AI Detector Database

Built by Edward Tian as a Princeton senior over winter break 2022-2023. Became the first AI detector with public traction. Most-used detector in academic settings as of 2026. Originally web-only; now integrates into Canvas, Schoology, and many LMS platforms via API.

Turnitin rolled AI detection into its existing plagiarism workflow rather than launching a standalone product. Vanderbilt University famously turned it off in August 2023, citing reliability concerns. The University of Texas at Austin, Northwestern's writing program, and several Cal State campuses followed. The MLA's 2024 guidance explicitly advised against standalone use.

Launched specifically targeting publishers, content agencies, and freelance content workflows. Strictest of the consumer detectors and the hardest to bypass cleanly. Jon Gillham has been a vocal industry voice on detector accuracy, including publishing periodic accuracy benchmarks against his own product.

Originally a plagiarism-detection company. Added AI detection in 2023 as enterprise demand grew. Strong Microsoft Word and LMS integration. The plagiarism layer remains the better-known product; AI detection is a side feature.

Free, fast, no-signup access made it the detector students run on themselves before submission. Particularly sensitive to formulaic openings ('In conclusion,' 'In today's society') and AI-typical transitions. Lowest accuracy of the major tools but useful as a quick gut-check.

Originally a writing assistant similar to Grammarly. Added AI detection as a side feature. Per-sentence detection is useful for finding inserted AI paragraphs in mostly-human documents but produces high false positive rates on dry, technical writing.

Marketed heavily to publishers, educators, and copyright lawyers. Includes OCR for scanned documents, useful for academic and legal contexts. The plagiarism layer is the stronger selling point; AI detection is solid but not class-leading.

Combines plagiarism with AI detection and adds a cross-language layer that flags AI-translation patterns. Strong in non-English and translated academic contexts. Less commonly used in US classrooms; more visible in European universities.

OpenAI built and shipped its own AI text detector on January 31, 2023, billed as a tool to help users identify AI-generated text. It was discontinued just six months later, on July 20, 2023, with OpenAI explicitly citing 'low rate of accuracy.' The shutdown was a major signal about the difficulty of the problem: the company that built the AI could not reliably detect its own output.

The classic open-source AI detector. Frequently cited in academic papers and used as a baseline in detector research. HumanGPT incorporates this detector as one signal in its internal pessimistic consensus, alongside perplexity statistics and LLM-based judgment.

An academic research tool rather than a commercial detector. Highlights each word in a passage by how predictable it was to GPT-2. Foundational for the field; many later detector papers built on its insights about token-level predictability.

A 2023 Stanford research paper that proposed measuring how much a model's likelihood of a passage changes under small perturbations. Influential in academic circles, less deployed in production. Open implementations exist on GitHub.

Notable research detector that uses two reference LLMs. The intuition: AI text has very similar perplexity across different models, while human text varies more. Cited in 2024 academic literature; reference implementation available.

Marketed primarily to SEO content teams alongside the company's AI writing platform. Useful as a built-in safety check in content workflows; not class-leading on accuracy.

Writer.com's free detector is a marketing surface for their enterprise AI writing platform. Not the company's main product. Useful as one signal among several but not as a primary tool.

Part of Smodin's broader writing-tools suite that includes a paraphraser and citation generator. The detector is a side feature; the rewriter is more frequently used. Common in student workflows.

Hive originally built moderation infrastructure for adult and harmful content classification. AI text detection was added as enterprise demand grew. Used by major social platforms and content marketplaces rather than by individual users.

Scribbr is an academic editing service acquired by QuillBot in 2023. Its public AI detector is essentially a free front-end for Turnitin's detection technology. Useful for students who want a Turnitin pre-check without an institutional subscription.

QuillBot is best known as a paraphraser. The AI detector is a side feature that often disagrees with QuillBot's own paraphrased output, an irony noted by users. Acquired by Course Hero in 2021; QuillBot acquired Scribbr in 2023.

Markets itself on the consensus angle: combining six different detection methods into one verdict. The approach is sound in principle but the underlying detectors share many of the same weaknesses, so consensus does not fully resolve the false positive problem.

Smaller commercial detector with branding focused on professional content workflows. Less academic visibility than the major players. Periodic updates but no published methodology.

Major French academic plagiarism vendor that added AI detection. Widely used in French universities. Less common in English-language workflows. Strong in French academic context.

Phrasly is primarily a humanizer; the detector is a sanity-check feature inside the humanizer's UI. Useful for users who want a built-in score before output but not a primary detection tool.

Unique in being a humanizer company that ships its own meta-detector. The meta approach (aggregating GPTZero, Originality, Copyleaks, etc.) provides a useful cross-check for users but does not eliminate the underlying detector weaknesses.

Smaller commercial detector with limited public benchmarking. Bundled with JustDone's broader writing-tool suite. Useful as a quick gut-check; not a primary detection tool.

StealthGPT is primarily a humanizer; the detector is a complementary feature inside the product. Used internally to validate the humanizer's output before showing it to the user.