Why an essay you wrote yourself can still get flagged
The software your school uses does not know who typed the essay. It looks at sentence lengths, word choices and punctuation, and compares them with the writing it was trained to call machine made. The trouble is that the writing it was trained on looks a lot like what a careful student produces when they follow the rubric.
In 2023 a team at Stanford ran TOEFL essays by students from other countries through seven of the popular tools, and more than 60% came back as machine written [Liang et al., 2023]. That same year another group tested fourteen tools and found that none of them reached 80% accuracy once you changed the conditions a little [Weber-Wulff et al., 2023]. So if you ran your draft past Grammarly a few times, or you used a very formal template, or English is not your first language, you are the kind of writer these tools get wrong most often. None of that is cheating, and it is better that you know about it before you hand the essay in.
How to check your own essay
- Finish the essay first. A scan is a picture of the text you are about to submit, so run it when the text is done. If you start changing your sentences to push a number down you will make the essay worse and prove nothing to the person reading it.
- Paste the whole thing into the essay checker, the full document and not just one paragraph, since the scan has more to work with when it has all of it. If you give it something shorter than our word floor it will tell you it cannot judge that, and it will not guess.
- Read the verdict and its limits. The score is an estimate. Our published evaluation reports error rates and verdict outcomes for the tested population; those figures do not establish the chance that your individual essay was written by AI. An inconclusive answer means the tool did not make a clear call.
- Save the result with the date on it. A screenshot from a tool that publishes its error rates, taken on the day you submitted, is worth having in a folder if the question ever comes up. It shows that you looked in good faith. It is not proof that you wrote the essay, and you should not present it as proof.
What a low score gives you, and what it does not
A low score from us is one piece of evidence you can keep, and we would not want you to treat it as more than that. The tools are not calibrated to each other. Each was trained on different data and draws its line in a different place, with no shared standard between them, so a low probability from us does not make Turnitin, or any other tool your school might use, agree with us [Weber-Wulff et al., 2023]. You get a number you can point to, with an error rate behind it, and a dated record that you had nothing to hide. A tool that promises your essay is guaranteed to pass is promising something it has no way of knowing.
The thing that protects you more than any score
Your version history is worth more than any number a scanner gives you, since it is evidence in a way that a score never is. Write in Google Docs, or in Word with AutoSave turned on, and the long messy record of the real writing builds up on its own while you work. Keep your outline and your notes and your list of sources as well. If someone ever does accuse you, that trail is what settles it. We have written up what to do when the accusation has already happened at falsely accused of using AI, and the process your school ought to be following is set out in the teachers' guide, and there is nothing wrong with sending that page to an instructor.
Why we do not sell a humanizer
When you search for AI detectors you will run into "humanizers", rewriting tools sold to get text past a scanner, quite often by the same companies that sell the scanner. We do not build them, we do not link to them and we do not give advice on getting around detection. A company selling both the detector and the way around it has told you what its verdicts are worth. And a humanizer takes writing that you did and turns it into text you can no longer say is yours, which is the problem you were worried about in the first place. If you wrote the essay, the evidence is on your side already. If you did not, no rewriting tool changes what you handed in.
What Cobalynx can and can't do here
We can give you a probability with published error rates behind it, a dated result you can keep, and an honest “we could not tell” when the evidence is too thin. We cannot promise that any other tool will agree with us, and we cannot certify that your essay was written by you. Be careful of any tool that says it can. The scans are free, you do not need an account, and your text is never stored, so checking your own work costs you nothing.
Sources
- Liang, Yuksekgonul, Mao, Wu, Zou, “GPT detectors are biased against non-native English writers,” Patterns (2023), arxiv.org/abs/2304.02819 — 61.3% average false-positive rate on TOEFL essays across seven detectors.
- Weber-Wulff et al., “Testing of detection tools for AI-generated text,” International Journal for Educational Integrity (2023) — all 14 tested detectors below 80% accuracy; tools disagree on the same texts, arxiv.org/abs/2306.15666.
Common questions
Will my essay get flagged as AI?
Nobody can promise you that it will not, because the checkers are not calibrated to each other and every one of them has its own false-positive rate. What we do know is what pushes the risk up, and it is heavy Grammarly polish, a very formal conventional structure, and the patterns of someone who writes English as a second language. What protects you is the record of how the essay was written, because your version history and your drafts will count for more than any score, whichever way the score went.
If Cobalynx says “likely human,” am I safe from Turnitin?
No, and we would be wary of any tool that told you that you were. The checkers were trained on different data and they draw their lines in different places, so a low score from us does not make another tool agree with us. What a Cobalynx check gives you is a probability with an error rate that we have published on the evidence page, and a dated result you can point to later, but it is not a guarantee of anything.
Will using Grammarly get my essay flagged?
It can, yes. When you run your writing through a grammar tool a few times, what you end up with is smoother and more even than what you started with, and that smooth, even writing is close to what the checkers were trained to call machine made. Writing that was polished with AI is one of the known ways these tools get it wrong. So keep your version history, so that anyone who looks can see the polishing happen as editing, say that you used the tool if your school asks you to, and treat any one score as one piece of evidence among the others.
Why don't you offer a humanizer to make my essay pass?
Because if we sold one, every verdict we publish would be worth nothing, since a company that sells both the checker and the way around it has already told you what it thinks of its own results. There is also a problem for you, which is that a humanizer takes writing you did and turns it into text you can no longer say is yours, and that is the real integrity risk. If you wrote the essay, then the thing that will protect you is evidence, and the playbook goes through what that evidence looks like.