How to Tell If a Student Essay Was Written by AI

How to Tell If a Student Essay Was Written by AI

You cannot reliably tell from the text alone, and a detector score on its own is not proof, least of all for the second-language writers many international counselors teach. A 2023 study found that the AI-text detectors it tested "consistently misclassify non-native English writing samples as AI-generated." Treat a suspicion as a reason for a 15-minute conversation, never as a verdict.

Most guides on spotting AI writing are written for college instructors grading coursework. A school counselor is in a different position. You know the student, you have seen their writing for years, and you write a letter that goes to every college on their list. That makes the conversation more useful than any score, and it makes your letter the place where an unresolved suspicion does the most damage. Our position: a suspicion you have not resolved with the student never goes into your letter. This page gives the rule that applies, why detectors misfire on second-language writers, which signals are worth a conversation, a script for that conversation, and what the counselor form invites you to write.

Common App's Fraud Policy (dated August 2023), the 2026 counselor recommendation form and the study quoted below were checked on October 30, 2026.

What the rules actually prohibit

Common App's Fraud Policy lists, among the circumstances in which fraud may occur:

"Submitting plagiarized essays or other written or oral material, or intentionally misrepresenting as one's own original work: (1) another person's thoughts, language, ideas, expressions, or experiences or (2) the substantive content or output of an artificial intelligence platform, technology, or algorithm;"

Note two words. "Intentionally": the policy is about presenting work as one's own, not about a tool touching a draft. "Substantive": the clause is about content and output, not about every use. Colleges set their own rules on top of this, and they differ. GradPilot's AI policy directory lists published college policies; read the specific college's rule, and if a college publishes none, the certification inside the application is what the student signs. Common App's AI fraud policy, decoded covers the policy in more detail.

The policy also names counselors among the people who may report suspected fraud: "These individuals may include, without limitation, an applicant's counselor, recommender, high school representative, school advisor, parent or guardian." It describes a process with written notice to the accused user, a determination, and an appeal "within ten (10) business days". It is a process, not an accusation on a hunch.

Why a detector score is not proof for your students

Liang and colleagues tested several widely used AI-text detectors on writing samples from native and non-native English writers (arXiv, 2023). Their abstract reports that the detectors "consistently misclassify non-native English writing samples as AI-generated, whereas native writing samples are accurately identified." The authors write that detectors "may unintentionally penalize writers with constrained linguistic expressions" and "caution against their use in evaluative or educational settings, particularly when they may inadvertently penalize or exclude non-native English speakers."

The reason matters for a counselor. A student writing carefully in a second language often uses common words, safe sentence patterns and textbook transitions. In our reading, that is what the authors mean by "constrained linguistic expressions". Our longer explainer covers why AI detection is biased against international students.

Our view: for a non-native writer, a high detector score on its own is not evidence of authorship you can act on. Do not act on the number alone. If a score exists, treat it as a prompt to talk. Detectors released after 2023 were not part of that study, so its finding describes the tools it tested, not every tool; the conversation below works whatever tool was used. Institutions that act on scores alone fall into what we call score-as-verdict in the AI-detection due-process ladder.

The countercase: if the student's essay contains events you know did not happen, or claims that contradict their record, that is a factual problem, and it is worth raising whatever any detector says.

Signals worth a conversation, and ones that are not

This is our view, not a published standard. No signal on the left proves anything; each is a reason to ask.

Constructed examples: one paragraph that reads "AI-like", one that is empty

Two constructed examples, written by us, not real student essays. Context (invented): both students are second-language English writers at an international school.

Paragraph A (58 words):

Furthermore, training for the Medan half-marathon taught me discipline. Every Saturday I ran the canal road at 5 a.m. and wrote my times in a blue notebook. In March, my times got worse for three weeks in a row. My coach had not noticed. I changed my shoes, not my plan, and by April the times came back.

Paragraph B (53 words):

Throughout my journey, I have always been passionate about helping others. This experience taught me the importance of resilience and perseverance. I learned that every challenge is an opportunity for growth. These invaluable lessons have shaped who I am today and will guide me as I pursue my dreams in college and beyond.

