For this benchmark, we submitted the same 129 documents to Academi.cx and Copyleaks through their APIs, then compared each tool’s scores with the documents’ actual Turnitin AI scores.
Copyleaks was only available for 129 of the documents in our full benchmark, since its pricing limits how many checks it can run at once. So this comparison is scoped to that same 129 document set for both tools, an even matchup with nothing cherry picked on either side. Those 129 documents add up to about 237,000 words total, averaging around 1,835 words each, ranging from short 350 word essays up to a 20,000 word thesis.
How close were the predictions
Across those 129 documents, Academi.cx landed within 1.45 points of the real Turnitin score on average. Copyleaks averaged 8.23 points off. The median error tells a similar story, Academi.cx sits at 0, meaning more than half the time it hit the Turnitin number exactly.
Breaking it down by document type
The 129 documents were split into three groups: 80 that Turnitin scored as 0% AI, 26 that scored as 100% AI, and 23 that fell somewhere in between.
On the clean 0% documents, Academi.cx averaged just 0.16 points of error, while Copyleaks averaged 5.71. On the fully AI written 100% documents, Academi.cx averaged 2.96 points off versus Copyleaks at 7.69. The mixed bucket is where every tool struggles most, since it takes more nuance to score partial AI content correctly. Academi.cx averaged 4.22 points of error there compared to Copyleaks at 17.58, by far the widest gap of the three groups.
False flags and missed AI
Two numbers matter most for anyone actually relying on this to make a judgment call: how often a tool wrongly accuses genuine human writing of being AI, and how often it clears writing that's fully AI generated.
Academi.cx had a 0% false flag rate and a 0% missed AI rate across this set, meaning it never scored a genuinely 0% document above 20%, and never scored a genuinely 100% document below 80%. Copyleaks came in at an 8.8% false flag rate and a 7.7% missed AI rate on the same documents.
Sentence level accuracy
Beyond the overall percentage, we also checked how often each tool correctly classified individual sentences as AI written or human written, matched against Turnitin's own sentence level calls. On this same 129 document set, Academi.cx correctly classified 98.17% of sentences. Copyleaks came in at 85.16%.
That gap matters because a document level percentage can be right for the wrong reasons. Getting the overall score close while flagging the wrong sentences still leaves a writer disputing highlights that don't actually match what Turnitin flagged in the first place.
The takeaway
Across 129 real documents scored against actual Turnitin results, Academi.cx came closer to Turnitin’s scores on every measure we checked: average error, consistency, false flags, missed AI, and sentence level precision. Copyleaks performed reasonably on the easy extremes but fell noticeably further from the actual Turnitin percentage as documents moved into the mixed range, which is where accuracy actually gets tested.
See the full benchmark results.
