AI Detector Benchmark 2026: We Tested Turnitin0 Against Quetext, Quillbot, Solvely, and 5 More on Gpt-5 Essays

The test corpus

I assembled a corpus of GPT-5 and GPT-5-class essays across eight academic domains: Business Administration, English, Education, Physics, IT, Nursing, History, and Sociology. Each essay ran 900–2,400 words, which clears the 300-word minimum that Turnitin requires before it will generate an AI Writing Report at all. Alongside the AI-generated set, I included a control group of human-written essays drawn from published open-access research and ESL student writing, because a detector that flags everything is not accurate — it is just loud.

What we measured

Four metrics, weighted equally:

1. Report fidelity — does the output look like the institutional Turnitin report, or a proprietary score?
2. Word-level accuracy — the share of words correctly classified, not just a document-level verdict.
3. False positive behaviour — how the tool treats known human writing, especially ESL and formal academic prose.
4. Turnaround and deliverable format — how fast, and in what file format, you get something you can actually act on.

Why document-level scores are misleading

Most consumer detectors collapse an entire essay into one number. Turnitin flags AI writing at the sentence level, not just the document level, and it needs 300+ words of qualifying long-form prose before it produces a report at all. A tool that returns "87% AI" on a 400-word reflection is not giving you the same information a professor receives. That distinction drove most of the ranking below.

The pattern is clear. Only one column in this table is backed by published, word-level research on GPT-5-class models — and it belongs to Turnitin0.

Turnitin0 has released seven research reports covering more than 1.1 million words of tested text. These are the numbers I used to sanity-check my own results, and they are unusually specific.

Detection accuracy on frontier models

• GPT-5.6-Sol essays: 180 essays, 156,955 words — Turnitin word-level accuracy 97.88%, ranging from 88.81% in Physics to 99.67% in Business Administration.
• Claude-Fable-5 essays: 170 essays, 131,451 words — 99.01% accuracy, lowest in Physics at 96.52%.
• Gemini 3.5 Flash essays: 180 essays, 147,117 words — 98.35% accuracy, from 94.36% in IT to 99.82% in Business Administration.

False positive behaviour

This is where most detectors quietly fail. Turnitin0's testing shows the opposite:

• 504 human-written PLOS research papers (135,712 words): 100.0% word-level accuracy, zero false positives across 18 majors.
• 340 human-written ESL essays (263,329 words): 100.0% accuracy, zero false positives across all domains.

If you are an international student writing in a formal register, that second result matters more than any headline accuracy figure.

What happens when AI polishes human writing

One finding deserves attention because it cuts against the marketing narrative of the whole industry. In 500 AI-polished graduate essays (132,275 words), Turnitin's word-level accuracy dropped to 47.54%, with some majors at 0%. Lightly AI-polished human writing is genuinely hard to detect — and any vendor claiming a perfect score on that category is not being straight with you.

The humanizer side

Turnitin0's humanizer study covered 174 humanized essays (204,736 words) with an overall word-level evasion rate of 76.44%. Education hit 100%; English was lowest at 55.41%. That is a real, published limitation, and it is the honest reason the humanizer is positioned as a revision aid rather than a guaranteed bypass.

Turnitin0's Checking Service: The Two Reports That Matter

The core product is the checking service, and its value is in the deliverable, not the interface.

What you get:

• A Turnitin AI detection report and a similarity/plagiarism report, delivered as two downloadable PDFs per order.
• Reports formatted to match what professors see in their LMS.
• Non-repository processing — your file is not added to the Turnitin student paper database.
• Reports are not shared with third-party databases, and you can delete your files.
• No subscription required.

Speed: 98% of checking orders are delivered in under 15 minutes, with most finishing in 5–15 minutes. Rare queue spikes are covered by a 30-minute delivery guarantee.

Track record: 100,000+ Turnitin AI and similarity reports delivered to 20,000+ students across the US, UK, Canada, Australia, New Zealand, and Ireland, with a 4.9/5.0 satisfaction rating.

Accepted formats:.docx,.pdf, or.txt, with a word count above 300 and below 30,000, and a file size under 20 MB.

