Quick answer. Originality.ai says its newest detector models (5.0 and 5.0 Max, October 2026) reach 99%+ accuracy, and its launch post reports its own benchmark results with sample sizes. The two independent tests we could read were run before those models existed and give lower numbers for the tool as it stood then. Scribbr (published 26 February 2024) scored Originality.AI at 76% accuracy, with one false positive, on 30 short texts. A Stanford-led study tested seven detectors, Originality.AI among them, in March 2023: on average they flagged 61.22% of 91 TOEFL essays written by non-native English speakers as AI-generated. Neither test used résumés, so neither gives an error rate for a résumé, and we did not find an independent test of the 5.0 models. We found no page from an applicant tracking system vendor that says it runs Originality.ai on incoming résumés.
What Originality.ai says about its own accuracy
The vendor page we checked is originality.ai/blog/ai-accuracy. We read it on 4 November 2026; it is titled “Why AI Detection Matters and How to Measure It” and its metadata gives 20 October 2026 as the last modification. It says:
- “Originality.ai 5.0 and 5.0 Max have 99%+ accuracy and set the new standard for AI detection.” Its release history gives the same “99%+ accuracy” for earlier models from October 2024 on.
- For an in-depth look at 5.0 and 5.0 Max it links to its launch post (metadata date 17 October 2026). That post reports the vendor’s own results on named public datasets, with sample sizes. Examples: human-text specificity of 99.954% for 5.0 on 25,996 PERSUADE 2.0 essays; 98.901% for 5.0 and 100% for 5.0 Max on a 91-essay benchmark it labels “Liang TOEFL”; AI-text recall of 97.69% (5.0) and 98.00% (5.0 Max) on 1,600 UChicago Booth AI texts. On 1,000 mixed human-and-AI samples it reports that 5.0 labelled 82.9% as mixed, 15.2% as human and 1.9% as AI.
- The launch post adds its own caveat: “Leading AI detectors are highly accurate but not perfect on human and AI content.” It also says its comparison results “come from separate evaluations with different samples and scoring rules; they are not a matched head-to-head test.”
These are the vendor’s own figures and we have not reproduced them. The release history on the same page lists models from April 2023, August 2023, February 2024, October 2024 and later, so a test from 2023 or early 2024 can describe a different model from the one advertised today. Scribbr’s page does not say which model it tested, so we have not tried to match each test below to a model version.
What the independent tests found
The Stanford-led study (Liang et al.)
Liang, Yuksekgonul, Mao, Wu and Zou published GPT detectors are biased against non-native English writers in Patterns (Cell Press) on 10 July 2023; Stanford HAI summarised it on 15 May 2023. The paper lists the seven detectors it tested, accessed on 15 March 2023: Originality.AI, Quil.org, Sapling, OpenAI, Crossplag, GPTZero and ZeroGPT. The texts were 91 TOEFL essays from a Chinese educational forum and 88 US eighth-grade essays from the Hewlett Foundation’s ASAP dataset.
- The paper says the detectors “demonstrated near-perfect accuracy for US 8-th grade essays” but “misclassified over half of the TOEFL essays” as AI-generated (average false positive rate: 61.22%).
- “All seven detectors unanimously identified 18 of the 91 TOEFL essays (19.78%) as AI-authored, while 89 of the 91 TOEFL essays (97.80%) are flagged as AI-generated by at least one detector.”
- When ChatGPT was asked to enhance the word choices of the TOEFL essays, the average false positive rate fell from 61.22% to 11.77%. When it was asked to simplify the word choices of the US eighth-grade essays as if written by a non-native speaker, the average misclassification rate rose “from an average of 5.19% across detectors to 56.65%”.
The 61.22% is an average over all seven detectors; the text we read gives no separate figure for Originality.AI. The authors link the pattern to perplexity: the TOEFL essays that all seven detectors flagged “show significantly lower perplexity compared to others, suggesting that GPT detectors might penalize authors with limited linguistic expression.” The texts were essays, not résumés.
Scribbr’s comparison
Scribbr’s page “Best AI Detector | Free & Premium Tools Compared” was published on 26 February 2024 and revised on 18 November 2024. It ran 12 detectors on 30 texts: six categories of five texts each (human-written, GPT-3.5, GPT-4, human text combined with GPT-3.5, paraphrased GPT-3.5, paraphrased human text), each 1,000 to 1,500 characters long. Each scan scored 1 point if the label was within 15% of the right answer, 0.5 if within 40%, and 0 otherwise. Its table gives Originality.AI 76% accuracy with 1 false positive, GPTZero 52% with 1, and Scribbr’s own premium detector 84% with none. Its conclusion: “The companies behind some tools make strong claims about their reliability, but those claims are not supported by our testing.” It also says “AI texts that have been combined with human text or paraphrased are hard to detect.”
