Will I Get Caught Using ChatGPT?

Will I get caught using ChatGPT? Structurally, 'caught' is not one event: it is the end of a sequence that starts with an instructor's concern, usually continues with a conversation, and only sometimes reaches a formal report, a review panel, and a decision.

UC San Diego's own published procedure is a useful way to see that sequence in detail. It names five phases in order, attaches a business-day deadline to almost every step, and at no point lets a percentage score from a detector decide anything by itself. The structure matters more than any list of supposed tells, because it shows what is actually at stake at each stage, and what happens if a case reaches one.

That level of formal process exists partly because instructors are anxious too. A 2025 College Board survey of more than 3,000 US college faculty found 92 percent concerned about AI-enabled dishonesty in their courses, while only 21 percent said they felt very confident guiding AI use in their own classroom.

Turnitin's AI writing report is the specific software most students picture here, and what it actually catches and misses is a separate, detailed question. What matters is what happens around a report like that in university procedure: a percentage on a screen is never the end of the process by itself.

What Actually Happens First

In the first stage, there's almost never a hearing. Most published processes (including UC San Diego's) have an instructor reviewing the specific work that raises concerns and having a conversation with the student before filing anything. This is what the University of Southern California's Office of Academic Integrity says in their own guidance to faculty: relying on an AI detection tool's output alone is 'insufficient to determine responsibility without more analysis or other supporting elements.'.

But this conversation isn't just a formality. If the student has a legitimate reason, this ends the discussion right now, and there're several published ways to handle such cases, so that the whole thing closes here without any formal record at all. But if it doesn't end here, then the instructor moves to a written report, and that is where the process gets procedural rather than personal.

How a Formal Report Moves Through the System

Once a report is filed, a named office takes over rather than the individual instructor. At UC San Diego, that office is the Academic Integrity Office, and its published procedure names five phases in order: Reporting, Decision and Resolution, Sanctioning, Appeals, and Closing. Each phase carries its own deadline, most of them measured in two to fifteen business days.

Phase	What happens
Reporting	The instructor files an Allegation Report with the Academic Integrity Office, normally within fifteen business days of the grades due date
Decision and Resolution	The office assigns the case and schedules a Resolution Meeting with three possible outcomes: acceptance of responsibility, a request for formal review, or withdrawal of the allegation
Sanctioning	If responsibility is accepted or found, the office confirms the consequence in writing and the instructor submits a grade once the case closes
Appeals	The student can appeal a suspension, a dismissal, or a review decision in writing, normally within ten business days of the outcome
Closing	The case closes in the office's records once a grade is submitted and any appeal window has passed

Whether a case moves past the Resolution Meeting depends on the student. Accepting responsibility moves straight to sanctioning. Contesting it moves the case to a review board, and UC San Diego runs two versions depending on how serious the possible outcome is: a two-person informal review when suspension is not on the table, and a five-person formal hearing, normally three faculty and two students, when it is.

Ignoring a notice does not make it disappear, which is worth knowing on its own. UC San Diego's procedure states that a student who does not respond to an Allegation Notification, or who misses a scheduled Resolution Meeting after a second attempt to reach them, can be presumed to have accepted responsibility. Responding, even just to ask a question, keeps every option on the table that silence closes off.

What a Review Panel Actually Weighs

A student accused of violating the code isn't afforded the same standards as in a criminal trial. UC San Diego's procedure states plainly that 'the rules of evidence used in legal proceedings do not apply' and that a panel decides based on 'the preponderance of evidence,' meaning what's more likely than not rather than proof beyond reasonable doubt.

The panel's only job is to decide whether a violation happened, not why. The same procedure instructs reviewers to decide on the evidence rather than 'intent or motivation of the student,' and gives the student the right to see every document in advance, question anyone testifying against them, and stay silent without that silence counting against them.

What the Range of Outcomes Actually Looks Like

The thing that worries most students is expulsion. This is the rarest consequence described in any published procedure. UC San Diego separates an academic sanction, which an instructor sets and which is usually a grade penalty on the assignment or the course, from an administrative sanction such as suspension or dismissal, which runs through the more serious review path and a much higher bar.

