Turnitin Check Accuracy Study: Testing 1,000 AI and Human Essays Across Turnitin0, Grammarly, Scribbr, and 6 Other Detectors

Most "Turnitin accuracy" content online is written by people who have never seen a Turnitin report. They repeat the same three claims: that detectors are unreliable, that false positives are rampant, and that any humanizer works. Turnitin0's research programme tests those claims directly, at word level, across academic majors.

Word-level accuracy matters because Turnitin does not flag whole documents. It highlights specific sentences and phrases. A tool that gets the document-level verdict right but the sentence-level highlights wrong will still land you in a misconduct meeting.

The Seven Turnitin0 Reports Behind This Article

Read that table twice, because it contains the single most important insight in this article: Turnitin is excellent at catching raw AI generation and excellent at clearing genuine human writing — but it is close to a coin flip on AI-polished human writing.

The 100% accuracy figures on human-written corpora are reassuring. Turnitin0's PLOS study found no false positives across 504 genuine research papers and 18 majors, and the ESL study found the same across 340 essays and 263,329 words. That directly contradicts the popular claim that Turnitin routinely flags international students. On this evidence, it does not — at least not on these corpora.

The generation studies are equally strong. Turnitin caught GPT-5.6-Sol output at 97.88% word-level accuracy, Claude-Fable-5 at 99.01%, and Gemini 3.5 Flash at 98.35%. If you paste raw chatbot output into an essay, Turnitin will find it.

Then comes TT0-2026-0006. Five hundred graduate essays that humans wrote and then ran through an AI polisher produced just 47.54% word-level accuracy. In some majors, accuracy fell to 0%. This is the scenario most students actually face: they wrote the draft themselves, then asked an AI to "improve the flow." Turnitin's model, trained to recognise machine-generated token patterns, struggles when human structure is overlaid with machine phrasing in patches.

What This Means Practically

• Raw AI text: assume detection. Accuracy above 97% across three model families.
• Genuine human text: assume you are safe. Two studies, 844 essays, zero false positives.
• AI-polished human text: assume uncertainty. Roughly even odds at word level.

That third row is why pre-submission checking exists as a category.

Turnitin0 is an independent service and is not affiliated with Turnitin, LLC. That disclosure matters and I will state it plainly: this is a third-party pre-submission service, not an official Turnitin product. What it does provide is the same style of AI and similarity report your instructor sees, delivered as two PDFs you can read before you decide what to change.

The operational details are worth knowing because they shape how you use it:

• Accepted formats:.docx,.pdf, or.txt — English only.
• Word count: greater than 300 and less than 30,000 words.
• File size: under 20 MB.
• Turnaround: under 15 minutes in 98% of cases, most orders within 5–15 minutes, with rare queue spikes guaranteed within 30 minutes.
• Privacy: non-repository. Your file is not added to Turnitin's student paper database, reports are not shared with third-party databases, and you can delete your files.
• No subscription required.

One honest limitation: when Turnitin's AI detection falls below its 20% confidence threshold, the report displays *% rather than an exact figure. That is Turnitin's behaviour, not Turnitin0's, and it is documented in the research materials. It means a "clean" report tells you that you are under the threshold, not that you scored precisely zero.

Reading the Similarity Report Correctly

The similarity score is the percentage of text matching other sources — nothing more. Turnitin's own guidance is explicit that the tool does not determine plagiarism; it highlights matches for a human to review [1]. National University's library guide makes the same point: the score is almost never zero, because titles and references match by default, and what matters is the nature of the match, not the number [2].

A Turnitin similarity checker report is therefore a diagnostic, not a verdict. Turnitin0's PDFs give you the same diagnostic your instructor will open.

The AI Humanizer Question

If TT0-2026-0006 shows Turnitin struggles with polished text, TT0-2026-0009 shows what happens when you polish deliberately. Turnitin0 humanized 174 GPT-5.6-Sol essays — 204,736 words — and measured a 76.44% overall word-level evasion rate. Education reached 100%. English was the weakest domain at 55.41%.

