Good English
is not evidence.

Why did a checker flag writing you wrote every word of?

Because the tool measures how predictable your words are. It does not measure whether you wrote them. And careful second-language vocabulary is predictable in exactly the way the tool penalizes.

Here is the mechanism — structure, not feelings. These tools rank text by "perplexity." In plain words: how surprising each next word is. Common, careful, textbook words are not surprising. They produce a low perplexity number. And the tool reads low perplexity as "AI." Why? Because an AI model, by design, also picks the most probable next word. You learned English carefully, from a smaller set of safe, high-frequency words. So your words land where the model's words land. Same place — on the only axis the tool can see. It is measuring your word list. Not your authorship. (Stanford / ScienceDaily)

Here is the proof that it's the vocabulary, not you. In a peer-reviewed study, researchers ran seven checkers on 91 TOEFL essays — all written by real people. On average, the tools flagged 61.22% of them as "AI" (S52). Then the researchers changed one thing. They enriched the vocabulary of the same essays. Same authors. Same ideas. Only the words changed. The average false-flag rate fell to 11.77% (S54). Nothing about the writers' honesty moved. And the same seven tools, run on 88 essays by native speakers (US eighth-grade students), false-flagged only 5.19% on average — about 5% (S52).

What actually pushes back — and what stands on its own?

Three moves. They stack. None of them is built to stand alone. Independent, documented evidence beats one black-box number. That is also the line the real cases turn on. → The cases — coming soon

Move 1 — your prior writing, and the same checker turned on it

This is the strongest single move for a second-language writer. It answers the actual theory against you: "this doesn't sound like you." A reviewer's first instinct is to set the flagged paper next to your earlier work. So give that instinct an answer. If this paper reads like the three you wrote last term — same formality, same careful vocabulary, same habits — the "not your voice" theory collapses. It sounds exactly like you. On a record made before anyone asked.

Then turn the tool on itself. Take an essay you wrote years ago, before these tools existed. Run it through the same checker. When a UC Davis student was accused, he cleared his name partly this way: he showed the tool flagged Martin Luther King Jr. and the Book of Genesis as "AI" (Rolling Stone). A 2019 essay cannot have been written by a 2023 model. So a "high AI" reading on your own old work proves one thing, by itself: the instrument misreads your style. It is not catching your authorship.

The catch, said plainly: this is strong support, not a standalone win. Pair the baseline with independent timestamps and genuine drafts. Those are things an outside system recorded — things you could not invent after the fact. → The evidence that actually holds up →

Move 2 — cite the bias finding, as support

It lands because it is a peer-reviewed measurement. Not a feelings argument. Not a claim about anyone's intent. The same seven tools read two groups of human-written essays. TOEFL essays by non-native writers: flagged 61.22% on average. Essays by native speakers: about 5% on average (S52). That gap is the argument. A tool this wrong on one group is not measuring authorship. It is measuring how you write English. Say it as a number, not a grievance.

The catch: it rarely carries a case standing alone. Schools have held firm when they had evidence beyond the tool. So it is your second sentence, not your first. Lead with your own record (Move 1). Cite the finding to show the tool is unreliable on writers like you. → The cases — coming soon

Move 3 — the Daubert frame

The reliability argument, structured. Its prongs — and how a black-box checker measures up — are in the next section. Read it before you reach for it. Where you use it decides whether it helps.

Does the Daubert argument actually work in a campus hearing?

Most students can skip this. It matters mainly for a formal hearing, an appeal, or a high-stakes / visa case — if you are still at the first-email stage, the three moves above are enough. ("Daubert" is just the U.S. courtroom test for whether an expert method is reliable enough to trust.)

As a court rule you can force on a panel — no. As the structure of your reliability challenge — yes. Here is how to use it.

Daubert is a courtroom test, under Federal Rule of Evidence 702. In plain words: it decides whether an expert method is reliable enough for a judge to allow it. A black-box AI checker fails it on every prong (Justice Speakers Institute):

The honest catch: Daubert binds courts. A campus hearing is not a court, and no rule forces a panel to apply FRE 702. So don't wave it like a law. Use the four prongs as the shape of your argument — a clean, named checklist the tool visibly fails. It turns "the number says AI" into "this number is not reliable enough to decide my future." It carries the most weight in a formal hearing, an appeal, or a high-stakes / visa case — argued by you or by counsel. → What happens at the hearing →

On a student visa — does a flag end your status?

No. A flag is not a finding. It does not end your status by itself. Preserve every piece of your evidence. Then talk to your DSO and an immigration attorney before any departure — and before any conversation that could be read as an admission. The stakes are higher for you. That is exactly why the record matters more. → The high-stakes clock →

If my formal style is the problem, does a record even help?

Yes — where it actually counts. But be clear about what it does and doesn't do. A record does not stop the flag. The checker will likely still read your careful, formal English as "predictable" — predictability is all it measures. What the record changes is the next step. A human reviews the case. A consistent baseline and a real writing process answer "this doesn't sound like you." That is what the cases students won turned on. The flag may fire either way. The record wins the review the flag triggers.

Two things a formal or second-language writer should do now:

• Treat your prior writing as a deliberate baseline, not an afterthought. Your past graded papers already are a baseline — but only if you can produce them on demand. Pull a few earlier pieces into a folder you control, today. "Here is how I have always written" should be one click away. Not a scramble.
• Lead with the independent record, not the bias finding. The 61.22% figure is real. Cite it. But on its own it rarely wins. Pair it with the timestamps and drafts that show this paper taking shape. The finding explains why you were flagged. The record proves who wrote it.

The unfair part this page exposes

Look back at the three moves. They share one thing: each had to already exist before you were accused. A baseline of prior writing. Genuine drafts. An independent timestamp. The tool turned your vocabulary into the case against you. Then the burden flipped to you — answer a number, with nothing but your word. That barehanded "prove you wrote it" falls hardest on a second-language writer. The instrument already mistook your careful English for a machine's.

The thing that actually answers it is the one thing you can't assemble under pressure after the flag: a record of the writing as it happened, already there before anyone asks.