The short answer
Partly. In Scribbr's own 30-text test, its premium AI detector was right 84% of the time and the free version 78%, with no human text flagged. A small independent test found the free version caught 4 of 9 AI samples. Plagino has not benchmarked Scribbr, so treat its score as a signal, not proof.
How accurate is Scribbr's AI detector?
Scribbr is unusually open about the limits of its detector, and the evidence it publishes is modest. The best figure available is Scribbr's own: 84% overall accuracy for the premium detector and 78% for the free one, on a small test set Scribbr built itself.
Plagino has not benchmarked Scribbr. Our March 2026 benchmark covered seven detectors and Scribbr was not one of them. The figures below are Scribbr's own claims or third-party tests, and we label each one.
What Scribbr's own test found
Scribbr's best AI detector research (its citation line is dated 21 July 2026) ran 12 detectors on 30 texts, each 1,000 to 1,500 characters long. There were five texts in each of six groups: human-written, GPT-3.5, GPT-4, human text mixed with GPT-3.5, GPT-3.5 text paraphrased with QuillBot, and human text paraphrased with QuillBot.
Scribbr's premium detector scored 84% and its free detector 78%, the highest premium and joint-highest free scores in that test. Scribbr gave each scan 1, 0.5 or 0 points depending on how close the percentage was to the right answer, and left the paraphrased human texts out of the score. Scribbr says neither tool labelled any of its human texts as AI, and both detected every GPT-3.5 and GPT-4 text. Mixed and paraphrased AI text was much harder: the premium tool caught 60% of those, the free tool 50%.
Keep the scale in mind. Five fully human texts cannot establish a false positive rate, and GPT-3.5 and GPT-4 are older than the models most students use in 2026. Scribbr also wrote and scored the test of its own product. Its detector page quotes a different free-tool figure ("68% in the best free tool") from the 78% in its research table, so the numbers have shifted between updates.
What independent testing found
In a multi-detector comparison, Pangram Labs ran Scribbr's free detector on nine roughly 1,000-word AI samples (three each from GPT-4o, Gemini 2.0 and Claude 3.7 Sonnet) and three human texts. Scribbr passed all three human texts and flagged 4 of the 9 AI samples at Pangram's 75% threshold. The free version caught all three Gemini samples, one of three ChatGPT-4o samples and none of the three Claude samples.
Pangram sells a competing detector and used twelve samples, so this is an indication, not a rate. Two of its three human texts were historical extracts, which is not typical student writing. It is consistent with Scribbr's own warning that the free version detects older models "with average accuracy".
Which Scribbr accuracy figures can you trust?
Several different Scribbr numbers circulate in reviews, and they do not measure the same thing. This table separates the ones we could trace to a source from the ones we could not. It is the quickest way to see why one review says 84% and another says 44%.
| Figure | Who produced it | What it measures | How far to trust it |
|---|---|---|---|
| 84% (premium), 78% (free) | Scribbr, in its own 30-text research | Closeness-scored accuracy on short GPT-3.5, GPT-4, mixed and paraphrased texts | Traceable, but vendor-run, tiny, and built on older models |
| 68% for the best free tool | Scribbr's detector page and research article FAQ | A "best free tool" summary figure | Conflicts with the 78% in the same article's table, and Scribbr has not explained the gap |
| 4 of 9 AI samples caught (44%) | Pangram Labs, a rival detector vendor | Free version against GPT-4o, Gemini 2.0 and Claude 3.7 text, 75% pass mark | Traceable, but twelve samples and a competitor's test |
| 84% "in ZDNet's independent test" | Some third-party reviews | Said to be an independent result | Not supported: the ZDNet article we read tested QuillBot, not Scribbr, and 84% is Scribbr's own figure |
| 72.8% detection, 9.2% false positives | A review citing Axis Intelligence | Said to come from 600 samples | Unverified: we could not find Scribbr figures on the Axis page that review cites |
| About 20% of human essays flagged | A review from a rival detector vendor | 4 of 20 human essays | Small, vendor-run, and the sample texts are not listed in full |
| 97.4% AI caught, 0.8% human flagged | Plagino's own March 2026 benchmark | Seven other tools, 500 AI and 500 human documents | Plagino's figures only; Scribbr was not tested |
The pattern is worth noticing. The only figure that holds up under a source check is Scribbr's own, and it is the one most flattering to Scribbr. Read it next to the independent comparison, then decide how much weight a single score deserves. Scribbr's same test also scored ZeroGPT, and what independent studies say about ZeroGPT is covered on its own page.
