Turnitin vs Copyleaks: the two detectors institutions buy

Why people compare these two

This is the comparison a university procurement committee actually makes. Almost every other pairing in this category is between consumer tools an individual picks; these two are bought by institutions, priced per student, and arrive integrated into the systems a campus already runs. If AI detection is happening to your work at scale, one of these two is overwhelmingly likely to be doing it.

It is also the pair where the evidence is most often misrepresented. Turnitin scored highest of fourteen tools in one study; Copyleaks scored highest of seven in another. Both facts get quoted as though they settle the question, and they cannot, because the two studies share no tools and no test corpus.

A note on who this page is for, because the two audiences want opposite things from it. An administrator is choosing, and needs the procurement considerations. A student is not choosing — they have been measured by whichever one their institution bought, and need to know what that result is worth. Both are below, and the second is the reason this page spends as long on appeals and data handling as on the comparison itself.

Side by side

No prices. Neither of these is a tool most readers choose on cost, and for one of them there is no public price at all — detector pricing, where it could be captured, is on our detector pricing page.

Every line above is summarised from our own examination of each tool, where the studies and vendor documents behind it are quoted and linked: Turnitin and Copyleaks.

What actually separates them

Start with what a student can do. Turnitin sells to institutions only and publishes no self-serve checker, so unless your institution has enabled the indicator for your class you cannot see your own score at all — and third-party services claiming to show it are showing you a different tool's opinion. Copyleaks sells to individuals as well as institutions, at $16.99 a month or $13.99 billed annually, so the same tool your university may be running is one you can also buy.

Then look at what happens to the text. This is the sharpest difference on the page and it is rarely part of the comparison. Copyleaks's paid plans include the Shared Data Hub, which it describes as a library of user-submitted documents that scans are compared against, and its privacy policy states it uses this information to train its models, with opt-out available to direct customers by contacting support. Turnitin's retention is set by the institution's configuration, which the student does not control either — but the default destination is a repository rather than a training corpus.

The handling of uncertainty differs too, and Turnitin's approach is the more unusual. Between 0% and 20% it attributes no score and shows an asterisk instead, because its own testing found more false positives in that band. That is a vendor deliberately withholding a number it could show, which is close to unique in this category. Copyleaks publishes 99.97% accuracy and a 0.026% false-positive rate for its V10 model with no equivalent suppression band.

On languages, Turnitin's AI indicator covers English, Spanish and Japanese, and needs 300 to 30,000 words of prose in .docx, .pdf, .txt or .rtf under 100MB. Copyleaks publishes no equivalent word floor and instead exposes sensitivity settings, the highest of which is described as designed to flag text run through a humanizer or spinner.

The integration story is the part procurement usually underweights and lives with longest. Turnitin's indicator appears inside a marking workflow instructors already use, so adoption requires no new habit and no training budget — which is also why it gets read by people who have never been told what it measures. Copyleaks arrives as a deliberate addition, which costs more to roll out and means somebody had to decide what the institution would do with the output.

What the evidence supports, and what it does not

Turnitin's result comes from Weber-Wulff et al. (2023), which tested fourteen tools across 756 tests and concluded that detection tools “are neither accurate nor reliable”, with every tool below 80% accuracy. Turnitin was the best of that fourteen. Copyleaks was not in it.

Copyleaks's result comes from Perkins et al. (2024), which tested seven tools. Copyleaks was the best of those seven — and in the same study it failed to identify 39% of the AI-generated cases and produced the highest false-accusation rate of the seven. Under simple adversarial conditions its accuracy fell from 73.9% to 58.7%. Turnitin was not in that study in the same configuration.

So the honest summary is that both sit at the better end of a field where the better end still means missing a third or more of AI text, and no measurement exists that puts them on the same scale. A procurement decision between them cannot be made on accuracy evidence, because there is none comparing them.

Both vendors' self-published figures deserve the same treatment, and it is worth applying it symmetrically rather than only to the one you like less. Copyleaks's 99.97% accuracy and Turnitin's sub-1% false-positive aim are each unverified by anyone but the company selling the product, neither is accompanied by a published corpus, and the gap between the vendor number and the independent one is roughly forty points on the only side where both exist.

Where each one came from

Turnitin's AI indicator is a feature bolted onto a business that already existed. The company sold similarity checking to institutions for two decades before generative AI, which means the indicator arrived inside a product universities had already bought, integrated into workflows already running, with a sales relationship already in place. Nobody made a purchasing decision about AI detection; it appeared in a tool they were using for something else.

That history explains the most unusual thing about the product — the suppression band. A company selling a new category has every incentive to show a number; a company protecting a twenty-year institutional relationship has an incentive not to show one it cannot stand behind. Turnitin attributes no score between 0% and 20% because its own testing found more false positives there, and that decision is easier to make when the indicator is not what you are selling.

