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.
People compare these two to answer one question — can I use the free one to predict the paid one? No. They are different classifiers, and only one of them has ever been tested by anyone but its maker.
Last reviewed 16 August 2026 · The HumanFlow team
This is the single most common detector comparison a student actually makes, and it is almost always made for the same reason: Turnitin is what will run on the submission, Scribbr is free and available right now, and the hope is that the second predicts the first. It does not, and understanding why is more useful than any ranking.
It is also a pair where the free option is unusually generous — unlimited checks, no sign-up, 1,200 words a submission — which makes the temptation to treat it as a dry run stronger than it is with tools that meter you.
It is worth being direct about what brings most people here, because the answer shapes everything below. It is not procurement. It is a student with a deadline wondering whether a free check will tell them what the institutional one will say. The answer is no, and the rest of this page is about why that is a property of how these systems work rather than a shortcoming of the free tool.
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 | Scribbr | Turnitin |
|---|---|---|
| Who it is sold to | Consumers, free. Owned by Learneo, Inc., which also owns QuillBot, Course Hero, CliffsNotes, LitCharts, Symbolab, Bartleby and LanguageTool. | Institutions only. A student or an individual instructor cannot buy it; it arrives inside the systems a university already licenses. |
| How you get at it | Unlimited checks, no sign-up, capped at 1,200 words per submission. | No public checker. If your institution has not enabled the indicator for your class, there is no way to see your own score. |
| What it needs | 1,200 words per submission on the free detector. | At least 300 words of prose, up to 30,000. .docx, .pdf, .txt or .rtf under 100MB, in English, Spanish or Japanese. |
| What independent testing found | None. It was not among the fourteen tools in Weber-Wulff et al. (2023), the seven in Perkins et al. (2024), the seven in Liang et al. (2023), or the four in the 2025 NBER working paper. Every public accuracy figure for it comes from its own testing of its own product. | 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. |
| What the vendor claims | 78% and 84% on its own thirty-text benchmark, in which its own brands took the top three places. | 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. |
| What the vendor says about its limits | Its premium tier is described as adding GPT-4 detection, with the free detector handling GPT-2, GPT-3 and GPT-3.5 at “average accuracy”. | States its indicator is not intended as the sole basis for an academic misconduct finding. |
| What happens to your text | Its detector page states “we do not store or share your data”. We found no statement there about training use in either direction, so treat that as unstated rather than denied. | Submissions may be retained in its repository depending on the institution's configuration, which the student does not set. |
Every line above is summarised from our own examination of each tool, where the studies and vendor documents behind it are quoted and linked: Scribbr and Turnitin.
They answer to different people, which shapes everything else. Turnitin sells to institutions only; there is no public checker, and unless your institution has enabled the indicator for your class there is no way to see your own score. Scribbr sells nothing to run its detector — it is a free tool on a site whose business is academic-writing services, and it exists to bring people to that site.
The evidence behind them is not comparable in kind. Turnitin scored highest of the fourteen tools in Weber-Wulff et al. (2023), the largest peer-reviewed multi-tool evaluation there is. Scribbr appears in none of the studies this site tracks — not that one, not Perkins et al. (2024), not Liang et al. (2023), not the 2025 NBER working paper. Every public accuracy figure for Scribbr's detector comes from Scribbr testing Scribbr.
Their handling of the uncertain middle is opposite. Turnitin attributes no score at all between 0% and 20%, showing an asterisk, because its own testing found more false positives there — a vendor withholding a number it could display. Scribbr's published position is about model coverage instead: its free detector handles GPT-2, GPT-3 and GPT-3.5 at what it calls average accuracy, with GPT-4 detection reserved for premium. That is a statement about what the free tool cannot see, and it is easy to miss.
The mechanical limits differ too and matter for planning. Turnitin needs 300 to 30,000 words of prose in .docx, .pdf, .txt or .rtf under 100MB, in English, Spanish or Japanese. Scribbr caps at 1,200 words per submission, which is shorter than most of the essays people want to check, so a long piece goes through in fragments — and a fragment is exactly the condition under which every detector in this category is least reliable.
The 1,200-word cap deserves more attention than it usually gets, because it quietly changes what the free tool can tell you. A 2,000-word essay has to be split, and detectors are least reliable on shorter passages — so the free check is performed under weaker conditions than the institutional scan it is being used to anticipate. That makes the comparison worse than merely imprecise; it is systematically biased toward the tool's weak spot.
