Pangram – AI detector for text and images

Try our Gmail integration An AI detector that actually works. Detect AI-generated content with remarkable accuracy. Trusted by universities, schools, and enterprises worldwide. Upload an image to check for AI Paste or drag and drop it here, or upload a file JPG, PNG, WebP • min 512×512px • up to 30MB each Try a sample image Proven the most reliable and accurate AI detector on the market by third party researchers, including the University of Maryland and the University of Chicago. What is an AI detector and should you use one? 01 AI writing contains structural, stylistic, and semantic patterns that can all be detected by a trained eye or sufficiently sophisticated detection software. AI detection is not magic. It can actually work. 02 Pangram's AI detector uses natural language processing and a massive data set of human and AI writing to analyze patterns in AI generated text created by popular models like ChatGPT, Gemini, Grok, Llama and Claude. Pangram's accuracy leads the industry and has been verified by researchers at the University of Chicago and University of Maryland. 03 In an era where AI-generated content is increasingly prevalent in academia, media, and business, the ability to distinguish between human and machine-authored text is critical. In education, it creates trust between students and teachers. It's important in journalism where integrity is essential. AI-generated text isn't inherently bad. However, society can benefit from being able to distinguish between AI and human text. AI writing contains structural, stylistic, and semantic patterns that can all be detected by a trained eye or sufficiently sophisticated detection software. AI detection is not magic. It can actually work. Pangram's AI detector uses natural language processing and a massive data set of human and AI writing to analyze patterns in AI generated text created by popular models like ChatGPT, Gemini, Grok, Llama and Claude. Pangram's accuracy leads the industry and has been verified by researchers at the University of Chicago and University of Maryland. In an era where AI-generated content is increasingly prevalent in academia, media, and business, the ability to distinguish between human and machine-authored text is critical. In education, it creates trust between students and teachers. It's important in journalism where integrity is essential. AI-generated text isn't inherently bad. However, society can benefit from being able to distinguish between AI and human text. Always up to date Does Pangram detect the latest AI models? Yes. Pangram is trained specifically to detect the latest AI models from frontier labs like OpenAI, Anthropic, Google, Meta, DeepSeek, xAI, and more. We benchmark against 26 models (so far) and they are available to see on our model card. We're accurate on 99.7% of these AI-generated samples. We test broadly on every Pangram release and retest on every major model release. Anthropic Claude Opus 5 99.8% accuracy Tested Jul 2026 Claude Opus 5 99.8% accuracy Tested Jul 2026 OpenAI GPT 5.6 99.6% accuracy Tested Jul 2026 GPT 5.6 99.6% accuracy Tested Jul 2026 Google Gemini 3 99.5% accuracy Tested Nov 2025 Gemini 3 99.5% accuracy Tested Nov 2025 xAI Grok 4.3 99.4% accuracy Benchmarked Jul 2026 Grok 4.3 99.4% accuracy Benchmarked Jul 2026 Also benchmarked Claude Sonnet 599.9% Claude Fable 599.7% GPT-5.499.7% GPT-OSS 120B99.8% DeepSeek V499.6% Llama 3.399.5% Mistral Medium 3.599.6% Kimi K2.699.6% Qwen 3.7 Max99.7% Gemma 499.6% Nemotron 3 Ultra99.7% GLM 5.299.5% Our product Detect AI and plagiarism Pangram combines cutting-edge AI and comprehensive plagiarism detection to give you the complete picture of text authenticity and get the information you need, all in one place. Pangram also detects AI-generated images. AI Segment Analysis Pangram won't just tell you how confident it is that the document is AI, but also how much of it is. AI Assistance Pangram will highlight segments that have been edited by AI rather than fully AI-generated. Plagiarism Checker Check for plagiarism and AI all at once. Perform an AI plagiarism check with one click. Features How is Pangram's AI detector different? There are many AI detectors available, but Pangram is different. It actually works. We also provide advanced features backed by peer-reviewed research that allow users to understand the origin of a piece of writing. Years of research Uses proprietary technology with years of research and development behind it, not an open source model or commercial LLMs covered with branding. Trained on