Understand the signal. Not a verdict.
Cortex asks GPTZero how it reads your passage, then shows you the parts a score leaves out: the confidence behind it, which paragraph carries the risk, which sentences drew attention, and what in your writing produced them. Up to 1,000 words, no account. It is not a plagiarism report, not a record of who wrote something, and not a prediction of what your school's software will say.
Your draft
What a detector sees
What lands here
- A whole-document estimate with its confidence
- The passage marked sentence by sentence
- Why each flagged sentence drew attention
Common questions
Which detector does Cortex use?
GPTZero's v2 text-prediction API. Cortex does not run Turnitin, does not search a plagiarism database, and is not connected to whatever your school uses.
Does a high score prove AI use?
No. A classifier can mislabel human work and can miss AI writing. Read its estimate alongside your drafts, your sources and your course policy, and treat a flag as a prompt to look, not a finding.
What are the writing signals you show?
Five properties Cortex measures itself from your text: sentence-length variety, word variety, stock connectives, passive voice and repeated openers. They are labelled as ours because they are not GPTZero's reasoning, which GPTZero does not publish. They are the things classifiers tend to react to, and unlike a score they tell you what to change.
How much text can I submit?
Up to 1,000 words a request on the free sample. Plans raise that and add a per-cycle allowance of checks.
Will the result match Turnitin?
There is no reason to expect it to, and Cortex does not promise it. Different tools, versions and settings disagree with each other routinely, which is itself a good reason not to treat any one of them as proof.