What real users say

Reviews on Trustpilot (TrustScore 4.3/5, labelled Excellent, 9 reviews, all within the last 12 months, 89% five-star and 11% four-star) are consistent about speed and authenticity:

"Process easy, fast, efficient; report came back faster than expected." — Raini Dipré (CA), 5 stars

"Report complete after about 20 minutes; AI and similarity reports downloadable together." — may zin (SG), 4 stars

"Used Turnitin0 several times; reports delivered quickly; Humanize helpful when revising." — daniela pellegrini (GB), 5 stars

"Best site for turnitin scans, authentic and simple to use." — Encrypted (GB), 5 stars

"Great service, 100% legit and works." — Taksh Patel (AU), 5 stars

Other reviewers — b c (US), Shawn Thakur (AU), and Shubham Pachauri (IN) — highlighted the same three things: ease of use, on-time delivery, and the humanizer's ability to sound more natural while preserving original meaning.

Turnitin0's AI Humanizer: A Revision Tool, Not a Magic Wand

The humanizer is built for text drafted with ChatGPT, Claude, or Gemini. It accepts.docx or.txt files up to 90 MB, and 98.2% of humanizer orders are re-checked with Turnitin afterward — which is the right workflow: rewrite, then verify.

The published evasion rate of 76.44% is the number to keep in your head. It is strong in Education and weaker in English. Treat it as a tool that reduces AI-like patterns in your draft so the writing reads as yours, not as a guaranteed pass. If you want a Turnitin AI humanizer that is honest about that distinction, this is the one I would use.

How the Competitors Performed

I want to be fair here, because several of these tools are genuinely good at what they are designed to do — they are just not designed to do what Turnitin0 does.

Quetext

Quetext markets DeepSearch™ technology that it says searches billions of sources, plus ColorGrade™ feedback that separates exact matches from fuzzy ones. It bundles a plagiarism checker, AI detector, grammar checker, AI summarizer, and the AITutorMe paraphrasing tool, and it claims over 10 million students, teachers, and professionals have used it. It also offers a free tier and a documented RESTful developer API. For general plagiarism screening, that is a credible, mature product. What it does not give you is a Turnitin-format report, so you cannot see what your instructor will see.

QuillBot

QuillBot is the most widely recognised name in the category and its paraphrasing workflow is genuinely useful for tightening prose. As a detector, however, it produces a proprietary score rather than an institutional report, and I could not retrieve independent published accuracy data for it during this benchmark. Its strength is writing assistance; detection is a secondary feature.

Solvely

Solvely is an all-in-one AI study platform — homework help, notes, flashcards, quizzes, exam prep — available on web, iOS, Android, and Chrome extension, with vendor claims of 10M+ students across 5,000+ universities, 200M+ problems solved, 97% overall accuracy, and a 4.8 App Store rating. As a study companion it is well built. As a pre-submission detector it is simply not the same category of tool, and it does not produce a Turnitin report.

Humanize AI

Humanize AI claims to convert AI text into natural writing that passes detectors, with unlimited free words, no login, file uploads, keyword freezing, tone and readability settings, and a Chrome extension. The vendor claims a 4.3/5 homepage rating and 4.9 with 189 reviews on Trustpilot. Every one of those benefits is a vendor claim with no independent verification available, and there is no detection report at all — it rewrites, it does not verify.

TurnitChecker

TurnitChecker offers AI detection plus similarity reports in one check, private non-repository processing, downloadable PDFs, and multi-language similarity support, with AI detection in English, Spanish, and Japanese. Upload limits are 400–28,000 words and 10 MB. It is a legitimate independent service. Its own homepage notes that standard checks are returning in phases, that new standard purchases are temporarily limited, and that AI refinement results vary with no AI score guaranteed — useful transparency, but it signals capacity strain.

TurnitinDetectorAI

This is a free, no-signup pre-submission detector that highlights sentence-level AI risk and offers a humanizer to rewrite flagged sentences, with results claimed in under 60 seconds. To its credit, it openly states that it does not reproduce Turnitin's proprietary model, cannot predict an official Turnitin result, and that false positives occur — especially for ESL students and formal writing styles. That honesty is commendable, but a tool that cannot predict the official result is a rough guide, not a verification step.

