Essay Flagged as AI but You Wrote It? Prove It

Essay Flagged as AI but You Wrote It? Prove It

A flag on your own essay before you submit is a measurement, not a finding. ChatGPT's developer withdrew its own AI text classifier in July 2023 "due to its low rate of accuracy"; in its tests it wrongly labeled human writing as AI 9% of the time. Keep a dated record of your process, check the flagged sentences, and do not rewrite good sentences to chase a lower number.

Most advice on this question is written for students who have already been accused. This page is for the earlier moment: you ran your Common App essay through a free checker, it came back high, and nobody has contacted you. If a college or teacher has contacted you, read what to do if you are falsely accused of AI cheating instead. That page covers your rights and the process.

Every source quoted below was checked on October 30, 2026.

First: a flag is not a finding

Two published sources set expectations for any detector score.

The first comes from the company that makes ChatGPT. When it launched its AI text classifier on January 31, 2023, it wrote that "In our evaluations on a 'challenge set' of English texts, our classifier correctly identifies 26% of AI-written text (true positives) as 'likely AI-written,' while incorrectly labeling human-written text as AI-written 9% of the time (false positives)." It added: "The classifier is very unreliable on short texts (below 1,000 characters)." The same page now says: "As of July 20, 2023, the AI classifier is no longer available due to its low rate of accuracy." (We read the archived copy; the live page returned HTTP 403 to our request.)

The second is a study by Liang and colleagues, posted on arXiv and published in the journal Patterns in 2023. It found that the detectors it tested "consistently misclassify non-native English writing samples as AI-generated, whereas native writing samples are accurately identified," and that they "may unintentionally penalize writers with constrained linguistic expressions." Our page on AI detection bias against international students covers that research in depth.

Those numbers describe specific tools tested in 2023, not the checker you used. They do show that a confident label on human writing is a known failure, reported by the people who built and tested these tools.

Why human college essays get flagged

Four features of real application essays can push scores up. The sources above document the first and the last; the middle two are our reading. None of them means AI wrote the essay.

• Short length. A 250-word supplement runs about 1,250 to 1,750 characters, not far above the 1,000-character line where the classifier above called itself "very unreliable" (our arithmetic, at 5 to 7 characters per word including spaces, the range of the constructed sentences on this page). Our guide on whether AI detectors are accurate on short essays works through the length math for each kind of answer.
• Formula. Essays that end with a three-part lesson, or open with a dictionary definition, follow patterns that machine text also follows.
• Tool smoothing. Run every sentence through a grammar or rewriting tool and the edited sentences may no longer be yours in rhythm or word choice.
• Plain, careful English. The Liang study's finding about "constrained linguistic expressions" applies to many careful second-language writers.

For the phrase-level patterns, see what flagbait is. We do not repeat that list here.

Triage in ten minutes

Run these four questions in order. Only the last one can tell you to change a sentence.

1. Which tool, and what does it claim? Write down the tool and the date. A free checker with no published accuracy data is a weaker signal than one that publishes its error rates. Either way, it is a signal.
2. How long was the text? Under about 1,000 characters (roughly 140 to 200 words), treat the overall percentage as noise and move to the highlighted sentences.
3. Which sentences are highlighted? List them. If the tool gives only one number for the whole essay, you have nothing to act on. Stop here and keep your process record (next section).
4. Did a tool edit those sentences? For each highlighted sentence, check your version history. If a grammar or rewriting tool changed it, restore your own wording and fix only the error. If the sentence is yours as you first wrote it, keep it.

Our view: step 4 is the only step that should change your essay. Basis: the published error rates above show that a high score does not tell you whether a sentence is human. Your own history does. The countercase: if your essay has a sentence you cannot explain, or that you do not remember writing, the problem is real whatever the score says. Rewrite it from your notes.

Constructed example: one flagged sentence, two diagnoses

Constructed example, written by us, not a real applicant statement. Context (invented): a student, Daniel, writes about working the register at his family's bakery. A checker highlights this sentence:

Working at the bakery taught me invaluable lessons about responsibility, patience, and the importance of community.

Diagnosis A, tool-smoothed: his version history shows his first draft said "The bakery is where I learned to count change in my head while Mrs. Ortiz told me about her grandson." A rewriting tool turned it into the sentence above. Revert it. The original is specific and his.

Diagnosis B, his own: his history shows he typed the flagged sentence himself in his first draft. Keep it or cut it on writing grounds, not because of the flag. It is a generic sentence, and the earlier specific one would serve the essay better, but that is a writing call.

Build your proof now: version history, step by step

The strongest evidence that you wrote your essay is a dated record of you writing it. If you draft in Google Docs, its version history help page gives the steps:

1. Open the document. At the top right, click Last edit to see earlier versions in the right panel.
2. Choose a version, click More, then Name this version. Google says: "You can create a named version to track your version history and make sure your versions aren't merged." You can add "up to 40 named versions per document."
3. Name a version at each milestone: brainstorm, first draft, the cut to length, after feedback, final.
4. To show only your milestones, turn on Only show named versions.

