Checking an essay for AI detection before you hand it in
What a pre-submission check can and cannot tell you, and which parts of an essay raise a score without anyone having generated anything.
Running your own essay through a detector before submitting is a reasonable thing to do, and it is worth knowing what you get back, because the number answers a narrower question than it looks like it answers.
What the score is
A classifier's estimate that the text has the statistical properties generated writing has. Nothing was matched against a source, nothing was looked up. It is a probability, converted into a percentage by a threshold somebody chose.
That means a score is a statement about how your essay reads, not about how it was produced. The two correlate, which is why detectors work at all, and they come apart often enough that the error rates are worth knowing. Turnitin publishes under 1% of documents and around 4% of sentences. Ours is 0.8% at the public setting and 0.4% stricter, on academic prose.
The parts of an essay that raise a score on their own
Detectors measure regularity: predictable word choice and low variation in sentence length. Several things you were probably taught to do produce exactly that.
Formal register. Essays are written formally, and formal writing uses conventional constructions. Conventional means predictable, and predictable is the signal.
Connective openings. Starting sentences with "Moreover", "Furthermore", "Additionally", "Overall". These are taught as good structure and they are also the single most recognisable marker in generated prose. In our corpus they open 2.4% of human sentences against 5.8% of generated ones.
Even sentence rhythm. Writing every sentence to a similar length reads as polished and scores as machine-like. Human writing measured across our corpus has a within-document sentence length standard deviation of 8.7 words against 6.4 for generated text.
Heavy editing. Every revision pass smooths prose, and smoothing removes the variation that reads as human. A well-edited essay looks flatter than a first draft. See does Grammarly get flagged as AI.
Writing in a second language. Published research repeatedly finds elevated false positive rates here, because learned English applies its rules more consistently. See why non-native English writing gets flagged.
Reading your own result
Look at where the score sits, not just what it is. A score concentrated in one section points at that section. Spread evenly across the essay, it points at your writing style as a whole, and those want different responses.
Discount short pieces. Under a few hundred words, every detector is substantially less reliable. A 300-word response gives a classifier far less to work with than a 2,000-word essay. See why detectors are unreliable on short text.
Run two tools if the result surprises you. Detectors disagree for structural reasons, and wide disagreement usually means the text sits near a boundary rather than that one tool is broken. See why AI detectors disagree.
Do not expect the number to match what the institution sees. Different tool, different threshold, different segmentation. Yours is a second measurement, not a preview.
If the score is higher than you expected
The fixes are the same handful of things, and none of them require changing what you are saying. Vary sentence length deliberately. Cut connective openings and let the logic carry the transitions. Keep the specific detail you were tempted to generalise, because specificity is the strongest human signal there is. The full list is in what makes text read as machine written.
Ours is free and unlimited with no account, takes .pdf and .docx, has no word cap and scores each paragraph separately so you can see which passages are flat.