Two cautions. Scribbr sells its own AI detector, and that detector tops its table. And 30 texts is a small set; the page does not say which Originality.AI model version it tested. We read the figures on a copy of the page as revised on 18 November 2024, so the live page may differ.
What this does and does not show about your résumé
Both independent tests used prose of essay length, not résumés, so neither tells you how a detector behaves on a résumé. The vendor’s own figures for 5.0 on its “Liang TOEFL” benchmark (98.901% of the 91 essays labelled human, 100% for 5.0 Max) are far from the 2023 seven-detector average of 61.22% flagged; these sources do not let us say how much of that difference is a better model and how much is a different test. In our view, short formulaic résumé bullets may be hard for a detector to judge, but that is our hypothesis, not a measured result. What the independent sources do support is narrower. Scribbr concludes that AI detectors “shouldn’t be treated as absolute proof that a text is AI-generated, but they can provide an indication in combination with other evidence.” The Stanford-led authors “caution against their use in evaluative or educational settings, particularly when they may inadvertently penalize or exclude non-native English speakers from the global discourse.”
Do applicant tracking systems run Originality.ai?
We found no Workday, Greenhouse, iCIMS or Lever page that says it runs Originality.ai or GPTZero on incoming résumés, but we did not check every product page and cannot rule it out for every employer’s set-up. In Enhancv’s interviews with 25 US recruiters (September to October 2025), 23 (92%) said their systems do not auto-reject résumés for formatting, content or design; Enhancv calls the sample intentionally small. That finding is about formatting rules, not about AI detectors. A recruiter could also paste résumé text into a public detector by hand; we have no data on how often that happens. See our 2026 AI résumé statistics hub for these figures and their sources.
What the 2025 surveys say about AI-written résumés
The four figures below come from résumé-builder companies’ own surveys. Each is the company’s statement.
- 74% of hiring managers have encountered AI-generated content in applications (Resume Genius, Pollfish survey, sample of 1,000, launched 8 Jan 2025).
- 62% of HR workers say AI-generated résumés without customization are more likely to be rejected (Resume Now, n=925 American HR workers, survey of 28 March 2025; an opinion, not a measured rate).
- 78% of hiring managers say they look for personalized details as a sign of genuine interest and fit (Resume Now, same survey; also a stated opinion).
- Resume.io says that after surveying 3,000 hiring managers, “nearly half (49%) of AI-generated resumes are automatically dismissed” (page last updated 22 Jan 2025; no survey dates or method given).
Nothing in these figures says recruiters used a detector. In our view they point to generic wording as something recruiters react to, but that is our reading, not a finding of any survey.
What we would do (JobCannon view)
The three points below are our advice, not findings from the sources above.
1. Use AI to tighten wording you wrote yourself
The NBER abstract for Working Paper 30886 (Wiles, Munyikwa and Horton, 2023) says that in a field experiment with nearly half a million jobseekers, treated jobseekers “received algorithmic writing assistance on their resumes” and “were hired 8% more often”. The paper reports 480,948 jobseekers, and those hired earned 10% higher hourly wages. The hiring effect for jobseekers from countries where English is not the native language was not statistically different from that for jobseekers from English-speaking countries. The study measured hiring, not detector flags. In our view, the safest use of AI is the one that study tested: help with the wording of a résumé you wrote. Do not hand a model a job description and ask it to invent your work history.
2. Be specific
If, as 78% of the hiring managers in Resume Now’s survey say, recruiters look for personalized details, then named tools, named projects and outcomes you can back up are worth including whoever or whatever drafted the sentence.
3. If someone says your résumé was flagged
Ask which tool was used and what it reported. The two tests above are published and you can point to them. Neither used résumés, so they do not prove your résumé was flagged wrongly; the Stanford-led finding concerns essays by non-native English speakers. In our view, offering to show your drafts is a reasonable response.
Bottom line
Originality.ai says its newest models (October 2026) are 99%+ accurate and publishes its own benchmarks. The independent tests we read, both older than those models, found 76% for Originality.AI on 30 texts (Scribbr, 2024) and an average false positive rate of 61.22% across seven detectors on essays by non-native English speakers (Stanford-led study, tested March 2023). Neither used résumés, and we cannot tell which model version Scribbr tested. Whether a given employer runs a detector on résumés is something we found no source for.
For the broader picture on AI in hiring, ATS rejection myths, and bias evidence, see our 2026 AI résumé statistics hub. To map your résumé strengths against real career options, take our free career match assessment; it ranks careers by how closely they match your work interests.
