually substantial / often prohibited: AI generating the thesis, argument, or whole passages you submit; producing analysis you can’t explain; writing sections you keep more or less verbatim.

The ladder isn’t a law — a class that’s specifically teaching writing may treat even light phrasing help as off-limits, while a coding course might allow generated code with disclosure. But it captures how the word “substantial” tends to bite: the higher up the ladder, the closer the AI’s output is to the finished product, and the more likely it counts.

The gray zone: brainstorming and editing

Two areas trip up honest students more than any others.

Brainstorming. Using AI to generate ten possible angles, then picking one and developing it entirely yourself, is defensible at most schools — it’s closer to talking through ideas with a classmate. It tips toward substantial when the AI’s outline becomes your outline and its framing becomes your framing. If your policy allows brainstorming, protect yourself by keeping notes that show how you turned a spark into your own argument.

Editing. “I only used it to edit” is the most common defense, and it’s true right up until it isn’t. Fixing commas and tightening a clunky sentence is editing. Having the model rewrite your paragraphs, swap in its own vocabulary, or reorganize your essay is rewriting — and rewritten text isn’t clearly yours anymore. It can also read as machine-generated afterward, which invites a detector flag on top of the integrity question. If you’re not sure whether a rewrite went too far, you can check whether a draft reads as AI, and if a section comes back flat, rework a passage in your own voice instead of leaning harder on the tool. The line between these habits is the same one covered in studying with AI versus submitting AI.

When the policy is vague, make it concrete

If your syllabus doesn’t answer the question, don’t interpret silence as permission or as a ban. Send the instructor a short, specific email: “For the argument essay, I’d like to use AI to brainstorm counterarguments and to check my grammar, but I’ll write and organize everything myself. Is that within your policy?” A described plan gets a clear yes or no, and a yes in writing is the best protection there is.

Substantial assistance isn’t a mystery once you reframe it. Ask whether the graded thinking is still yours and whether the final words came from you. Keep the answer to both a confident “yes,” and the label rarely applies. For more on drawing these lines, keep reading the blog.

Frequently asked questions

What does ‘substantial AI assistance’ actually mean in an honor code?

There’s no single legal definition, but most codes use ‘substantial’ to mean AI that produced or shaped the ideas, structure, or wording you’re being graded on — as opposed to help with mechanics you’d get from a spell-checker or a friend proofreading. If the AI wrote your argument, organized your essay, or generated passages you kept, that’s usually substantial. If it fixed commas or reminded you of a citation format, usually not. The key question graders ask is whether the intellectual work is still yours.

Is using AI to brainstorm considered substantial assistance?

It depends on the policy and on what you did with the ideas. Brainstorming that sparks your own thinking, which you then develop and write yourself, is treated more leniently at most schools than having AI produce the actual thesis and outline you submit. The risk rises the closer the AI’s output gets to the finished product. Because policies vary widely, the safe move is to check the specific course rule and, if brainstorming is allowed, keep a record of how you turned those sparks into your own work.

Does fixing my grammar with AI count as substantial assistance?

At most institutions, no — grammar, spelling, and light phrasing fixes are treated like using a spell-checker or a writing center, which are normally permitted. It becomes substantial when ‘editing’ turns into rewriting: if the AI restructures your paragraphs, replaces your wording wholesale, or changes your argument, the output is no longer clearly yours. The dividing line is whether the words and ideas are still recognizably your own after the tool touches them.

How can I tell if my AI use crosses my school’s line?

Ask two questions: could you reproduce and defend this work without the AI, and does the finished product contain the AI’s words or ideas rather than yours? If you couldn’t explain your own paper in a meeting, or if kept passages came from the model, you’re likely past the line. When a policy is vague, email the instructor with a specific description of what you plan to do and get their answer in writing. A quick confirmation beats a wrong assumption.

Why don’t honor codes just give an exact rule?

Because AI tools and their uses change faster than policy documents can, and because ‘assistance’ spans a huge range — from a spell-check to ghostwriting. A rigid rule would be outdated in a semester and would miss context that matters, like whether the assignment was meant to test your writing or your analysis. So codes use flexible words like ‘substantial’ and ‘unauthorized’ and lean on instructors to set course-specific expectations. The practical effect is that the real rule lives in your syllabus, not the campus-wide code.