AI Detector Database.

OrganizationGPTZero, Inc. (A Princeton University spinoff company)

BaseIt uses a combination of perplexity and burstiness statistics, alongside a fine-tuned classifier, to identify AI-generated text.

Vendor claimThey claim impressive accuracy, stating up to 99% on longer pieces of content.

Real accuracyOur 500-sample test showed a 69% overall accuracy rate. It's decent, but not perfect.

False positive rateYou'll see a 12% false positive rate for native English texts, which jumps to 38% for non-native English writing. That's a significant difference.

PricingThere's a free tier available for basic use, with paid plans starting at $10 per month.

OrganizationTurnitin, LLC

BaseThis detector employs a proprietary classifier, built on top of Turnitin's established plagiarism indexing system.

Vendor claimTurnitin asserts a 98% accuracy rate, with a false positive rate reportedly under 4%. These numbers are quite high.

Real accuracyIn our tests, it achieved 68% overall accuracy. However, this figure is often disputed by independent academic research.

False positive rateIt flagged 6% of native English texts incorrectly, and a higher 24% for non-native English.

PricingYou can't buy this one standalone; it's always part of Turnitin's institutional licenses for educational organizations.

OrganizationOriginality.AI Inc.

BaseOriginality.ai uses a custom transformer model, which is frequently retrained to keep up with new AI developments.

Vendor claimThey claim over 98% accuracy. That's a bold statement.

Real accuracyAmong the major tools we tested, it scored the highest at 75% overall accuracy. Pretty good!

False positive rateIt showed a 9% false positive rate for native English and 31% for non-native English content.

PricingMonthly plans start at $14.95, with more comprehensive agency tiers available, priced up to $209 per month.

OrganizationCopyleaks Inc.

BaseCopyleaks integrates a whole-document classifier directly into its existing plagiarism detection engine.

Vendor claimThey advertise an impressive 99.1% accuracy.

Real accuracyOur testing found an overall accuracy of 68%.

False positive rateThis tool had a 14% false positive rate for native English and a higher 42% for non-native English.

PricingIndividual users can get a plan starting at $8.33 per month.

OrganizationZeroGPT

BaseIt uses a statistical detector combined with a lightweight classifier.

Vendor claimThe company claims over 98% accuracy.

Real accuracyThis was the lowest performer among the seven major tools we tested, achieving only 55% overall accuracy.

False positive rateNative English texts had an 18% false positive rate, while non-native ones hit 47%. That's pretty high.

PricingZeroGPT offers a free tier, and Pro plans begin at $7.99 monthly.

OrganizationSapling Intelligence Inc.

BaseSapling employs a per-sentence classifier, which then aggregates these scores to provide a document-level assessment.

Vendor claimThey claim 97% accuracy.

Real accuracyIt scored 62% overall in our tests.

False positive rateNative English texts saw a 16% false positive rate, and non-native texts jumped to 44%.

PricingA free tier is available, with Pro plans starting at $25 per month. These plans often bundle other Sapling writing assistant features.

OrganizationWinston AI

BaseWinston AI combines a plagiarism and AI classifier, notably including Optical Character Recognition (OCR) capabilities. This means it can even scan text from images.

Vendor claimThey state a remarkable 99.98% accuracy on documents up to 80,000 characters.

Real accuracyOur tests showed an overall accuracy of 67%.

False positive rateIt produced an 11% false positive rate for native English, and 33% for non-native English.

PricingYou can try it out with a free tier, limited to 2,000 words. Paid plans then start from $12 per month.

OrganizationCrossplag Inc.

BaseThis tool uses a classifier that features cross-language semantic analysis, designed to work across various languages.

Vendor claimThey claim 96% accuracy across multiple languages.

Real accuracyIndependent tests place its accuracy in the mid-tier, around 65%.

False positive rateIt tends to be aggressive, often over-flagging academic and formal texts as AI-generated. This can be problematic for students.

PricingA free tier is available, and institutional pricing can be requested directly from them.

OrganizationOpenAI

BaseThis classifier was built upon a fine-tuned GPT model, as one might expect from OpenAI.

Vendor claimAt its launch, OpenAI itself stated it had only a 26% true positive rate.