Copyleaks came up as a plagiarism and content-integrity company serving enterprises as well as education, and its AI detection is closer to the centre of what it sells. The consequences are visible in the product: sensitivity settings the customer chooses, a documented position on humanized text, and the Shared Data Hub — a corpus built from customer submissions, which is a business model as much as a feature.

If one of these has produced a result about you

With either of these, you are inside an institutional process rather than a conversation with a vendor, which is better than it sounds. There is a policy, there is normally an appeals route, and there is someone whose job includes getting this right. Ask for the policy in writing before you argue about the score.

For Turnitin specifically, two facts are worth knowing and are Turnitin's own. It states the indicator is not intended as the sole basis for an academic misconduct finding, and it suppresses scores below 20% because of false positives in that band. If your report shows an asterisk rather than a percentage, that is the tool declining to give a number — it is not a low score, and it should not be presented to you as one.

For Copyleaks, ask which sensitivity setting was used. Its highest setting is explicitly designed to flag text run through a humanizer or spinner, which means it is tuned to accept more false positives in exchange for catching more disguised text. An institution that has turned that on has made a policy choice about which error it prefers, and you are entitled to know it was made.

In both cases the evidence that actually decides an appeal is your drafting record. Version history in Google Docs or Word, notes, outlines, the reading you did — and the guide to proving you wrote it is linked under Related below. A percentage is an input to a human decision, and drafting history is the thing that answers it.

What we could not establish

The central question — which of these two is more accurate — has no answer, and it is worth being blunt that this is not modesty. Turnitin scored highest of fourteen tools in Weber-Wulff et al. (2023) and Copyleaks scored highest of seven in Perkins et al. (2024). The two studies share no tools, no corpus and no year. Combining them produces a ranking that looks like evidence and is not.

Turnitin publishes no false-positive rate for the band it does report on beyond an aim of under 1%, and no independent test has verified that figure. Copyleaks publishes 99.97% accuracy and a 0.026% false-positive rate for its V10 model, which no independent test has verified either.

Turnitin's pricing is not public and it returns 403 to automated requests, so no figure for it appears anywhere on this site. What an institution actually pays is negotiated per contract and we have not seen one.

We could not establish how long Turnitin retains submissions, because that is set by each institution's configuration rather than by Turnitin. Your registry can answer it; we cannot.

Which to pick

Pick Turnitin if you are an institution that wants the tool with the widest peer-reviewed testing behind it and a vendor willing to withhold a score it does not trust. The asterisk band below 20% is a real feature, not a limitation.

Pick Copyleaks if you need the same tool available to individuals as to the institution, you want a documented position on adversarial and humanized text, or you need coverage beyond Turnitin's three languages — provided you have read what the Shared Data Hub does with what you submit.

Turnitin vs Copyleaks: common questions

Which is more accurate, Turnitin or Copyleaks?

Nobody knows, and anyone who tells you otherwise is extrapolating. Turnitin was the best of fourteen tools in Weber-Wulff et al. (2023); Copyleaks was the best of seven in Perkins et al. (2024). The two studies used different tools, different corpora and different years, and neither included the other's winner. There is no test that ranks these two against each other.

Can I check my own work in Turnitin or Copyleaks before submitting?

Copyleaks, yes — it sells to individuals from $16.99 a month. Turnitin, only if your institution has enabled a self-check facility such as Draft Coach for your class. Third-party sites offering to show you your Turnitin score are showing you a different detector's output, which is a rough signal and not the number your instructor will see.

Does Copyleaks keep the documents I submit?

Yes, and it is a product feature rather than a footnote. Paid plans include the Shared Data Hub, which Copyleaks describes as a library of user-submitted documents that scans are compared against, so a document you submit can become part of a corpus other customers are checked against. Its privacy policy also states it uses this information to train its models, with opt-out available to direct customers by contacting support.

Why does my Turnitin report show an asterisk instead of a percentage?

Because the detected share fell between 0% and 20%. Turnitin suppresses the figure in that band on purpose — its own testing found a higher incidence of false positives at low percentages, so it shows an asterisk rather than a number that would read as more certain than the evidence supports.

Do both detect text that has been through a humanizer?

Both claim to and neither should be relied on for it. Copyleaks's highest sensitivity setting is explicitly described as aimed at humanized and spun text, though independent testing found its accuracy fell to 58.7% under simple adversarial techniques. Turnitin announced humanizer detection in August 2025, a claim that still awaits independent verification. Across fourteen detectors, Weber-Wulff and colleagues measured 26% accuracy on machine-paraphrased text.

Which one do most universities use?

Turnitin has by far the larger installed base, largely because it was already in place for similarity checking long before AI detection existed and the indicator arrived inside a product institutions had already bought. Copyleaks is the realistic alternative and integrates with Canvas, Moodle, D2L Brightspace, Blackboard, Schoology, Sakai and Edsby.