Check your essay before you submit it
Five minutes, before the version that counts. This is the only point in the process where the information is worth anything.
The asymmetry of AI detection is simple. After submission, a number exists that somebody else generated and you are discussing it. Before submission, you can see what your writing does to a classifier and change the paragraphs that read flat.
Same information, completely different value depending on when you have it.
What a pre-submission check gets you
Not a prediction of what your institution's tool will say. Different detector, different threshold, different segmentation, so the numbers will not match.
What it gets you is the shape: which paragraphs in your document carry the properties detectors read as machine-written. That is stable across tools, because they are all reading the same underlying regularity.
A section that comes back flat in ours is a section that will read flat in theirs.
How to run it properly
Run the whole document. More text means a more reliable number. Every detector is far better per document than per paragraph, and per paragraph than per sentence, so checking an extract gives you the least reliable version of the answer. See why detectors are unreliable on short text.
Use the final draft. Editing changes the score, and it changes it upward, because every revision pass smooths out variation.
Score the prose, not the file structure. If you write in LaTeX, check the text rather than the source, or a large part of what gets scored is your markup. See why LaTeX breaks AI detectors.
Read per paragraph, not the document number. One figure over 2,000 words is an average that hides the thing you need.
What to look at in the result
Concentrated or spread. Concentrated in one section points at that section, and the first question is what kind of section it is. Methods, procedures and literature reviews score high for structural reasons. Spread evenly points at your writing style as a whole.
Which paragraphs sit highest. These are the ones to work on, and usually there are two or three rather than twenty.
What to change, if anything
Three things move a score, and none of them touch your argument.
Sentence length variation. The strongest single signal. Human academic writing in our corpus has a within-document standard deviation of 8.7 words, generated text 6.4. Find three sentences of similar length, split one and merge another pair.
Connective openings. "Moreover", "Furthermore", "Additionally", "Overall". These open 2.4% of human sentences and 5.8% of generated ones. Most can be deleted outright.
Specific detail. Name the method, give the number, use the real example. Generalising is the habit that reads as generated, and it is usually introduced while tightening a draft.
Full list in what makes text read as machine written.
Doing it without a word cap
A cap that forces you to check a 3,000-word essay in fragments does not just annoy you, it degrades the answer, because you are asking for several unreliable small results instead of one reliable large one.
Ours is free and unlimited with no account. It takes .pdf, .docx and .tex, has no word cap, scores every paragraph separately, publishes its wrong-flag rate at 0.4%, and never stores your text.