How Do Teachers Prove Students Used AI to Write for Them?

Even if a teacher cannot prove that AI like ChatGPT wrote something, evidence may still suggest it is more likely than not that the student did not write all of it. The truth is: AI is not special under most student codes of conduct. When it comes to the consequences, an academic integrity violation using AI may be much the same as an academic integrity violation using any other prohibited tool: human or AI.

However, it is important to know what evidence is specific to AI and what evidence is not. It may be that a school’s policies regarding AI are not nearly as clear as its policies regarding hiring another person to write a paper, for example. More critically: it may be that a teacher’s accusation relies upon unreliable—and sometimes biased—methods of AI detection.

Evidence Teachers Use to Show a Student Did Not Write Their Own Paper

Some methods used to prove cheating via AI are not dissimilar from those used to prove any kind of cheating or plagiarism, like:

Inconsistency in Writing Style

An assignment written in a tone, style, or vocabulary that is markedly different from previously submitted work is a red flag to teachers.

High-Level Work

Submitting work at a level above and beyond what is typically expected from students in class may indicate that the student did not write the essay themselves. This does not only apply to a high-level writing style or well-crafted argumentation. Research is a common indicator. For example, including citations to sources not typically accessed by (or accessible to) students of a certain level may trigger suspicion.

Missing Citations

If a paper provides information that should come with a citation but doesn’t, it is unclear where that information came from. This may result from carelessness or a technical glitch like a printer cutting off endnotes. However, it can also indicate a reluctance to credit sources which may trigger further scrutiny. Failure to cite one’s sources is problematic even without AI involvement or malintent. Crediting sources is essential to upholding academic integrity standards.

Variations in Font, Spelling, and Citation Style

Inconsistencies in font, spelling conventions, and citation styles can suggest multiple authors. Such inconsistencies may appear when:

• One section of a paper uses American spelling, while another uses Canadian.
• One paragraph cites sources with Chicago-style footnotes, but another uses MLA parenthetical citations.
• Some words or sentences are in Times New Roman, but another is in Cambria.

The above methods are used to recognize cheating and plagiarism in many forms, be it copy/pasting from content written by another person, having someone else write portions of an assignment, or reliance upon GenAI.

Methods specific to detecting AI-based cheating in particular include:

Hallucinations

Hallucination refers to instances where AI presents misleading, false, nonsensical, or completely made-up information with an air of total confidence and authoritativeness. Because hallucination is a common—and commonly known—feature of GenAI, it is a relatively reliable indicator of AI-produced content. AI cannot critically evaluate or understand information. It can only determine the statistical likelihood of one word following another. It can present a convincing mimicry of trustworthy information in some instances, but it is only a mimicry.

Say you prompt ChatGPT to write a paragraph: The impact of AI use on Long Island students K-12 that cites three sources. ChatGPT reaches a point in a paragraph where a citation usually would, statistically speaking, appear. It will insert something that looks like a citation that would usually appear there. That citation may reference a real source, but it may also reference material in no way related to students or Long Island. Its citation might be completely made up, or, “hallucinated.”

Note: Even if not AI-produced, falsifying citations is another form of academic dishonesty.

Outdated Information and Knowledge Cutoff

Large language models (LLMs) are trained using vast amounts of information, but it is still a limited set sourced from a limited span of time. Meaning, if Gemini has only been trained using data up until April 2023, it cannot reproduce knowledge of events after that date. This is knowledge cutoff. If an assignment asks about the development of student rights over the last three years, and an essay only references four years ago, that is a red flag for knowledge cutoff, signaling possibly AI-generated content. If an essay cites sources not easily accessible to present-day users, that is another indicator that AI and not a human was navigating the website or database whence the information derived.

AI-Detection Tools

Unless a school has already banned the use of AI-detection tools like GPTZero for student disciplinary matters (as some have), odds are high that such AI-detection reports make up part of their case. It is entirely possible that a school recognizes how problematic and unreliable AI-detectors are, but the individual instructor accusing a student of submitting AI-generated work may not.

Responding to Allegations of Academic Dishonesty Involving AI

As GenAI use among students becomes more commonplace, so will allegations that students used AI to cut academic corners. Unfortunately, not every student accused of inappropriate AI use will have done anything wrong. False allegations do occur; the consequences can be long-lasting and severe. If you or your student is facing unfair allegations of academic dishonesty involving AI, it is highly advised to consult an attorney experienced in education to discuss options and coordinate your strategy.

Navigating the student disciplinary and appeals process can be overwhelming. Tully Rinckey’s education attorneys will handle your matter with the attention and tact it deserves. If you have additional questions about the rights of parents and students in New York state, our team of attorneys is available to assist you today. Please call 8885294543 to schedule a consultation, or schedule a consultation online.

Greg T. Rinckey is one of Tully Rinckey PLLC’s two founding partners. He worked with Founding Partner and fellow Hofstra University alum Mathew B. Tully in 2004 to build the firm from the ground up into the coast-to-coast, full-service powerhouse that it is today. As Founding Partner, Greg collaborates with Mat in all areas of strategic planning and law practice management to develop and deploy innovative business solutions that continue to grow the firm.