Real accuracyOpenAI publicly admitted its accuracy was low right from the start. It never really got competitive.

False positive rateIt registered a 9% false positive rate on the original test set used by OpenAI.

PricingIt was free to use for its entire, albeit brief, lifespan.

OrganizationHugging Face

BaseIt's a RoBERTa large model, specifically fine-tuned on outputs from the older GPT-2.

Vendor claimInitially, it showed approximately 95% accuracy on the original GPT-2 test set.

Real accuracyIts effectiveness drops sharply when applied to text generated by more modern Large Language Models like GPT-4o, Claude 3.5, or Gemini 2.5. It's just too old for today's AI.

False positive rateThis one is quite variable; it frequently over-flags formal academic writing as AI.

PricingIt's freely available through the Hugging Face inference API.

OrganizationMIT-IBM Watson AI Lab

BaseGLTR is a visual perplexity inspector. It's built upon GPT-2 to highlight predictable word choices.

Vendor claimIt's designed as a visualization tool to assist humans, not as a standalone classifier to give a definitive "AI or not" answer.

Real accuracyIts core purpose is to support human judgment by showing patterns, not to replace it with an automated score.

False positive rateThis metric isn't applicable here, as it doesn't give a pass/fail judgment.

PricingIt's free to use, hosted by Harvard NLP.

OrganizationStanford University

BaseThis method analyzes probability curvature using local perturbations within the text.

Vendor claimThe researchers claimed it outperformed previous detectors in their published paper.

Real accuracyWhile it performed strongly in laboratory settings, its efficacy hasn't been widely validated in real-world production environments.

False positive rateIt showed lower false positive rates compared to simpler statistical baseline methods.

PricingIt's available as open-source code.

OrganizationUniversity of Maryland

BaseBinoculars operates by comparing the perplexity scores generated by two different large language models.

Vendor claimAt the time of its publication, the researchers reported that it surpassed the performance of existing detectors.

Real accuracyIt demonstrated strong performance on the specific test corpora used in their study.

False positive rateIt showed lower false positive rates when compared against detectors that rely on a single model.

PricingThe project is open source and available on GitHub.

OrganizationContent at Scale, Inc.

BaseThis detector uses a proprietary classifier trained specifically on content generation patterns.

Vendor claimIt claims high accuracy for detecting AI-generated content produced for web articles and blogs.

Real accuracyIndependent evaluations typically place this detector in the mid-tier for accuracy. It's not the best, but not the worst.

False positive rateIt can sometimes over-flag highly optimized SEO content written by humans.

PricingThere's a free tier to get started, but the paid detector is usually bundled with the broader Content at Scale AI writing platform.

OrganizationWriter.com Inc.

BaseIt uses a custom-trained classifier, designed to integrate with their enterprise writing platform.

Vendor claimThey market it as highly accurate for brand-specific content.

Real accuracyThis tool generally performs in the mid-tier. However, it's not frequently benchmarked publicly, making definitive comparisons difficult.

False positive rateCan be inconsistent, especially with highly polished, repetitive content.

PricingA free public detector is available, but the enterprise platform, which includes this detector, is priced separately and tailored to specific business needs.

OrganizationSmodin.io

BaseThis detector uses a statistical analysis model combined with linguistic pattern recognition.

Vendor claimThey claim high accuracy for academic and professional writing.

Real accuracyIndependent tests have shown it to be on the lower end of the accuracy spectrum.

False positive rateIt often over-flags formal academic text as AI-generated, which can be frustrating for students and educators.

PricingThere's a free tier for basic checks, with Pro plans starting from $10 per month.

OrganizationHive Moderation

BaseHive uses a proprietary deep learning model, focused on identifying patterns in raw AI outputs.

Vendor claimThey claim high accuracy for detecting AI across various content types, including text, image, and video.

Real accuracyIt performs strongly when detecting raw, unedited AI text. However, its accuracy shifts closer to the mid-tier once text has been humanized or edited.

False positive rateGenerally low on raw AI, but can increase with heavily edited human text that still retains some AI characteristics.

PricingThis is typically an enterprise API solution, with custom pricing tailored to client needs.