Those are real, published, unflattering-in-places numbers. A 55.41% evasion rate in English means roughly two in five flagged words survived. Anyone promising universal bypass is not reading their own data.

Turnitin0's AI humanizer accepts.docx or.txt, English only, files under 90 MB, and is designed for text drafted with ChatGPT, Claude, or Gemini. It preserves meaning, citations, headings, and.docx formatting. The score promise is specific: lower the Turnitin AI score to *% or below 20%, or even 0%, or a full refund. Turnitin0 reports that 98.2% of humanizer orders are re-checked with Turnitin.

Honest Limitations of the Humanizer

• No free word quota and no free trial.
• Domain results vary widely — do not assume English performs like Education.
• The *% threshold display applies here too.
• Trustpilot shows a TrustScore of 4.3/5 from 9 reviews, all within the last 12 months, and Trustpilot notes the company has not recently invited customers, so those reviews may not be representative. The 4.9/5.0 satisfaction rating on Turnitin0's homepage is a separate, self-reported figure.

I would rather you know all of that than discover it later.

What Real Users Report

The review sample is small but geographically spread, and the pattern is consistent: speed and completeness.

• Raini Dipré (CA), 5 stars: described the process as easy, fast, and efficient, with the report arriving faster than expected.
• may zin (SG), 4 stars: report complete after about 20 minutes, with AI and similarity reports downloadable together.
• daniela pellegrini (GB), 5 stars: a repeat user who found reports delivered quickly and the Humanize feature helpful when revising.
• b c (US), 5 stars: used the site for assignments, plagiarism checking, and general awareness of AI.
• Shawn Thakur (AU), 5 stars: easy to use and arrived on time.
• Encrypted (GB), 5 stars: called it the best site for Turnitin scans — authentic and simple.
• Shubham Pachauri (IN), 5 stars: praised the speed and liked Humanize for sounding more natural while keeping the original meaning.
• Taksh Patel (AU), 5 stars: "100% legit and works."

Note the one 4-star review cites a 20-minute wait. That is consistent with the stated turnaround rather than contradicting it.

How the Other Eight Tools Compare

Fair comparison requires acknowledging that these tools are not all trying to do the same job. Here is what each actually offers, based only on the supplied research.

Grammarly

Grammarly positions itself as free AI writing assistance, with AI agents, an AI detector, an AI humanizer, a plagiarism checker, and a paraphrasing tool, plus a citation generator and word counter. It claims 50,000 organisations and 40 million users. Its strengths are breadth and integration across apps, browsers, and mobile. Its limitation for this use case is that it is a writing assistant first — its detector is a feature, not a Turnitin-equivalent report, and it does not produce the instructor-facing AI and similarity PDFs.

Scribbr

Scribbr offers proofreading, plagiarism checking, citation generation, and AI detection, with human proofreading claimed in three hours and AI proofreading in five minutes. Its citation generator covers APA, MLA, Chicago, AMA, IEEE, and ACS, and it maintains a substantial academic writing knowledge base. It is a strong academic support platform. What it does not claim to provide is a report matching what your specific instructor will see.

turnitindetector.ai

This vendor claims a free AI checker with unlimited words, no signup, instant scores, and sentence-level highlights. The stated score bands run from 0–30% ("mostly human-like") to 81–100% ("very strong AI-like signals"). There is no independent user feedback available and no evidence of accuracy or reliability. Treat the bands as heuristic, not diagnostic.

SupWriter

SupWriter is an AI humanizer and detector aimed at students and non-native English writers, available via Apple App Store, Chrome extension, Claude, OpenAI, and MCP, with a stated free tier of 300 words and a built-in detector on paid plans. It claims 9,000+ writers worldwide. Notably, it includes a disclaimer telling users to follow school or employer rules on AI use — a responsible touch. No independent feedback is available.

turnitindetectorai.com

This vendor is unusually candid. It states outright that it does not reproduce Turnitin's proprietary detection model and cannot predict an official Turnitin result. It also notes that Turnitin requires 300+ words of qualifying long-form prose to generate an AI Writing Report, that Turnitin advises its model should not be the sole basis for adverse action, and that false positives occur — especially with ESL students and formal styles. That last claim sits in tension with Turnitin0's ESL study, which found zero false positives across 340 essays. Both can be true: false positives may occur in the wild while remaining absent from a controlled corpus.