Which AI models can the Scribbr AI checker detect?
Scribbr documents accuracy for older models only. Its detector page says the tool finds text from ChatGPT, Gemini and Copilot, but the only model-level claims are GPT-2, GPT-3 and GPT-3.5 for the free version and GPT-4 for premium. Scribbr's own test used GPT-3.5 and GPT-4 texts and nothing newer.
The independent result points the same way. In Pangram's small test the free version caught every Gemini 2.0 sample but only one of three ChatGPT-4o samples and none of the three Claude 3.7 Sonnet samples. Nine samples cannot give a rate. They do show that a clean Scribbr result depends on which model wrote the text, and that matters for any work written with newer tools.
Plagino's benchmark mixed GPT-4o, Claude and Gemini documents into one set and reports a single combined detection rate. It does not break results out by model, so it cannot settle the Claude question either. A detector that has not published results for newer models should not be used as the only check.
Is Scribbr's AI detector powered by Turnitin or QuillBot?
Not by Turnitin. This is a common mix-up. Scribbr's plagiarism checker uses Turnitin technology: Turnitin's own partner story quotes Scribbr saying it included "Turnitin technology in the Scribbr platform". Scribbr's AI detector is a separate product. Scribbr's plagiarism page says the add-on AI detector "is powered by Scribbr's proprietary software".
The closer link is QuillBot. Scribbr's FAQ calls QuillBot and Scribbr sister companies within Learneo, and Scribbr's Spanish-language FAQ says its AI detector works with the QuillBot service. Scribbr's English detector page also points readers who want a more advanced and reliable check to QuillBot's own detector. In Scribbr's research the free QuillBot and free Scribbr detectors both scored 78%, and in Pangram's test both caught 4 of 9. We found no page that explains exactly how the free tool, the premium add-on and QuillBot's detector relate, so treat them as connected but not proven identical.
Some reviews say Turnitin shares a parent company with Scribbr. Learneo's own site lists Scribbr and QuillBot among its businesses and does not list Turnitin. So a Scribbr plagiarism report can come close to a Turnitin similarity report, but a Scribbr AI score tells you nothing about what Turnitin's AI detector will say. Even on plagiarism, Scribbr notes that universities may check work against private student-paper databases that Scribbr cannot access. For more on Turnitin's detector, see does Turnitin detect AI, or Plagino as a Turnitin AI checker alternative.
How do you read a Scribbr AI score?
A Scribbr AI score is a percentage from 0% to 100% showing how much of the text is likely to be written or refined using AI tools. Scribbr's FAQ says each section is labelled AI-generated, AI-generated and AI-refined, human-written and AI-refined, or human-written, and the tool gives paragraph-level feedback so you can see which parts were flagged.
"AI-refined" describes text that looks edited by an AI tool, not text written from scratch by one. That distinction matters when you have used a grammar or paraphrasing tool on your own sentences. Scribbr recommends scanning longer pieces rather than single sentences or paragraphs, and says its AI detector gives no improvement suggestions and cannot detect plagiarism.
Why a percentage can look more certain than it is
Scribbr's own research found that detectors often give results close to 100% or close to 0% even when a text is about half AI and half human. A tidy-looking number is not the same as a precise one. Treat the percentage as a ranking of which paragraphs to read again, not a verdict on the author.
How do Plagino and Scribbr compare side by side?
Scribbr charges per document and is strongest on plagiarism. Plagino is a monthly plan that puts AI detection and source matching in one report. Here is how they line up as of October 2026.