diverse data sets Pangram relies on diverse datasets, hard negative mining, and active learning to achieve industry-leading FPRs rather than perplexity and burstiness, which often fail in practice. Detects all LLMs Pangram can detect content from all major language models including ChatGPT, Claude, Gemini, Llama and more, making it a comprehensive solution for AI detection. Supports 20+ languages Our AI detection works across more than 20 languages, making it a truly global, multilingual solution for institutions and businesses worldwide. Out-performed trained humans Independent studies have shown that Pangram's AI detection outperforms trained human readers in identifying AI-generated content. Detects humanized text Pangram can detect AI-generated text even after it has been "humanized," or processed by tools that attempt to evade AI detection, ensuring reliable detection. Verified accuracy Our accuracy and reliability have been independently verified by third-party researchers and educational institutions. Who we are About Pangram At Pangram, our mission is to rebuild trust amidst the increase in generative AI content. Founded in Brooklyn, NY, in 2023 by AI researchers from Tesla and Google, we continue to contribute to the research community to increase transparency. If you're interested, read more about our contributions and collaborations. How it works How Pangram's AI detector works 01 Input your text Copy and paste your text into the field or upload files. Pangram supports PDF, DOCX, RTF files and up to 100 files at a time. 02 Scan for AI content Pangram analyzes your text using advanced AI detection algorithms to identify AI-generated content with high accuracy. 03 View results Review detailed results showing AI detection scores, highlighted sections, and comprehensive analysis of your text. 04 Access past records Access your complete history of scans, download reports, and manage your past detection records all in one place. Pangram is also integrated into Chrome browser, Google Docs, and many learning management systems like Google Classroom and Canvas. 01 Input your text Copy and paste your text into the field or upload files. Pangram supports PDF, DOCX, RTF files and up to 100 files at a time. 02 Scan for AI content Pangram analyzes your text using advanced AI detection algorithms to identify AI-generated content with high accuracy. 03 View results Review detailed results showing AI detection scores, highlighted sections, and comprehensive analysis of your text. 04 Access past records Access your complete history of scans, download reports, and manage your past detection records all in one place. Pangram is also integrated into Chrome browser, Google Docs, and many learning management systems like Google Classroom and Canvas. Solutions Protect authenticity in every industry Pangram provides AI detection solutions for educators, publishers, law firms, HR departments, foundations, and more. For teachers Pangram's transparency tool reliably detects AI content and checks for plagiarism to help teachers understand the origin of written text and maintain integrity and fairness in assessment. For higher ed Pangram's LMS integrations allow universities to create informed AI policies. Our AI detector's tool accuracy gives them the peace of mind to implement those policies at scale. Features AI detection solutions for enterprise and schools 01 LMS Integration Our AI detector works with the most popular LMS systems making it easy to use across multiple platforms like Canvas, Moodle, Google Classroom, Brightspace, and more. Learn More → 03 API Solution Our API is perfect for high-volume projects with different pricing options to suit your needs. We have custom API solutions for educators, developers, and enterprise. Learn More → 02 Google Doc & Browser Extension Our browser extension is a quick and convenient way to use our AI detection tool while browsing the web or within web-based applications such as Google Docs. Download Now → See what our users say Users love Pangram Alex Imas, Professor Pangram is pretty incredible. Much better than other AI detection tools (by a country mile). I've used it and tested it thoroughly and seen minimal false positives and negatives. University of Chicago Booth School of Business For me, Pangram Labs has become the most student-centered option I’ve found for my courses. Students see the same report I see, directly in our Canvas shell, and Pangram’s documented false-positive rate (1 in 10,000) allows me to comfortably treat a flag as an invitation to talk, never as a presumption of