Solvely.ai

Solvely is an all-in-one study tool — homework help, notes, flashcards, quizzes, exam prep — claiming 10M+ students, 5,000+ universities, 200M+ problems solved, 97% overall accuracy, and a 4.8 App Store rating. It is not an AI detector or a Turnitin report service. It belongs in a different comparison entirely, and I include it only because students often conflate study tools with integrity tools.

humanizeai.pro

A free online humanizer claiming to bypass all detectors including Turnitin, produce 100% original output, and retain meaning and SEO value. It supports.txt,.docx,.pdf, and.md uploads, offers style and tone settings, and requires a Captcha. Its own documentation warns that processing may take time and may require multiple iterations. No independent reviews are provided.

aiturnitinchecker.com (SubmitSense)

SubmitSense claims official Turnitin AI and similarity reports in 5–15 minutes, 99% within 30 minutes, files deleted within 24 hours, metadata stripped, one-time payment, 30-day report access, and a money-back guarantee. It accepts DOCX, PDF, TXT, or RTF at 400–30,000 words and up to 40 MB. It also states it is not affiliated with Turnitin LLC — which makes the word "official" worth scrutinising. To its credit, it explicitly warns that AI detectors produce false positives and false negatives and should not be the sole basis for decisions.

Comparison Summary

What Turnitin Itself Says

Turnitin's own documentation is clear that the similarity score is not a plagiarism verdict — it is the percentage of matching text, and quoted material with citations still registers as a match [1]. Turnitin's platform materials frame its academic integrity products around upholding standards and guiding responsible AI use rather than punishing students [3]. Institutional guidance echoes this: the report is a review tool, and matched text must be interpreted in context [2][4].

That framing is important. A Turnitin check service is not a way to "beat" a system; it is a way to see your own work the way it will be seen, while you still have time to fix genuine problems — an uncited paraphrase, a missing quotation mark, or a passage that drifted too far toward machine phrasing.

Recommendations

If you have written your essay yourself: you are statistically safe. Both human-corpus studies returned 100% word-level accuracy with zero false positives. A pre-submission check is still worth it for the similarity report alone, because references and titles always match.

If you used AI to polish your own draft: this is the highest-risk category. TT0-2026-0006's 47.54% accuracy means Turnitin's verdict here is genuinely unpredictable. Check before you submit.

If you generated text with ChatGPT, Claude, or Gemini: assume it will be caught. Raw generation was detected at 97.88%–99.01% word-level accuracy. Either rewrite substantially in your own voice or use the humanizer and re-check.

If English is your second language: the ESL data is on your side — 340 essays, 263,329 words, zero false positives. Do not let internet folklore convince you that Turnitin is targeting you.

Conclusion

Across 2,048 essays and more than 1.17 million words, the pattern is unambiguous: Turnitin is highly accurate on raw AI generation and on genuine human writing, and unreliable on AI-polished human writing. That single gap is why pre-submission verification is worth doing, and it is why Turnitin0 is my recommendation. It returns the same AI and similarity reports your instructor opens, it is non-repository so your file never enters Turnitin's student paper database, it delivers in under 15 minutes in 98% of cases, and it publishes its own accuracy research — including the results that do not flatter it. Its AI humanizer carries a specific score promise with a full refund if it is not met, and 98.2% of humanizer orders are re-checked with Turnitin.

The honest caveats stand: no free humanizer trial, a small Trustpilot sample, English-only documents between 300 and 30,000 words, and the *% threshold display that Turnitin itself imposes. None of that changes the core conclusion. If you want to know what your professor will see before your professor sees it, use a Turnitin check service that shows you the real report — and on the evidence assembled here, that service is Turnitin0.