We built Academi.cx to help students understand how Turnitin is likely to assess their writing before they submit it. Our AI detector is trained to closely match Turnitin’s results, and this comparison puts that work to the test. By checking the same documents with both tools, we show where our results agree and where differences remain.
Watch the full comparison on YouTube
How the test worked
We sourced five different academic papers across five disciplines, including English, political science, economics, and nursing, so the detector couldn't lean on a single writing style. For each document, a portion of the paragraphs were run through ChatGPT with instructions to rewrite them naturally and bold whatever was added. That gave us a clear record of exactly which sentences were AI-generated before either detector ever saw the document. From there, each document was uploaded to Academi.cx and Turnitin at the same time and formatted the same way, so the comparison stayed one to one.
The results from each detector
- Document 1: Academi.cx detected 14%, predicting 0 to 19%. Turnitin's exact score was hidden below 20%.
- Document 2: Academi.cx detected 34%, predicting 27 to 38%. Turnitin returned 39%, 1 point above the range.
- Document 3: Academi.cx detected 64%, predicting 58 to 68%. Turnitin returned 58%, inside the range.
- Document 4: Academi.cx detected 21%, predicting 15 to 25%. Turnitin returned 24%, inside the range and 3 points above Academi.cx.
- Document 5: Academi.cx detected 22%, predicting 14 to 27%. Turnitin's exact score was hidden below its 20% reporting threshold.
Where the detectors agreed
Going paragraph by paragraph, Academi.cx and Turnitin flagged the same AI-generated sentences the overwhelming majority of the time. When they didn't align, it usually came down to how each detector defines a sentence, or one detector catching a borderline sentence the other didn't quite flag. Neither detector caught everything, but the overlap between the two was consistently close, often close enough that the highlighted sections appeared nearly identical when placed side by side.
The one thing Turnitin won't show you
Two of the five documents in this test scored under 20% AI, and in both cases Turnitin didn't show a percentage or any highlights at all, just an asterisk. That's intentional. Turnitin holds back detailed AI results at low percentages to avoid falsely implicating students who show a low, possibly false positive, reading. Students get zero visibility into what their document actually flagged in that range.
Academi.cx doesn't hide that. Even when a document scores low, you still see the percentage and exactly which sentences are flagged, giving you a look at your own risk level that Turnitin simply won't provide.
What this means for you
Across five completely different papers and disciplines, Academi.cx's predictions lined up with Turnitin's real results closely enough that the small differences came down to how each detector counts a sentence, not any meaningful difference in accuracy.