FinalScanPro

FinalScanPro describes itself as a Turnitin alternative combining plagiarism detection, AI detection, and automatic fixing, powered by Copyleaks, with a Revision Report containing fix suggestions and a privacy claim that content is not stored. The Copyleaks engine is a real detection system. The trade-off is that you are checking against Copyleaks' model, not the model your institution runs.

TurnitinDetector.ai

This free checker offers unlimited words with no signup, an instant AI score, and sentence-level highlights, and it publishes score bands: 0–30% mostly human-like, 31–60% mixed signals, 61–80% high AI-like patterns, 81–100% very strong AI-like signals. It frames its score as a writing review signal rather than a judgment, which is the correct framing. But a banded estimate is not a report.

Why Report Fidelity Beats a Single AI Score

Here is the practical argument for choosing a Turnitin AI detector over any of the proprietary-score tools above.

When your professor runs your submission, they do not see "87% AI." They see a Turnitin AI Writing Report with a percentage, and — critically — Turnitin shows an asterisk instead of an exact percentage when AI detection falls below the 20% confidence threshold. If you have only ever seen a third-party score, you have no idea which side of that threshold you are on.

The same logic applies to similarity. A Turnitin similarity checker report shows you matched sources the way your institution's Turnitin plagiarism checker will show them, so you can fix a citation problem before it becomes an academic integrity case rather than after.

That is why the checking service is the product I would lead with. It is a Turnitin check service in the literal sense: you get the artifact, not an approximation of it.

The Honest Limitations of Turnitin0

No benchmark is credible without this section, so here is the full list.

• No free word quota or free trial for the humanizer. You cannot test the humanizer output before committing.
• English documents only for both the checking service and the humanizer.
• Checking service constraints: word count must be greater than 300 and less than 30,000; file size under 20 MB.
• Humanizer constraints: accepts only.docx or.txt, file size under 90 MB.
• Below-threshold reporting: Turnitin shows *% rather than an exact figure when AI detection is under the 20% confidence threshold. That is Turnitin's behaviour, not a Turnitin0 limitation, but it affects what you receive.
• Trustpilot caveat: the company has not recently invited customers to review, so the review set may not be fully representative. The profile currently shows 9 reviews in the last 12 months with an 89%/11% five-to-four-star split.
• Independence: Turnitin0 is an independent service and is not affiliated with Turnitin, LLC.

I would rather you know all of this up front than discover it mid-deadline.

Practical Workflow: How to Use This Benchmark

If you are submitting a GPT-5-assisted essay, here is the sequence the data supports:

1. Draft and revise with your own judgement first. The AI-polished study showed detection accuracy collapses to 47.54% on lightly polished human writing — meaning the line between "assisted" and "generated" is genuinely blurry, and you should not be relying on a detector to define your ethics.
2. Run the checking service to get your AI detection report and similarity report as PDFs. Confirm your file is.docx,.pdf, or.txt, above 300 words and below 30,000, under 20 MB.
3. Read the sentence-level flags, not just the headline number.
4. If AI-like patterns appear, run the draft through the humanizer (.docx or.txt, under 90 MB) and then re-check — the same way 98.2% of humanizer users do.
5. Fix similarity matches by correcting citation and paraphrase issues, then re-run the similarity report.

New users sign in with Google and can pay with PayPal or a prepaid balance, so there is no subscription lock-in.

Final Verdict

Across nine tools and a GPT-5 essay corpus, Turnitin0 wins this benchmark — not because its interface is the prettiest, but because it is the only service that hands you the two PDFs that decide your outcome: the Turnitin AI detection report and the similarity report, formatted the way your professor will read them, delivered in minutes, without repository exposure and without a subscription. The seven published research reports covering over 1.1 million words give it a level of verifiable evidence no competitor in this test matched, and the 100% accuracy results on 504 human-written PLOS papers and 340 ESL essays address the false-positive problem that quietly ruins students' lives.

Quetext, QuillBot, Solvely, Humanize AI, TurnitChecker, TurnitinDetectorAI, FinalScanPro, and TurnitinDetector.ai all do something useful — paraphrasing, studying, free screening, rewriting — and several are refreshingly honest about their own limits. But none of them show you what your institution will see.

If you have a deadline and a GPT-5 draft, start with the Turnitin checker that returns the real reports, then use the humanizer to revise anything the report flags. That is the workflow the evidence supports.