OrganizationScribbr B.V.

BaseScribbr's detector is known to utilize similar underlying technology as Turnitin, focusing on academic integrity.

Vendor claimThey often align their accuracy claims with what Turnitin presents.

Real accuracyIts accuracy profile closely mirrors that of Turnitin, reflecting similar strengths and weaknesses.

False positive rateSimilar to Turnitin, it may show higher false positives for non-native English or complex academic structures.

PricingIt's available as a free tool.

OrganizationQuillBot (a Course Hero Inc. company)

BaseIt uses a classifier integrated with QuillBot's extensive linguistic processing engine, designed to assess text originality.

Vendor claimThey claim high accuracy for identifying AI-generated content, especially when used in conjunction with their paraphrasing tools.

Real accuracyIndependent tests generally place it in the mid-tier for overall accuracy.

False positive rateCan sometimes flag highly structured or repetitive human-written text.

PricingThere's a free tier available, and the Premium plan, which includes the AI detector, starts from $9.95 per month and bundles it with their popular paraphraser.

OrganizationGPTKit

BaseGPTKit uses a combination of deep learning models and statistical analysis to identify AI patterns.

Vendor claimThey claim robust detection capabilities for various LLMs.

Real accuracyIt generally performs in the mid-tier. However, there's a broader consensus that these detectors haven't fully solved the underlying technical challenge of AI detection.

False positive rateCan vary, with some reports of over-flagging human text that mimics AI-like patterns.

PricingA free tier with daily caps is available, and paid plans offer more extensive usage.

OrganizationAI Detector Pro

BaseThis tool relies on a proprietary AI model designed to analyze stylistic and structural characteristics.

Vendor claimThey claim high accuracy rates for various AI-generated content.

Real accuracyThere has been limited independent testing available for this detector, making it harder to provide a definitive accuracy assessment.

False positive rateInsufficient public data to provide reliable false positive rates.

PricingIt's a subscription-based service, with plans starting from $14 per month.

OrganizationCompilatio SAS

BaseCompilatio integrates an AI detection module into its established plagiarism detection platform, often used in European academic institutions.

Vendor claimThey emphasize robust detection for academic integrity.

Real accuracyIt generally performs in the mid-tier for AI detection.

False positive rateAnecdotal evidence suggests it can be strict on formal, structured writing typical in academia.

PricingThis product is exclusively available through institutional pricing, targeting schools and universities.

OrganizationPhrasly.ai

BasePhrasly uses a proprietary AI model designed to evaluate text against common AI generation patterns.

Vendor claimThey claim high accuracy, particularly when paired with their humanization features.

Real accuracyLimited public benchmarking makes a precise assessment challenging.

False positive ratePublic data is scarce regarding its false positive rates.

PricingIt's typically bundled with Phrasly's humanizer plan, with subscriptions starting from $9.99 per month.

OrganizationUndetectable.ai

BaseThis detector uses a proprietary model focused on identifying characteristics of raw AI output.

Vendor claimThey claim high accuracy on raw AI text.

Real accuracyIt performs strongly on raw, unedited AI content. However, by design, it becomes less effective on text that has been "humanized" by their own platform. This is intentional, as their humanizer aims to bypass detectors.

False positive rateGenerally low on raw AI, but its purpose is to be bypassed by humanized text.

PricingA free tier is available, and the paid detector is bundled with the Undetectable.ai humanizer.

OrganizationJustDone

BaseIt uses a proprietary machine learning model to analyze text for AI indicators.

Vendor claimThey claim high accuracy for various content types.

Real accuracyIndependent tests have placed its performance on the lower end compared to more established detectors.

False positive rateLimited public data means specific false positive rates are not widely available.

PricingA free option is available, with paid plans offering additional features.

OrganizationStealthGPT

BaseThis detector utilizes a proprietary algorithm specifically designed to identify AI text that attempts to evade detection.

Vendor claimThey claim to detect even "undetectable" AI content.

Real accuracyThere's limited public testing or independent verification for its accuracy claims.

False positive rateInsufficient public data to provide reliable false positive rates.

PricingThe detector is bundled with StealthGPT plans, which start from $14.99 per month.