| Plagino | Scribbr | |
|---|---|---|
| AI docs caught | 97.4% in Plagino's March 2026 benchmark (n=500) | Not tested by Plagino. Scribbr's own test: 84% overall accuracy (premium), 78% (free), on 30 texts |
| Human docs wrongly flagged | 0.8% in the same benchmark (n=500) | Not tested by Plagino. None of 5 human texts in Scribbr's own test |
| Free tier | 3 scans/month, no card | Unlimited AI checks up to 1,200 words each, with ads |
| Paid price | Pro: $19/month for 20 scans a day | Per document: $19.95 (to 7,499 words), $29.95 (to 49,999), $39.95 (50,000+), premium AI check included |
| Plagiarism check against | CORE open-access research index | Turnitin-powered: 99.3 billion web pages and 8 million publications; no private student-paper databases |
| Input | Upload a .docx file | Paste text (free AI); Word or PDF upload (plagiarism check) |
| AI report detail | Sentence-level highlights in a PDF report | Paragraph-level feedback with AI-generated, AI-refined and human labels |
Detection rate and false positives
In Plagino's benchmark, run in March 2026 on 500 AI-written documents from GPT-4o, Claude and Gemini (some lightly edited by people), Plagino flagged 97.4% (95% interval roughly 95.6% to 98.5%). Turnitin flagged more, at 98.2%. On 500 human-written documents, Plagino wrongly flagged 0.8% (roughly 0.3% to 2.0%), the lowest of the seven tools, against Turnitin's 1.4%.
Scribbr's 84% is an overall accuracy across mixed text types, not a detection rate on unedited AI essays, so the two numbers do not compare directly. Its clean result on human text is encouraging but rests on five documents. Any detector can misread formal or second-language writing: a 2023 Stanford study found several detectors misclassified more than half of the TOEFL essays it tested as AI-written. A 2023 study of 14 tools by Weber-Wulff and colleagues concluded that detection tools are "neither accurate nor reliable" and lean towards calling AI text human. Our guide to AI detector false positives covers what to do if you are flagged.
Privacy and data retention
Scribbr's AI detector page says "we do not store or share your data". For the plagiarism checker, Scribbr says submissions "will never be added to our content database", and its storage FAQ says you can delete your documents from its server once you have your report.
Plagino keeps reports in your own account. Whichever tool you use, a report is a prompt to review the writing, not evidence of misconduct.
Where Scribbr is the better choice
For one high-stakes plagiarism check, such as a thesis or dissertation, Scribbr is the stronger choice. Its checker uses Turnitin technology and searches billions of web pages and millions of publications. Plagino only searches open-access research through CORE, and does not search the open web or paywalled journals. Our guide to the best plagiarism checker for students compares more options.
Scribbr's free AI detector also suits a quick check of a short passage. It needs no account, has no monthly cap, and takes pasted text. Plagino's free plan allows 3 scans a month and needs a .docx file.
If you only check a few long documents a year, paying per document can work out cheaper than any monthly plan. And Scribbr's AI-refined label is useful if you used a grammar or paraphrasing tool on your own writing and want to see how that reads.
Where Plagino is the better choice
Plagino fits regular checking of essays across a term. Pro costs $19 a month for 20 scans a day, and every scan returns AI highlights at sentence level and matched academic sources in one PDF report. Scribbr's free tool gives paragraph-level feedback, and its premium AI check comes only with a paid plagiarism check per document.
Plagino also publishes a document-level false positive rate from 500 human-written documents, with confidence intervals. If being wrongly flagged is your main worry, that is the number to compare. You can try Plagino's AI detector on a draft for free.
How should you use a Scribbr AI result?
Use it as one input in a short checking routine, not as an answer. These steps work for a student checking a draft and for an instructor reading a flagged paper.
- 1
Paste enough text. Check at least several paragraphs, not a single sentence. Scribbr itself recommends scanning longer pieces for better results.
- 2
Read the paragraph labels. Look at which paragraphs are marked AI-generated or AI-refined, not only the overall percentage. A grammar or paraphrasing tool can produce an AI-refined label on your own writing.
- 3
Match the tool to the text. Remember that the free version documents older models only, and that independent tests found it missed Claude and ChatGPT-4o samples. A clean result is not proof of human writing.
- 4
Get a second reading. Run the same draft through a detector that publishes its false positive rate, such as Plagino, and compare which passages both flag.
- 5
Back the score with evidence. Keep drafts, version history, notes and sources. If you are an instructor, use a flag to start a conversation about how the work was written, not to decide guilt.
If you have been flagged and wrote the work yourself, read what to do if you are falsely accused of using AI. Instructors can see how Plagino fits a marking workflow on the page for teachers.
