What it is
Originality.ai launched in November 2022 — before ChatGPT’s public release — and describes itself as an AI and plagiarism detector for “serious content publishers”. Its primary market is SEO agencies, publishers and brands checking freelance work at scale. Education is a secondary and, until recently, disavowed motion. The company is Originality AI Inc., based in Collingwood, Ontario; the founder and CEO is Jonathan Gillham.
The reversal on academic use
This is the part worth knowing before an instructor runs your essay through it.
In June 2024, Gillham told Gizmodo: “we advise against the tool being used within academia, and strongly recommend against being used for disciplinary action.” His reasoning was statistical and, we think, correct: a student submits a handful of essays a year, so a single false positive is catastrophic, whereas a professional writer producing high volume gives the algorithm more chances to be right on balance.
By 2026 the company markets an “Academic Model for Educators” claiming 99%+ accuracy and under 1% false positives, alongside a Moodle plugin and institutional data handling. Its terms still forbid using the output as sole grounds for discipline.
We are not calling that dishonest — models improve, and the claimed false-positive rate did fall over that period. But if you are an educator weighing this tool, the strongest argument ever made against using it in your setting came from the person who sells it, and it was made on the basis of maths that has not changed.
What happened to writers it got wrong
The same Gizmodo investigation followed two freelancers. Kimberly Gasuras, a news reporter of 24 years in Bucyrus, Ohio, was flagged and later suspended from a writing platform for “excessive use of AI”. She does not use it.
The second case matters for anyone relying on draft history as proof. A copywriter the piece calls Mark was flagged, produced timestamped documentation showing he wrote the article by hand, and it was not enough — he lost roughly 90% of his income. Gillham, to his credit, did not dodge it: “We hear these stories more than we wish we did, and we understand the pain that false positives cause writers when the work they poured their heart and soul into gets falsely accused.”
We tell students that version history is their best evidence, and we stand by that — it is the strongest thing available. Mark’s case is the honest caveat: evidence only works where someone is obliged to weigh it. A commercial platform can drop a freelancer without a hearing. An academic institution generally cannot, which is why the same evidence is worth much more inside a formal process than outside one.
The accuracy claims, and their conditions
Originality publishes per-model figures: Lite at 99% accuracy and 0.5% false positives, Turbo at 99%+ with up to 97% accuracy “on the latest AI humanizers and AI bypassers” and a 1.5% false-positive rate, and an Academic model at 99%+ with under 1%. What these lack is the scaffolding that would let anyone check them: no named corpus, no sample size, no test date.
Gizmodo also caught an arithmetic problem in the 2024 figures — the company claimed 98.8% accuracy and a 2.8% false-positive rate, which sums past 100%. Gillham said the numbers came from two different tests. That is a plausible explanation and also precisely the problem: two numbers from two unnamed tests, presented together as one specification.
Be careful with the independent record here, because it is widely miscited. Originality.ai was not among the fourteen tools in Weber-Wulff et al. (2023), and was not among the seven in Perkins et al. (2024). Any figure from those papers attributed to Originality.ai is wrong. It was one of seven detectors in Liang et al. (2023), but that paper publishes no per-tool breakdown — the well-known 61.22% false-positive rate on TOEFL essays is a seven-detector average and cannot be assigned to any single tool.
Where it was genuinely tested, in the September 2025 NBER working paper (not peer-reviewed, and the paper says so), it placed second of four behind Pangram — solid on long-form text, but in a “secondary tier” that degrades on short passages and against humanizing tools.
What it costs, and what it keeps
No free trial. Pro is $14.95/month, or $12.95/month billed annually, for 2,000 credits; Enterprise is $179/month, or $136.58 annually, for 15,000. One credit buys 100 words, so Pro covers 200,000 words a month. Monthly credits expire monthly.
On data: the privacy policy states that “unless you opt out, we may use your scan history and scan results to help train and improve our models,” with no published retention period. If you are pasting unpublished client work or an unsubmitted dissertation into it, that is the sentence to read twice.
The Google claim
Originality’s marketing has led with “future proof your site on Google”. Google rejected that reading directly. Its spokesperson told Gizmodo: “It’s inaccurate to say Google penalizes websites simply because they may use some AI-generated content,” adding that low-value content produced at scale to manipulate rankings is spam “however it is produced”.
That distinction is the whole thing, and it is the same one we make about our own product: the objection is to content with no value, not to the involvement of a machine.
Where we stand
We sell a humanizer and a detector, so we are an interested party. We have published no measured pass rate against Originality.ai and will not claim one until our benchmark produces it. To their credit, Originality’s own homepage tells buyers to “use AI detection as one signal, not a final decision,” and its education page says a detection score should not be the only measure used to identify cheating. We agree with both sentences, and would rather quote them than caricature the company.
Compared against
- Turnitin vs Originality.ai — Originality.ai against Turnitin on what independent testing found, who each is sold to, and what happens to the text you submit.
- Pangram vs Originality.ai — Originality.ai against Pangram on what independent testing found, who each is sold to, and what happens to the text you submit.
- GPTZero vs Originality.ai — Originality.ai against GPTZero on what independent testing found, who each is sold to, and what happens to the text you submit.
Related
- Our fuller review of Originality.ai — the product beyond the detector: bulk scanning, the Chrome extension and who it is actually built for.
- Compilatio, reviewed — a vendor figure and an independent one, twenty points apart
- Scribbr, reviewed
- GPTZero, reviewed
- How accurate AI detectors are
- Falsely accused of using AI?