Module 3 · Read

AI Detection: Myths and Realities

As AI writing tools spread, so did "AI detectors" that claim to tell whether a piece of text was written by a human or a machine. They are widely used and widely misunderstood. This page separates the common myths from what the evidence actually shows, so you know what these tools can and cannot do, whether you are a student or a future educator.

Where Your Campus Stands

(ADD YOUR CAMPUS POLICY.) Many institutions treat AI detectors as not reliable and not proof: a detector score can start a conversation, but it cannot by itself determine that academic dishonesty occurred. Studies have found the same thing, and so have the toolmakers, from OpenAI to Turnitin.

Four Common Myths

Each of these sounds reasonable, which is exactly why it spreads. For each one, take a moment to think about why someone might believe it, then reveal what the evidence shows.

Myth 1: "A detector flagged it, so the student definitely used AI."

Reality

Detector results are probabilities, not evidence. They produce false positives, flagging human writing as AI, often enough that schools have stopped trusting them. Vanderbilt University disabled Turnitin's AI detector in 2023, noting that even a 1 percent false-positive rate would mean roughly 750 of their 75,000 yearly papers could be wrongly flagged. A flag is a starting point for a conversation, not a verdict.

Myth 2: "AI detectors treat everyone the same."

Reality

They do not. A Stanford study found detectors wrongly labeled more than 61 percent of essays by non-native English writers (TOEFL essays) as AI-generated, while rarely misflagging native English writers (Liang et al., 2023). Writing that uses simpler or more predictable word choices, common for multilingual writers, looks "AI-like" to these tools. That makes detectors a fairness problem, not just an accuracy problem.

Myth 3: "Detectors are imperfect now, but a better one will fix this soon."

Reality

Do not count on it. OpenAI, the maker of ChatGPT, launched its own AI-text classifier and then shut it down in July 2023 because of its "low rate of accuracy." Its tool correctly identified only about a quarter of AI-written text and still flagged some human writing by mistake. If the company that built the AI could not reliably detect it, claims of near-perfect detection deserve real skepticism.

Myth 4: "Running my own work through a detector will protect me."

Reality

Detectors are inconsistent: the same text can score differently on different tools or even on the same tool at different times. A "human" score is not a guarantee, and a false "AI" score on your own writing can cause needless worry. A far stronger protection is keeping evidence of your process, which we cover below.

The Numbers

26%

of AI text correctly caught by OpenAI's own detector before it was retired

Source: OpenAI, 2023

61%+

of non-native English essays wrongly flagged as AI in a Stanford study

Source: Liang et al., 2023

12+

universities that have disabled Turnitin's AI detector over reliability and fairness

Source: Vanderbilt, Johns Hopkins, and others

Why Detectors Get It Wrong

Most detectors guess by measuring how "predictable" writing is. AI models tend to choose common, expected words, so text that is smooth, simple, or formulaic can look machine-made. But plenty of human writing is also smooth and simple, including writing by multilingual students, by people using clear plain language, and by anyone following a tight assignment template. The detector cannot tell the difference between "predictable because AI wrote it" and "predictable because the person writes clearly." There is also no way for it to see your actual process: the drafts, the research, and the thinking that produced the work.

What This Means for You

You do not need to fear detectors, but it helps to protect yourself and to understand your rights:

Keep your process. Write in a tool that saves version history (such as Google Docs or Microsoft Word), keep your notes and drafts, and you will have clear evidence of how your work developed.
Know that a flag is not a finding. At many institutions, a detector score is a prompt for a conversation, not proof. If you are ever questioned, you have the right to explain and show your work.
Follow the policy that applies. Whether AI use is allowed is set by your syllabus and each assignment, and unauthorized AI use is addressed under the Academic Integrity Policy. You looked at how to read those rules on the "Navigating College in the Era of AI" page.

The "keep your process" habit above is your strongest protection, so make it concrete. Tap each habit you already practice to see where your record is strong, and pick up any you do not.

You are tracking 0 of 6 habits.

That is a strong evidence trail. If your work is ever questioned, you can show exactly how it developed, which is far more convincing than any detector score.

For future educators in this course: the same evidence argues for designing assessments and conversations that do not hinge on a detector score. Australian Catholic University shows the human cost of treating these tools as proof: it opened almost 6,000 academic-integrity cases in 2024, around 90 percent tied to AI, and roughly a quarter were later dismissed, many because a detector score was the main evidence. Some students had their results withheld for months before their names were cleared, and the university stopped using Turnitin's detector in 2025.

Knowledge Check

Select an answer to see feedback. Each option explains why it is or is not correct.

Question 1 of 3

An essay gets an "80% AI" score from a detector. What can you correctly conclude?

Question 2 of 3

Research found AI detectors are especially likely to wrongly flag:

Question 3 of 3

Why did OpenAI take down its own AI-writing detector in 2023?

Sources

Liang, W., Yuksekgonul, M., Mao, Y., Wu, E., & Zou, J. (2023). GPT detectors are biased against non-native English writers. arXiv:2304.02819

OpenAI. (2023). New AI classifier for indicating AI-written text (update on discontinuation). openai.com

Inside Higher Ed. (2023). Turnitin's AI detector: Higher-than-expected false positives. insidehighered.com

Vanderbilt University. (2023). Guidance on AI detection and why we're disabling Turnitin's AI detector. vanderbilt.edu

Futurism. (2025). University using AI to falsely accuse students of cheating with AI (reporting on Australian Catholic University). futurism.com

Johns Hopkins University. (n.d.). Detection tools: Limitations and alternatives. teaching.jhu.edu