Test conditions and limitations of this comparison
Plagino's figures come from one benchmark run in March 2026: 1,000 documents (500 AI-written, 500 human-written), each tool used through its own web app on default settings, and scored on whether a document was flagged at all. The 95% intervals are Wilson intervals on 500 documents per rate. Your own results will differ, and a different corpus could rank the tools differently. See the full benchmark method.
Scribbr was not in that run. Scribbr's own test used 30 short texts and older GPT models and was run by the vendor; Pangram's used twelve samples and comes from a competitor. None of these figures is directly comparable with another. Prices and features were re-checked on 5 October 2026.
Comparing other tools? See Plagino vs Grammarly, Plagino vs Winston AI and Plagino vs Originality.ai.
Frequently asked questions
Is Scribbr's AI detector accurate?
Moderately. In Scribbr's own test of 30 texts, its premium detector was right 84% of the time and its free detector 78%, with no human texts flagged. A small independent test by Pangram Labs found the free version caught 4 of 9 AI samples. Scribbr says no AI detector can guarantee 100% accuracy.
Is Scribbr's AI detector free?
Yes. The free version allows unlimited checks of up to 1,200 words each without an account. The premium AI detector, which Scribbr says is more accurate and can detect GPT-4, comes free with a premium plagiarism check, priced from $19.95 per document.
Does Scribbr use Turnitin for AI detection?
No. Scribbr's plagiarism checker uses Turnitin technology, but Scribbr says its AI detector runs on its own proprietary software. A Scribbr AI score will not predict a Turnitin AI score.
Is Scribbr's AI detector the same as QuillBot's?
Not confirmed. Scribbr and QuillBot are sister companies under Learneo, and Scribbr's Spanish FAQ says its AI detector works with the QuillBot service. Scribbr's English plagiarism page calls its add-on AI detector proprietary software, and we found no page that explains how the tools relate.
Does Scribbr store the text I check?
Scribbr says its AI detector does not store or share your data. For the plagiarism checker, Scribbr says submissions are never added to its content database and that you can delete your documents after receiving your report.
What languages does Scribbr's AI detector support?
Scribbr's AI detector page lists English, Spanish, German and French, and its plagiarism page lists Dutch for the premium add-on. Check a non-English result with extra care, since Scribbr does not publish accuracy figures by language.
Has Plagino tested Scribbr's AI detector?
Not yet. Scribbr was not part of Plagino's March 2026 benchmark. This page uses Scribbr's own published test and third-party results instead, and labels each one.
What is the word limit on Scribbr's AI detector?
Scribbr's AI detector page says the free version accepts up to 1,200 words per submission, with no limit on the number of checks. Some third-party reviews quote 1,200 characters instead, which Scribbr's page does not say. Scribbr's research article mentions 500 words per check, so confirm the limit in the tool before pasting a long text.
Can Scribbr detect ChatGPT, Claude and Gemini?
Scribbr says its detector finds text from ChatGPT, Gemini and Copilot, but it only documents accuracy for GPT-2 to GPT-4. In Pangram Labs's small test, the free version caught all three Gemini 2.0 samples, one of three ChatGPT-4o samples and none of three Claude 3.7 Sonnet samples. Treat a clean result on newer-model text with caution.
Can Scribbr detect paraphrased or AI-edited text?
Only partly. In Scribbr's own test, the premium detector caught 60% of texts that mixed AI and human writing or were paraphrased, and the free one caught 50%. Scribbr also uses an AI-refined label, which describes AI-assisted editing rather than text written from scratch by an AI.
Can Scribbr's AI detector flag human writing?
Yes, any AI detector can flag human writing. None of the five human texts in Scribbr's own test was flagged, but five texts cannot give a false positive rate. Formal and second-language writing is flagged more often by some detectors, so a flag should start a review, not end one.
Will a clean Scribbr result mean I will pass Turnitin?
No. Scribbr's AI detector is a separate product from Turnitin's, so the two scores are not linked. Scribbr's plagiarism checker uses Turnitin technology, but even that can differ from your university's result because universities may check private student-paper databases.
What do AI-generated and AI-refined mean in Scribbr?
AI-generated means text that looks written by an AI tool, while AI-refined means text that looks improved or rewritten by one. Scribbr labels each section as AI-generated, AI-generated and AI-refined, human-written and AI-refined, or human-written. The overall percentage shows how much of the text was likely written or refined using AI tools.
