HumanFlow is not one of the two tools on this page. We sell a rewriter and run a detector of our own, so weigh this accordingly — our methodology sets out how we source what appears below and what we refuse to claim.
Originality.ai's own CEO said in 2024 that the company advised against using it in academia and strongly recommended against disciplinary use. By 2026 it sells an Academic Model for Educators.
Last reviewed 16 August 2026 · The HumanFlow team
Turnitin is what a university already has; Originality.ai is what turns up when someone at that university goes looking for an alternative, usually because Turnitin's indicator is unavailable to them or because a department wants a second opinion. Both are paid, neither is a casual consumer tool, and both are pitched at people who will act on the result.
The pair is also the clearest case in this category of a vendor changing its stated position on what its own product should be used for — which is worth knowing before you buy either.
One framing to set aside first, because it is how this comparison is usually presented elsewhere: that the newer, cheaper, self-serve tool is the challenger and the incumbent is the safe default. Neither half holds up. Originality.ai is not primarily an academic product and once said so itself, and Turnitin's advantage is a peer-reviewed result inside a study that judged the whole category unreliable. The interesting comparison is about evidence and stated positions rather than about incumbency.
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.
| Dimension | Turnitin | Originality.ai |
|---|---|---|
| Who it is sold to | Institutions only. A student or an individual instructor cannot buy it; it arrives inside the systems a university already licenses. | Publishers, agencies and SEO teams first; education second and recently. Its CEO told Gizmodo in June 2024 that the company advised against academic use; by 2026 it markets an Academic Model for Educators and a Moodle plugin. |
| How you get at it | No public checker. If your institution has not enabled the indicator for your class, there is no way to see your own score. | No free trial. Pro is $14.95 a month, or $12.95 a month billed annually, for 2,000 credits at 100 words each. |
| What it needs | At least 300 words of prose, up to 30,000. .docx, .pdf, .txt or .rtf under 100MB, in English, Spanish or Japanese. | Not published as a word floor. The NBER paper found it struggles on short passages. |
| What independent testing found | Scored highest of the fourteen tools in Weber-Wulff et al. (2023) — in a study whose own conclusion was that detection tools “are neither accurate nor reliable”, with every tool below 80% accuracy. | Ranked second of four in the 2025 NBER working paper, in a “secondary tier” that struggles on short passages and on text run through humanizing tools. It was not among the fourteen tools in Weber-Wulff et al. (2023). |
| What the vendor claims | Aims to keep false positives under 1% above a 20% detected share. Below 20% it attributes no score at all, showing an asterisk, because its own testing found more false positives in that band. | 99%+ accuracy and false-positive rates between 0.5% and 1.5% depending on the model — published with no named corpus, no test-set size and no date. |
| What the vendor says about its limits | States its indicator is not intended as the sole basis for an academic misconduct finding. | Its terms forbid using the output as sole grounds for discipline. |
| What happens to your text | Submissions may be retained in its repository depending on the institution's configuration, which the student does not set. | Its privacy policy states that unless you opt out, it may use your scan history and results to help train and improve its models. No fixed retention period is published. Institutional agreements exclude student data from general-purpose training. |
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 Originality.ai.
The change of position is the fact that should shape a decision here. Jonathan Gillham, Originality.ai's CEO, told Gizmodo in June 2024: “we advise against the tool being used within academia, and strongly recommend against being used for disciplinary action.” His stated reason was the arithmetic — students submit few enough essays that the false-positive risk is unacceptable. That arithmetic has not changed. By 2026 the company markets an Academic Model for Educators and a Moodle plugin, though its terms still forbid using the output as sole grounds for discipline.
Turnitin's position on the same question has been consistent and is built into the product rather than stated in terms of service. It attributes no score at all between 0% and 20%, showing an asterisk instead, because its own testing found more false positives in that band. Above 20% it aims to keep false positives under 1%. A vendor withholding a number it could display is rare enough in this category to be worth naming.
Who can run them differs completely. Turnitin sells to institutions only — no self-serve checker, and if your institution has not enabled the indicator for your class there is no way to see your own score. Originality.ai sells to anyone: Pro is $14.95 a month, or $12.95 billed annually, for 2,000 credits at 100 words each, with no free trial. So a department can pilot Originality.ai without procurement, which is a genuine advantage and also how a detector ends up in use with no institutional policy attached to it.
What each does with the text differs too. Originality.ai's privacy policy states that unless you opt out, it may use your scan history and results to help train and improve its models, with no fixed retention period published; institutional agreements exclude student data from general-purpose training. Turnitin's retention is set by the institution's configuration, which the student does not control either.
The credit model is a practical difference that shapes behaviour rather than just cost. Originality.ai meters in credits at 100 words each, so scanning is a decision with a running total attached, and organisations ration it toward what they suspect. Turnitin's indicator runs on every submission as part of a workflow already paid for, which means it is applied indiscriminately — and a tool applied to everybody generates its false positives across everybody.