Scale matters too. UC San Diego's own reporting shows roughly 6,433 allegations of all kinds across the five academic years from 2020-21 to 2024-25, involving 5,761 unique students. The university does not publish a separate AI-only count, so this figure spans every kind of violation rather than ChatGPT specifically. It still shows that a report is a common, mostly administrative event rather than the rare catastrophe it can feel like.

What Happens at the Hearing Itself

A formal hearing is procedural before it is anything else. The student is normally notified in writing, told what is being alleged, and given the evidence the panel will consider, with enough time to prepare a response before the meeting happens.

At the hearing, the student typically gets to explain their side, answer questions from the panel, and sometimes bring a supporter, though usually not a legal representative in a straightforward academic case. The panel then deliberates privately, without the student present, and communicates its decision afterward in writing rather than announcing it on the spot.

None of this happens overnight. Once a case gets to a formal panel, you're looking at several weeks between the first flag and the final decision letter. It can take longer if a case is contested, if it sits over a holiday break, or if there's an external check (such as verifying a source). An informal resolution at department level is usually much faster, sometimes settled within days.

The Appeal Route

An appeal is not a chance to re-argue whether AI use should be against the rules. Almost every procedure limits an appeal to a short list of grounds:

• A procedural error in how the original case was handled
• New evidence that was not available at the time of the original decision
• A penalty clearly disproportionate to the finding

Simply disagreeing with the panel's judgment is rarely, on its own, a valid ground. Appeals typically go to a different, often more senior body than the one that made the original decision, and run on a tighter deadline than the original case did, commonly a matter of days to a few weeks from the date of the decision letter. An appeal can uphold the original decision, reduce the penalty, or send the case back for a fresh hearing; it rarely increases a penalty, since the point of the route is review, not re-litigation.

None of this is a substitute for knowing where a specific use stood before submission. TextPulse's own guidance on ethical AI use is a reasonable place to check that in advance, since resolving doubt before a deadline is considerably cheaper than resolving it in front of a panel afterward.

Whether humanizers count as a policy breach in their own right is answered separately. The heaviest cases involve fabricated citations, which are treated as misconduct on a different footing entirely.

None of these stages move quickly. A case that starts with one flagged paragraph in October can still be open in February, which is itself worth planning around: a resubmission deadline, a graduation date, or a reference request does not pause just because a panel has not met yet.

Will I Get Caught Using ChatGPT? The Realistic Answer

Put plainly, using ChatGPT on work your course does not permit carries a real chance of a conversation you would rather not have, and a smaller chance of a formal process with a documented, appealable outcome. It does not carry one fixed, universal chance of expulsion. No article, including this one, can answer 'will I get caught using ChatGPT' with a single number, because the answer depends on your institution's specific AI use policy, your instructor, and what you actually submitted.

The more durable move, regardless of any of this, is treating your own draft the way a careful reader would before you submit it, rather than after a report has already started. A free AI word cleaner will flag vocabulary and phrasing patterns that read as generic or machine-typical in your own writing, the same patterns an instructor is already primed to notice.

Paraphrasing is the usual next move, and whether Turnitin detects paraphrased text has its own page. Models other than ChatGPT raise the same question, and whether Turnitin detects DeepSeek is covered separately.

TextPulse's AI humanizer for students exists for exactly that earlier stage: it reports an estimated Human Score built from your own draft, meant to be read before submission rather than argued about after the fact, since no tool can speak for a specific professor's judgment on a specific case. The more reliable move is still the plainest one: write the draft so it is unmistakably built from your own reading, your own seminar, and your own argument, until the entire question stops being one you need to ask.

Related research: whether an AI model can be told to write like a person is tested in Do AI Models Speak Human?, a TextPulse Research working paper in which four flagship models were given a detailed style brief and a human example, then scored on a stylometric spectrum and on GPTZero against real journal prose.