Paragraph A opens with a stiff transition and a formal register, the features a detector or a quick reader might flag. It also contains things only a person who was there would know: a city, a day of the week, an hour, a blue notebook, three bad weeks. You can ask about every one of them. Paragraph B has no fact you could check. It may be the student's own writing; it is also the paragraph a conversation should focus on, because it gives colleges nothing, whoever wrote it.

The 15-minute conversation

Frame it as a conversation about the essay, not about cheating. The goal is to learn how the essay was written.

Constructed example: an eight-question script

Constructed example, written by us, not a transcript of a real conversation.

1. "Walk me through how you wrote this. Where did you start?"
2. "Can I see an earlier draft, or the document's version history?"
3. "Tell me the scene in your second paragraph out loud, as if I wasn't there."
4. "What did you cut from this essay, and why?"
5. "Who has read it, and what did they change?"
6. "Did you use any tool for ideas, translation or grammar? Which parts did it touch?"
7. "This sentence is very polished. What did your first version of it say?"
8. "Can you rewrite this paragraph now, in your own words, in five minutes?"

Question 6 is asked flatly, without judgment. A student who used a grammar or translation tool should be able to say so, and colleges' rules on that differ, so the answer is information, not a confession. Question 8 is the most useful for a second-language writer: a student who wrote the original should be able to reproduce its facts, even in plainer English.

After the conversation: if the student shows drafts and can talk through the details, the matter is closed. If the student says a tool or a person wrote it, the essay is the student's to fix before submission, and the guide to an essay flagged as AI that you wrote covers the student side if the concern was wrong. Make sure the student has read the rules that apply to them.

What goes in your letter, and what never does

The 2026 Common App counselor recommendation asks for "a short evaluation (1000 words or less) that will help colleges differentiate this student from others" and suggests three things to consider, including:

"Observed problematic behaviors, perhaps separable from academic performance, that an admission committee should explore further."

Our view: a suspicion you have not resolved is not an "observed" behavior. Three rules follow.

1. Write only what you have verified. If a student admitted that someone else wrote the essay and then replaced it with their own, there is nothing to report about the essay that was never submitted.
2. Do not hint. A phrase like "her writing has improved remarkably this term" can read as a signal you did not intend. Write about what you know.
3. Remember the letter goes to every college. Common App's recommender guide says submission is "a one-and-done process", sent "to all the colleges you're assigned to."

The countercase: if you have direct evidence that a submitted essay is not the student's work, and the student has refused to fix it, the form's wording covers it, and so does the fraud policy's reporting route. That is a decision for your school's leadership, made on evidence, not on a score.

Questions about spotting AI in student essays

Can you tell if a college essay was written by AI?

Not reliably from the text alone. The 2023 study found that the detectors it tested misread non-native writing as AI. The best evidence is process: drafts, version history and whether the student can explain the details.

Should a counselor use an AI detector on student essays?

We advise against acting on a score alone, especially for second-language writers. The study's authors "caution against their use in evaluative or educational settings."

Do I have to mention AI suspicions in my counselor letter?

The form invites "observed problematic behaviors"; it does not ask for suspicions. Resolve the question with the student first, and write only what you have verified.

What if the student admits they used AI?

Then the essay should be rewritten by the student before it is submitted. Check the rules of each college on the list in the AI policy directory. Parents can read how parents can raise AI use without accusing, which turns the Coalition's four questions into a conversation.

Get a free review of your Common App essay

Your student runs the review on their own draft. Get a free Quick Review of your Common App essay against GradPilot's Common App personal essay rubric, which reads whether the essay builds one main moment, stays inside the scene, and uses concrete details that do work in the story. For background and identity essays, use the Common App background and identity rubric.

More for counselors and teachers of second-language writers:

• Teaching the college essay to ESL students: lesson structure for second-language writers.
• Do colleges use AI detectors?: what colleges say about detection.
• Why AI detection is biased against international students: the research in more depth.

Sources

Retrieved October 30, 2026.

• Common App, Fraud Policy (date of last update August 2023; read through Common App's page data): AI and plagiarism clause; reporting individuals; notice and appeal.
• Liang, Yuksekgonul, Mao, Wu and Zou, study of AI-text detectors on non-native English writing (arXiv abstract, 2023): misclassification of non-native writing; authors' caution.
• Common App, counselor recommendation (2026 PDF): evaluation prompt and the problematic-behaviors bullet.
• Common App, Recommender guide: one-and-done submission.