For students and educators

Can I use AI on this?

Four questions about the specific thing in front of you. Most people get an answer in two, and every answer — including the ones that say no — tells you what to do next.

Nothing you tap is recorded, sent anywhere, or seen by your instructor. Most people who look this up are trying to get it right rather than get away with something — and asking has never been the thing that gets anyone in trouble.

What the three lights mean

The same three you already know. Amber is the one people get wrong: at a junction it means stop and wait, not carry on carefully.

Stop

Red. Safety questions, anything timed, and the part of the task being marked. No amount of checking or disclosing makes these acceptable.

Ask first

Amber. Anything your course has not settled in writing. Treat it as red until someone with the authority to say yes says yes.

Go, with conditions

Green. Checked by you, cited to sources you have read, and disclosed. Break one of those and the light changes.

And one rule underneath all four questions: if you are hesitating about whether a use is acceptable, that hesitation is amber. Ask someone before you go.

Why the questions come in this order

Safety is checked first because it cannot be checked at all

Everywhere else, a student can use an answer and verify it afterwards. With a hazard, a quantity or a waste route, the verification is often the accident. Putting it first is also the clearest way to teach the general lesson: the tool is least trustworthy exactly where you are least equipped to audit it.

The second question is about what the assignment is for

Every task aims at one or two specific skills. If the tool performs that skill, the mark describes the tool, and the gap does not stay where you left it — mechanisms, stoichiometry and data interpretation are assumed by everything that follows, including the practical work.

The last question is the honest one

Policy and verification can both be satisfied and you can still end up with a submission you cannot talk about. “Could you explain it with no notes and no AI” is the question that separates using the tool to learn from using it to finish.

Common questions

Am I cheating if I use AI for my homework?

Often, yes. If the tool did the thing the assignment was set to assess — worked the problem, built the argument, interpreted your data — that is academic misconduct at essentially every institution, whether or not your syllabus names AI specifically. It is not misconduct if it explained a concept you then worked through yourself and your course permits that. The dividing line is not how much you used it but whether it performed the skill you are being marked on. If you cannot tell which skill that is, treat the answer as yes and ask before you submit.

My syllabus does not mention AI at all. Does that mean I can use it?

No. Silence is an unanswered question, not a permission, and misconduct is judged on what you did rather than on what you assumed was allowed. That is why “the syllabus did not mention it” is a weak position and “I asked first” is a strong one. Email your instructor and treat it as prohibited until they reply.

Can I use AI to look up a chemical hazard or how much of something to use?

No. Hazard values are measured, not derived, so you cannot check them by reasoning, and fabricated ones arrive in the right units and the right format. Worse, a missing warning is indistinguishable from a warning that does not exist. Use the safety data sheet for the exact substance and strength you have, the lab manual for quantities, and your demonstrator for anything else.

How do I tell if an AI answer about chemistry is wrong?

Rework the calculation by hand, run dimensional analysis and a mass balance, count electrons and formal charges in any mechanism, check significant figures against your least precise measurement, look up every physical value in a handbook, and resolve every DOI. If you cannot do at least one of those for a given answer, you cannot use it, because you have no way to separate correct from confidently wrong.

Does AI make up references and DOIs?

Yes, frequently. This is among the most easily detected forms of AI use in science writing, because a reference either resolves or it does not. Generated citations often pair real authors in the right field with a real journal, a plausible volume and a DOI that leads nowhere. Search for and open every source yourself, and cite only what you have read.

Do I have to tell my instructor I used AI?

Assume yes. Many courses now require it explicitly, but the obligation does not begin with the policy: handing in work shaped by a tool while letting a marker believe otherwise misrepresents whose thinking they are assessing, whatever the syllabus happens to say. A disclosure names the tool and version, what you used it for, where it appears, how you checked it, and what you changed. If declaring it would change how the work is marked, that is not a reason to leave it out — it is precisely why it has to go in. And if you are unsure whether a particular use needs declaring, declare it and say you were unsure; nobody has ever been penalised for that.

Is it different if my course is graded pass/fail?

No. The integrity standard is identical whatever the grading scheme, and the practical consequence is unchanged too: a satisfactory grade earned by a tool still leaves you without the skill in the next course and at the bench.

How do I cite ChatGPT in APA or MLA?

APA updated its guidance in September 2025. Because most AI tools now produce a shareable link for a conversation, a specific chat is cited like any other source, following the usual author–date–title–source pattern with the company as author and a bracketed descriptor such as large language model. MLA treats the tool as a container, with your prompt described in place of a title. Chicago prefers a note or an in-text mention over a bibliography entry. Whichever you use, a citation is not a disclosure and does not replace one.

Are AI detectors accurate?

No — not reliably enough to stand alone as evidence, and the vendors’ own documentation says so. Detectors measure how statistically predictable text is, not where it came from, so plain, templated or heavily edited writing scores as machine-like. A Stanford study published in 2023 found seven detectors wrongly flagged 61.3% of TOEFL essays by non-native English speakers, and Vanderbilt disabled Turnitin’s detector in 2023 after calculating that even a 1% false-positive rate across 75,000 submissions would wrongly flag around 750 students a year. None of that makes undeclared use safe or acceptable. Fabricated citations, values that match no handbook, and being unable to explain your own submission are all far more telling than a percentage, and unlike a percentage they hold up.

I have been falsely accused of using AI. What should I do?

If the work is genuinely yours, what helps is evidence of it being made: version history, dated drafts, notes, photographs of your lab book, library and browser history. Ask in writing what the allegation actually rests on — a detector percentage alone is not proof, and most institutions say so in their own policies — and take an adviser or student representative to any meeting. Do not admit to something you did not do in order to end the process; that stays on your record for a long time. If the work is not entirely yours, the honest route is genuinely the better one here: say so now rather than assembling a defence. Undeclared use is treated far more leniently than undeclared use plus a denial, and manufacturing evidence of authorship you do not have turns a misconduct case into something much more serious. This site will not help anyone make a false denial look convincing.

