Comparisons

Academi.cx vs Winston AI Accuracy Test

We compared each detector’s overall AI scores and sentence-level highlights with Turnitin’s results across five essays.

Brayden · 2 min read

Winston AI puts accuracy at the center of its pitch, marketing its detector to educators, publishers, and writers. It also emphasizes reducing false positives, a concern for anyone checking their own writing. With Turnitin as our reference, we compare Winston AI and Academi.cx to see which better reflects the results students can expect.

Watch the full comparison on YouTube

How the test worked

Five essays were sourced from IvyPanda, a free essay database with no affiliation to Academi.cx. Selected paragraphs in each essay were rewritten using ChatGPT to introduce AI-generated content. The same version of each document was then submitted to Academi.cx, Winston AI, and Turnitin. We compared how closely Academi.cx and Winston AI matched Turnitin’s overall AI scores and sentence-level highlights.

The results from each detector

Results for five documents with blue Academi.cx scores and predicted ranges, Winston AI scores, and navy Turnitin scores. Document 1: Academi.cx 41%, range 31–47%; Turnitin 39%; Winston AI 0%; Document 2: Academi.cx 26%, range 19–30%; Turnitin *%, exact result hidden; Winston AI 0%; Document 3: Academi.cx 22%, range 15–26%; Turnitin *%, exact result hidden; Winston AI 21%; Document 4: Academi.cx 32%, range 23–36%; Turnitin 26%; Winston AI 20%; Document 5: Academi.cx 54%, range 45–60%; Turnitin 41%; Winston Human Score 52%, shown as 48% on the graph. *%: Turnitin hides results above 0% and below 20%. Winston: 100 minus Human Score.
  • Document 1: Academi.cx detected 41%, predicting 31 to 47%. Turnitin returned 39%, inside the range. Winston's human score was 100% (0% on the graph).
  • Document 2: Academi.cx detected 26%, predicting 19 to 30%. Turnitin returned *%, with no exact score or highlights. Winston's human score was 100%, and the round remained undecided.
  • Document 3: Academi.cx detected 22%, predicting 15 to 26%. Turnitin's exact score was hidden as *%. Winston's PDF gave a 79% human score (21% on the graph).
  • Document 4: Academi.cx detected 32%, predicting 23 to 36%. Turnitin returned 26%, inside the range. Winston's human score was 80% (20% on the graph), putting both detectors 6 points from Turnitin.
  • Document 5: Academi.cx detected 54%, predicting 45 to 60%. Turnitin returned 41%, below the range. Winston's human score was 52% (48% on the graph).

Where the sentence highlights differed

On the first essay, Winston marked passages as possibly AI-generated despite giving the document its highest human score, while on the fourth it highlighted whole paragraphs that Turnitin had only marked in part. Academi.cx generally followed Turnitin's highlights more closely on those essays, although it still added some extra sentences and left part of a conclusion unflagged.

Average deviation from Turnitin: Academi.cx 7.0 percentage points; Winston AI 17.3 percentage points. Paired rounded bars share a common zero and scale. Documents 1, 4, and 5; displayed score gaps, using 100 minus Winston Human Score.

The limits of the percentage comparison

Turnitin withheld the exact scores and highlights on two essays, leaving three documents in the paired average. On the final essay, Winston came closer to Turnitin, while Academi.cx's prediction was too high, with additional highlighted sentences and part of a paragraph pushing its score above Turnitin's and leaving the actual result below the predicted range.

The final scorecard

Academi.cx generally matched Turnitin's sentence highlights more closely on the essays where both reports were available, although it also flagged extra text and overestimated the final result. Winston picked up some of the same passages, but its broader paragraph highlights were less precise about which sentences Turnitin had marked. Across the three essays with exact Turnitin scores, Academi.cx had the lower average deviation, while Winston came closer on the final essay.

Related reading