Turnitin scored highest of the fourteen tools in Weber-Wulff et al. (2023) — the largest peer-reviewed multi-tool evaluation there is, and one whose own conclusion was that detection tools “are neither accurate nor reliable”, with every tool below 80% accuracy. Originality.ai was not in that study.
Originality.ai's independent showing comes from the 2025 NBER working paper, which ranked it second of four and placed it in a “secondary tier” that struggles on short passages and on text run through humanizing tools. That paper is not peer-reviewed and says so.
Its own published figures are 99%+ accuracy and false-positive rates between 0.5% and 1.5% depending on the model. The conditions attached are thin: no named corpus, no test-set size, no date. A figure without those is not checkable, and this site treats it as a claim rather than a measurement — the same standard applied to every vendor here.
One further claim is worth correcting because Originality.ai's marketing leans on it. Google has said directly that it does not penalise content for being AI-generated: a spokesperson told Gizmodo it is “inaccurate to say Google penalizes websites simply because they may use some AI-generated content.” Google's position is that low-value content produced at scale to manipulate rankings is spam however it was produced — a claim about value and intent, not about machines.
Applying the CEO's own arithmetic to the academic case is worth doing explicitly, because it is the clearest argument on this page and it is his rather than ours. Take the company's published false-positive range of 0.5% to 1.5%. Across a cohort of 400 students submitting one essay each, that is between two and six people wrongly flagged per assignment — which is precisely the calculation he described in 2024 when advising against academic use.
Originality.ai was built for publishers and SEO agencies, not for universities, and its whole design reflects the customer it was built for. Bulk scanning, site-wide scans, a Chrome extension, per-credit pricing measured in words: these are the tools of someone checking a freelancer's invoice-worth of copy, not someone reading one student's essay. Academic use arrived later, as an expansion into a market with different stakes.
That expansion is documented in an unusually explicit way, which is why it belongs on this page. Its CEO told Gizmodo in June 2024 that the company advised against academic use and strongly recommended against disciplinary use, giving a specific reason: students submit few enough essays that the false-positive risk is unacceptable. By 2026 the company sells an Academic Model for Educators and a Moodle plugin. The arithmetic he described has not changed.
Turnitin's position is the mirror image. It has only ever sold to institutions, its indicator arrived inside a product universities already ran, and its most distinctive design decision is a refusal — no score at all between 0% and 20%, because its own testing found more false positives there. One company built for publishers and moved toward education; the other has never sold to anyone else.
If the result came from Originality.ai in an academic setting, the company's own terms are the most useful document you can bring. They forbid using its output as sole grounds for discipline. That is not a student's argument about detector reliability; it is the vendor's licence terms, and an institution acting against them is acting outside the conditions it agreed to.
The 2024 Gizmodo quote is worth having alongside it, and worth using carefully. The point is not that the company is untrustworthy for changing its position — companies are allowed to change positions. The point is the reason it gave, which was about the mathematics of applying a percentage false-positive rate to a population submitting a handful of essays each. That reasoning applies to your case regardless of what the company now sells.
If the result came from Turnitin, start with which band it falls in. An asterisk means the detected share was between 0% and 20% and Turnitin has declined to give a number — that is not a low score, and it should not be characterised as one. Above 20%, Turnitin states it aims to keep false positives under 1% and that the indicator is not intended as the sole basis for a misconduct finding.
With either, the evidence that resolves an accusation is a record of how the document was made rather than an argument about the number. Version history, notes, outlines, the sources you actually opened. Both vendors say a score is an input to a human decision; the drafting record is what that human should be deciding on.
No study has tested these two on the same corpus. Turnitin was best of fourteen tools in Weber-Wulff et al. (2023), which did not include Originality.ai; Originality.ai was second of four in the 2025 NBER working paper, which did not include Turnitin. There is no measurement that ranks them against each other and we have not manufactured one.
Originality.ai's published accuracy figures — 99%+ accuracy, false positives between 0.5% and 1.5% — carry no named corpus, no test-set size and no date. That is not a claim that they are wrong. It is a statement that they are not checkable, which is the property this site cares about.
We could not establish how long Originality.ai retains scan history. Its privacy policy states that unless you opt out it may use scan history and results to train and improve its models, and publishes no fixed retention period.
Turnitin's pricing is not public and its documentation returns 403 to us, so nothing on this site quotes a Turnitin price or a Turnitin support article directly.
Pick Turnitin if the setting is academic and a result may lead to a misconduct process. It has the widest peer-reviewed testing behind it, and the 20% suppression band means the tool declines to give a number in exactly the range where it is least trustworthy.
Pick Originality.ai if you are checking published or commissioned content rather than student work — the publisher and agency use case it was actually built for — and you have read its scan-history training clause. Do not buy it for disciplinary use; its own terms forbid that and its CEO once did too.
A detector reports how statistically machine-typical a piece of prose reads. It has no access to how the text was produced, so it cannot establish authorship in either direction — a flag is not evidence of AI use, and a clean result is not a clearance. If you have been accused on the strength of one, drafting history is what actually answers it. And if your institution requires you to disclose AI assistance, disclose it — nothing on this page changes that obligation.