I already used AI when I should not have. What now?

Tell your instructor before anyone asks. Every code of conduct worth the name treats coming forward differently from being found, and it is a much shorter conversation. Do it in writing, be specific about what the tool did, and ask what they want you to do about the submission.

The whole test on one page

Every question and every possible answer, for anyone who would rather read it than tap through it — and for printing, pasting into a syllabus, or pinning up in a lab.

The questions

Question 1

What are you working on?

Pick the closest match to the task in front of you.

  • An exam, quiz, or anything timedgoes to No
  • Hazards, amounts, or waste for a labgoes to No
  • Something I’m handing in — Problem set, lab report, essay, homeworkgoes to next question
  • Studying — I’m not handing anything ingoes to next question

Question 2

Would AI be doing the part you’re marked on?

The mechanism, the calculation, your own data, your own conclusions.

Question 3

Does your course allow it, in writing?

Two places: the syllabus, and this assignment’s wording.

Question 4

Could you explain it with no notes and no AI?

Out loud, to someone asking why each step follows.

Question 5

Can you check what it says against your notes?

Not plan to check. Actually check, before believing any of it.

The nine answers

No

Not in an exam, quiz, or anything timed.

An assessment measures you under stated conditions. There is no version of this that is a grey area, and it is the easiest form of misconduct to prove.

Instead

  • Close the tool before you open the assessment.
  • If you are not ready, ask about a deferral. That route exists and this one does not.
  • If you have already used it, say so now. Coming forward is treated very differently from being found.

No

Never for hazards, amounts, or waste.

This is the one answer you cannot catch by reworking it. A pH you can recompute; a flash point you cannot derive. Invented hazard data arrives in the right units and the right format, and a missing warning looks exactly like a warning that does not exist.

Instead

  • Read the safety data sheet for the exact substance and strength you have. Sections 2, 7, 10 and 13 answer most bench questions.
  • Use the amounts in the lab manual, not a scaled version of them.
  • Ask your demonstrator or the stockroom. Then you have a source you can name afterwards.

What is fine, once you have read the safety data sheet: asking why peroxide formers get dangerous as they concentrate, or what makes an oxidiser and a reducing agent a bad pair. Concepts yes, numbers no.

No

That is the part being marked.

If the tool does this, the mark describes the tool. And it is not a gap that stays where you left it, because mechanisms, stoichiometry and data interpretation are assumed by everything that comes after them.

Instead

  • Do the first step yourself, badly if necessary, then take the broken attempt to office hours. That is far more useful to a teaching assistant than a blank page.
  • Use it backwards: ask for three more problems of this type, then solve them unaided.
  • Find the worked example in your textbook and follow it line by line with your own numbers.

If you answered “I am not sure”, that is worth sitting with. Not knowing which skill an assignment is testing is a common route into a misconduct finding, because it feels like a technicality right up until it is not.

No

Your course has already answered this.

Nothing on this page can overturn that, and a good reason for wanting to is not the same as a permission.

Instead

  • Ask for what you actually need: an extension, office hours, a worked example, a study group.
  • If you think the rule is wrong for this task, argue that with your instructor before the deadline, not with your submission.

Ask first

Silence is not permission.

Misconduct is judged on what you did, not on what you assumed was allowed. That is why “the syllabus did not mention it” is a weak position and “I asked” is a strong one. This is a one-email problem.

What to do now

  • Read the assignment wording and the syllabus properly first. The answer is often already there.
  • Send the question in writing, so that the answer is in writing too.
  • Plan the work as though the answer will be no, and treat it as no until you hear back.

Not yet

This finishes the task without learning it.

The mark arrives now and the exam arrives later. A grade earned this way is bought; the understanding behind it is only rented.

Instead

  • Go back to the point where your understanding stops, and work forward from there without the tool.
  • Explain it out loud to someone, or to an empty room. Where you stall is what to study.
  • Ask for an extension if the honest version of this needs more time. That request is normal.

Yes

With three conditions.

Break any one of these and you are somewhere else on this page. The conditions are what make this a permitted use rather than a lucky one.

All three, not two

  • Check every number, structure, sign and source yourself, before you submit.
  • Cite only sources you have opened and read. The tool is never the source.
  • Disclose it: what tool, what you used it for, and what you changed.

Things worth checking specifically, because they arrive looking correct: physical values invented outright, the wrong strength or form of a substance answered as though it were yours, mechanisms with the right product and impossible arrows, spectra assigned to peaks that are not in your trace, and DOIs that resolve to nothing.

Yes

This is studying. Use it, and check it.

The best use of it while studying is not answers at all. Ask it to quiz you, or to generate more problems of a type you keep getting wrong, then work them yourself.

Keep to these

  • Check what it tells you against your notes or textbook, every time.
  • Stop if the answer goes past what your course covers. That is a signal you have left the syllabus, not a bonus.
  • Never for hazards, amounts, or waste, even when you are only reading.

Not for answers

Use it to make questions instead.

If you cannot check it, you cannot tell a good explanation from a plausible detour. An answer you cannot place inside your course is one you cannot defend, mark, or build on.

What works

  • Ask it what you should be able to explain about the topic, then take that list to office hours.
  • Read the lecture notes and textbook first, then come back to it.
  • Ask your instructor how far this topic goes in your course. That one answer is worth more than any explanation.

Keep reading

How to cite AI Current APA, MLA and Chicago formats — and when not to cite it at all. Declaring AI use The five-line template, and what each line has to say. For educators Syllabus wording and a version carrying your own policy.