Your professor sees Turnitin’s report, but you don’t get that same opportunity beforehand. If you’re a student, you know the uncertainty of submitting a paper without knowing what it will flag. We built Academi.cx to put that information in your hands, with an AI detector trained specifically to reproduce Turnitin’s results.
What is Academi.cx?
Academi.cx is built to show students what Turnitin will flag before their professor checks their paper. We trained our AI detector on real Turnitin reports to reproduce its overall AI scores and sentence-level highlights, achieving the highest accuracy against Turnitin of any detector in our benchmark. Our separate similarity checks return an authentic report directly from Turnitin, giving you both results in one place before submission.
You can upload a PDF, DOC, DOCX, or TXT file, or paste your writing into our detector. We extract the readable text and analyze it with our AI model, returning your result within seconds. Your report highlights individual sentences inside the document so you can read them in context and review anything unexpected before running another check or submitting your paper. Subscribers receive unlimited AI checks on documents up to 50,000 words.
How we trained our AI detector
We started with a pretrained language model that had already learned general patterns in written language, then adapted it to predict which sentences Turnitin would flag. This process, called fine-tuning, used over 300,000 real Turnitin reports. Each training example paired writing with labels taken from Turnitin’s highlights, identifying the sentences it flagged and those it left clear.
During training, our model predicted which sentences Turnitin would flag, and we compared its predictions with the labels in the training reports. We used the differences to adjust the model, with further training focused on the current version of Turnitin’s detector. We also kept separate reports out of training so we could evaluate how well our model applied what it had learned to unfamiliar writing.
What the AI score and highlights mean
Your overall AI percentage estimates how much of the analyzed writing Turnitin is likely to identify as AI-generated. A 40% result describes the assessed text; it does not mean there is a 40% probability that you used AI. We handle references and other excluded material separately from the prose being assessed.
Your sentence scores and highlights help you see where our model’s signals are stronger or weaker. Cyan highlights indicate stronger signals, while light purple identifies weaker signals where that category appears in your report. Sentence scores are not verified probabilities of authorship, and we calculate the overall percentage separately, so averaging the sentence scores will not reproduce it.
Surrounding sentences can also affect how our model assesses a passage, which is why you can inspect a highlight within its original paragraph. Document length matters too: a difference in which sentences our detector and Turnitin flag can change the overall percentage more noticeably in a short essay, where each sentence represents a larger share of your writing.
How closely does it match Turnitin?
We test our detector by checking the same documents with Turnitin and comparing both the overall percentages and the individual sentences each tool flags. Our published benchmark of 429 documents, totaling approximately 944,000 words, recorded an average deviation of 1.92 percentage points from Turnitin’s AI scores.
Across 57,478 evaluated sentences, our detector matched Turnitin’s classification on 55,547, giving 96.64% sentence-level accuracy against Turnitin. Within the subset of documents Turnitin scored at 0% AI, sentence-level accuracy against Turnitin was 99.8%. These figures measure how closely we reproduce Turnitin’s assessment; they do not independently establish who wrote a document or guarantee an identical result on every submission.
How similarity checks work
While our AI report estimates how Turnitin will assess AI-generated writing, your similarity report identifies text matching material in Turnitin’s comparison sources. It comes directly from Turnitin and can take up to about three minutes to arrive. Once it is ready, you can switch between the two reports, inspect highlighted passages, and follow available links to the matched sources.
A match does not automatically mean plagiarism, because quotations, references, and common phrases can also appear in the results. Reading each passage alongside its source helps you determine whether the overlap is expected, properly attributed, or something that needs attention. Your institution’s filters and exclusions can affect the final similarity percentage it displays.
What happens to your documents
We never add your documents to Turnitin’s global repository, so checking a document does not create a repository entry that could later match against your submission. We store your files to provide the service and make completed checks available in your account. When you select similarity checking, we send your document to Turnitin to generate the report.