The prediction question deserves a direct answer, because it is what brings most readers here. Detectors disagree with each other on identical text, routinely and by wide margins. Two tools trained on different data with different thresholds will not converge, so a clean Scribbr result is not evidence about what Turnitin will report, and a Scribbr flag is not a reason to rewrite work you wrote yourself.
Scribbr's own ranking of AI detectors is worth reading with its ownership in view. Scribbr and QuillBot are both owned by Learneo, Inc., and in Scribbr's comparison those two brands take the top three positions, with no relationship disclosed. In that test GPTZero scored 52% and Copyleaks 66% — figures that sit oddly beside peer-reviewed work in which Copyleaks was best of seven and Turnitin best of fourteen. The sample was thirty texts of 1,000–1,500 characters, so each one moves the score by 3.3 points.
None of which makes Turnitin's result strong in absolute terms. Weber-Wulff and colleagues concluded that detection tools “are neither accurate nor reliable”, with every tool below 80% accuracy. Best of that fourteen is a real finding about a field the authors judged unfit for the purpose it is being put to.
The absence of independent testing is worth sitting with rather than noting in passing. Sixteen detectors have been evaluated by researchers with no commercial interest in the result. Scribbr, despite being one of the most-used free detectors in academic circles and appearing in countless student recommendations, is not among them — so its reputation rests entirely on distribution and on its owner's own benchmark.
Scribbr sells academic editing, proofreading and citation help to students, and its free detector is a door into that business. Understanding this resolves most of what is otherwise puzzling about the product — why it is unlimited and requires no account, why it caps each submission at 1,200 words rather than metering you monthly, and why it sits inside a large library of academic-writing guidance. It is designed to be used repeatedly by someone who may later buy proofreading.
Turnitin sells to the institution that will assess that student, and has done since long before generative AI. Its indicator is a feature inside a similarity-checking product universities had already bought, which is why no purchasing decision about AI detection was ever made by most of the institutions now running one.
The asymmetry that follows is the reason this comparison exists at all. One of these tools is optimised to be run by you, as often as you like, for free. The other is the one that actually reports on you, and you cannot run it. That is a difference in access rather than in quality, and it is what makes the free tool feel like a preview when it is nothing of the kind.
If a Turnitin report is the accusation, nothing Scribbr told you beforehand is relevant to it, and it is worth setting that aside early rather than building an argument on it. They are different classifiers trained on different data with different thresholds, and a clean result from one says nothing about the other.
Ask which band the score falls in. Turnitin attributes no score between 0% and 20% and shows an asterisk instead, because its own testing found more false positives at low percentages. An asterisk is the tool withholding a number, not the tool reporting a small one, and it is regularly presented to students as though it were the latter.
Then produce the drafting record, which is what actually answers the question being asked. Version history, notes, outlines, the sources you opened, the search history if you have it. Turnitin itself states the indicator is not intended as the sole basis for a misconduct finding — so the decision is meant to rest on evidence of this kind, and providing it is cooperating with the process rather than resisting it.
If you are tempted to pre-check future work through Scribbr to feel safer, be clear about what that buys. It is a reading on how conventional and predictable your prose is, which is worth something as writing feedback and nothing as a prediction of an institutional result.
Nobody independent has ever tested Scribbr's detector. It appears in none of the four studies this site tracks — not Weber-Wulff et al. (2023), not Perkins et al. (2024), not Liang et al. (2023), not the 2025 NBER working paper. Every public accuracy figure for it, including the 78% and 84% quoted widely across the web, comes from Scribbr testing Scribbr.
We cannot say how closely a Scribbr result tracks a Turnitin one, because no study has compared them and we have not run that test ourselves. The advice on this page not to treat one as a preview of the other rests on how these classifiers work rather than on a measurement of this specific pair, and that distinction is worth being honest about.
Turnitin publishes no false-positive rate beyond an aim of under 1% above the 20% threshold, and no independent study has verified it.
Scribbr's detector page states that it does not store or share your data. We found no statement there about whether submissions are used to train or improve its models, in either direction, so treat that as unstated rather than as a denial.
Pick Scribbr if you want a free, unlimited, no-account read on a passage and you will treat it as a rough signal about how conventional your prose looks — which is genuinely what it measures.
Pick Turnitin if the question is what your institution will see, in which case there is nothing to pick: Turnitin is what runs, you cannot access it yourself, and preparing a drafting record is more use than any pre-check.
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.