Be honest about what this proves. Google's page says "As the file owner, you can manage your document's version history to remove historical content," and that deleting history "is available only for users with a personal Google Account." Because an owner can delete history, a version record is evidence of process. It is not tamper-proof. It is still far stronger than a reread of the final essay.

Paper counts too. Brainstorm lists, outlines on notebook paper and a teacher's marked-up printout all show where the essay came from. Keep them until your decisions arrive.

Constructed example: a named-version timeline

Constructed example, written by us, not a real applicant record.

Five named versions over eight weeks show the essay changing in the way essays change. A single version pasted in at 641 words shows nothing.

What colleges actually compare your essay against

A detector score is not the only thing that sits next to your essay. Colleges publish other checks:

If the problem is that someone else wrote the draft, read if an agent or consultant wrote your essay instead.

• A graded paper. Princeton's application checklist says "A graded written paper is required, preferably in the subjects of English or history," and that it "will help the Office of Admission assess the student's written expression in an academic setting."
• Verification. Caltech's verification page says "Our team will verify all parts of your application: essays, letters of recommendation, creative work, etc. If you submit it, we will verify it."
• Authentication requests. UC's Statement of Application Integrity says "UC conducts regular screenings to verify the integrity of the responses, may request authentication of the content or writing as the student's".
• Fit with the rest of the file. Coalition's counselor article on AI says "Essays that sound generic or disconnected from the rest of an application often raise concerns, regardless of how they were produced."

None of these pages describes how a detector score is used, and we do not guess. What they do show is that the record that matters is consistency: your essay, your graded paper, your activities and your teachers' letters should read like one person. A version history backs that up if anyone asks.

What not to do

• Do not run your essay through a humanizer or paraphraser. The Common App fraud policy covers "the substantive content or output of an artificial intelligence platform, technology, or algorithm." Text a paraphraser produced is that platform's output, even if you wrote the input. Our page on whether AI humanizers work covers the evidence on those tools.
• Do not write worse on purpose. Adding typos or breaking good sentences to lower a score makes the essay weaker for every human reader, and the score still is not proof of anything.
• Do not keep re-running checkers. Five tools giving five numbers will not settle the question. Your version history will.

Questions about essays flagged as AI

Will my college essay be flagged as AI?

Nobody can promise it won't be. Published error rates show human writing does get labeled as AI, and short, formulaic or heavily tool-edited text is at higher risk. What you control is the record: draft in a tool that keeps version history and name versions as you go.

Does Google Docs version history prove I wrote it?

It is strong evidence of process, not proof. Google lets the owner of a personal account delete version history, so a record can be edited. A record with many dated, named versions that show the essay growing is still the best evidence a student usually has.

Should I send my version history with my application?

Do not send it unless a college asks. Keep it ready. UC's integrity statement says UC "may request authentication of the content or writing as the student's," and Caltech says it may verify "all parts of your application." That is when a version record helps.

Why does my essay say AI-generated when I wrote it?

The sources above name two causes. The classifier's maker warned about short texts and wrote that "Text that is very predictable cannot be reliably identified." The Liang study found non-native English writing misclassified. Our reading: common essay formulas and grammar-tool smoothing push text the same way. Check whether a tool edited the highlighted sentences. If not, the flag tells you about the limits of the tool, not about who wrote the essay. If the concern came from your counselor, see what a counselor should do with an AI suspicion.

Get a free review of your Common App essay

Get a free Quick Review of your Common App essay against GradPilot's Common App narrative craft rubric. It reads whether the essay develops one main moment, whether the details shape the story, and whether the reflection is specific, which is the fix for a generic sentence whatever a checker says. You can also start from the Common App essay review page.

More from this series:

• Can I use ChatGPT for college essays?: what six platform and college texts allow, use by use.
• How to make your college essay not sound like AI: the rewrite-from-notes method.

Sources

Retrieved October 30, 2026, except where noted.

• ChatGPT's developer, New AI classifier for indicating AI-written text (January 31, 2023, with July 20, 2023 update; archived copy, live page HTTP 403): 26% / 9% evaluation figures, short-text warning, withdrawal.
• Liang et al., arXiv 2304.02819 (2023; published in Patterns): misclassification of non-native English writing.
• Google, Find what's changed in a file: Last edit, named versions, 40-version limit, deleting history.
• Princeton University, Application checklist: graded written paper.
• Caltech, Verification of Submitted Materials: verification of all parts of an application.
• University of California, Statement of Application Integrity (last updated August 2023): screening and authentication.
• Coalition for College, AI and the College Essay (dated February 18, no year shown): generic or disconnected essays.
• Common App, Fraud Policy (dated August 2023): AI output clause.