AI Output Disclaimer

Effective date: 4 September 2026 Last revised: 4 September 2026

This is part of the Terms of Service and expands Section 7 of it. Read it with the Academic Integrity Policy, which covers the part your school cares about.

In one sentence

Everything CourseLite generates is a draft written by a language model, it is often wrong, and checking it is your job — not ours and not the model's.

1. What "generated" means here

CourseLite's answers, flashcards, practice tests, homework drafts, slides, speaker notes and study material are produced by a third-party language model (currently OpenAI's API). They are not:

• written or reviewed by a person at CourseLite,
• fact-checked against anything,
• verified against your syllabus, your instructor's expectations, or a marking rubric,
• proofread, or
• guaranteed to be original.

Nobody at CourseLite reads your generated work before you do.

2. The specific ways it goes wrong

Not a generic caveat. These are the actual failure modes, and knowing which one you are looking at is what makes checking tractable.

1. It invents facts and states them confidently. A fabricated date, a statistic that does not exist, a formula with a wrong constant — delivered in exactly the same tone as a correct one. There is no signal in the writing that tells you which is which.
2. It misattributes. A quote given to the wrong author, an idea credited to the wrong theorist, a source that says nearly but not quite the thing it is cited for.
3. It gets arithmetic and derivations wrong. Confidently, and often with a correct-looking method around a wrong step.
4. Grounding is not correctness. CourseLite answers from your own course material and shows where an answer came from. That makes it traceable, not right. The cited passage may not support the claim, and it may not be the passage your instructor had in mind.
5. It cannot see what it never read. Extraction fails on some material: a scanned page, a handwritten note, a diagram with no text, a file your Canvas account cannot open, a page your instructor has not published. CourseLite answers as though it read everything available, and does not know what it is missing.
6. It is not stable. The same question can produce different answers on different days, because the model changes and generation is not deterministic.
7. It does not know your course's conventions. Notation, preferred method, the specific definition your lecturer uses in week 3 — unless that is in the material it retrieved, it will use whatever is most common in general.
8. It may reproduce phrasing from your course material. A draft assembled from your readings can echo them closely. Quoting and citing that properly is your job; handing it in unmarked is plagiarism regardless of who or what assembled it.

3. What CourseLite deliberately does not fake

Some of these are worth as much as the warnings above.

• A reference entry is constructed, never written by the model. Entries are built from a document's own metadata, from an identifier you supplied and checked against Crossref, or from the course material as your instructor's own work. An entry that cannot be resolved is rendered as unfinished so that you finish it, rather than being invented so that it looks complete.
• There is no literature search, and none is faked. If you ask for one, the answer is a refusal saying what can be resolved instead.
• Provenance is not a citation. "Found through Lecture 8" and "cite the Smith 2024 article" are different claims and are kept in different fields. The formatter that builds a bibliography entry cannot read the provenance field at all, so a module label physically cannot end up in a bibliography.
• The model has no tools. It returns text. It cannot call a function, change a record, fetch a web address, or write to your Google Drive. A document in your course containing instructions aimed at an AI reaches a model with no way to act on them.
• CourseLite refuses quizzes. When Canvas says an assignment is a quiz, or is delivered through an external assessment tool, CourseLite will not draft it and says so.

4. What you have to do

Before you rely on anything generated, and always before you hand it in:

• Read every word. Not skim.
• Check every factual claim against the course material or another source you trust.
• Redo the arithmetic.
• Open the citations and confirm the passage says what the answer says it says.
• Check it answers the question actually asked, in the required format, at the required length.
• Check for unattributed copying, and quote and cite properly.
• Rewrite it into your own work.

Submitting a CourseLite draft unread and unchanged is the single most likely way to come to harm with this product.

5. What CourseLite is not

Not a source of academic, legal, medical, financial, tax, safety, immigration or any other professional advice. Not a proctored-assessment tool. Not a plagiarism or AI detector, and not a way to evade one. Not a submission mechanism — CourseLite has no way to hand anything in, and is not building one.

No guarantee of a grade. Nothing CourseLite produces is a prediction of, a promise about, or a contribution to any mark. We do not guarantee that generated material is accurate, complete, current, original, appropriate for your course, or acceptable to your instructor.

On AI detectors. Institutions use tools that claim to identify machine writing. They are unreliable in both directions — they flag human writing and miss machine writing — and they are used to open misconduct cases anyway. CourseLite makes no claim about whether its output will be flagged, does nothing to evade detection, and would not help you if it could.

6. Where your material goes to make this happen

To generate anything, CourseLite sends the relevant part of your course material, and what you type, to a third-party AI provider's API. That is how the product works; there is no version of it that keeps everything on one machine. Section 7 of the Privacy Policy says exactly what is sent, what is not, and what the provider publishes about what it does with it.

If you would rather your coursework were not processed by a third-party model, do not use the features that generate text. Adding and browsing a course does not require it.

7. Liability

Section 16 of the Terms of Service limits our liability, and specifically excludes liability for academic consequences — a grade, a failed course, a misconduct finding, disciplinary action, a lost scholarship or place. That exclusion is there because the decision to hand something in is one only you can make, and Sections 2 and 4 above are why that decision needs you to have read the thing first.

Nothing here excludes liability that the law applying to you does not allow to be excluded.

8. Questions

support@courselite.org. If your question is "what exactly does CourseLite send where, and what can it reach" — ask, and we will tell you plainly. If it is "does my professor allow this" — we cannot answer that one, and the Academic Integrity Policy explains why.