guilt. — Susan Ray, Associate Professor Of English, Delaware County Community College There's a lot of overhyped, me-too products flooding the AI space, but this detector is almost supernaturally good. I have yet to experience a single false positive or false negative. Cloud Architect — Ryan Nicolace, Cloud Architect Jenna Russell, PhD We compare human experts with five AI detectors, including Pangram and GPTZero. Among automatic detectors, Pangram significantly outperforms the rest. Tuhin Chakrabarty, Professor Pangram is really one of the best AI detectors we have, and y'all should use it. Based on our experiences in the fall term, we have high confidence in Pangram as a detector of AI content in Wikipedia articles. — LiAnna Davis, Chief Programs Officer and Deputy Director at Wiki Education Pangram noticeably outperforms other LLM content detection services I've tried. It reduces our moderation workload enough to pay for itself many times over. If you run a platform with user-generated content, you should probably be using it. — Robert Mushkatblat, CTO LessWrong Dan Simon, CEO of Qwoted The future of journalism depends on trust. This is why we're delighted to be partnering with Pangram, which has set the gold standard for Al detection and attribution Our model Pangram's AI detection is highly accurate How does Pangram achieve these results? Learn more. 01 Initial training process for AI detection Our classifier uses a traditional language model architecture. It receives input text and tokenizes it. Then, the model turns each token into an embedding, which is a vector of numbers representing the meaning of each token. The input is passed through the neural network, producing an output vector. A classifier head transforms the output vector into a prediction of human, AI, or AI-assisted. We train an initial model on a small but diverse dataset of approximately 1 million documents composed of publicly licensed human-written text. The dataset also includes AI-generated text produced by GPT-5 and other frontier language models. The result of training is a neural network capable of reliably predicting whether text was authored by human or AI. 02 Continued improvement through iteration After the initial training, our model undergoes continuous improvement through iterative refinement. We collect feedback from real-world usage and incorporate new data patterns to enhance detection accuracy. This iterative process ensures our AI detection remains at the forefront of technology, adapting to new language models and writing styles as they emerge. 03 Consistent retraining of AI detector models To maintain the highest level of accuracy, we consistently retrain our AI detector models with updated datasets that include the latest AI-generated content. This regular retraining process ensures our detection capabilities stay current with evolving AI technologies and can reliably identify content from the newest language models. 01 Initial training process for AI detection 02 Continued improvement through iteration 03 Consistent retraining of AI detector models Our classifier uses a traditional language model architecture. It receives input text and tokenizes it. Then, the model turns each token into an embedding, which is a vector of numbers representing the meaning of each token. The input is passed through the neural network, producing an output vector. A classifier head transforms the output vector into a prediction of human, AI, or AI-assisted. We train an initial model on a small but diverse dataset of approximately 1 million documents composed of publicly licensed human-written text. The dataset also includes AI-generated text produced by GPT-5 and other frontier language models. The result of training is a neural network capable of reliably predicting whether text was authored by human or AI. Third Party Evaluation Why Pangram has the lowest false positive rate Based on independent research from the University of Chicago Booth School of Business. Learn more. 01 Product review Pangram 0.5% GPTZero 2.4% Originality.ai 1.7% RoBERTa 30.6% False positive rate — Product review 02 Blog post Pangram 0% GPTZero 0.5% Originality.ai 0% RoBERTa 64.9% False positive rate — Blog post 03 News article Pangram 0.1% GPTZero 1% Originality.ai 0.3% RoBERTa 77.8% False positive rate — News article 04 Novel excerpt Pangram 0% GPTZero 0.5% Originality.ai 0.3% RoBERTa 53.2% False positive rate — Novel excerpt 05 Restaurant review Pangram 0.8% GPTZero 1% Originality.ai 2.2% RoBERTa 51.7% False positive rate — Restaurant review 06 Résumé Pangram 0% GPTZero 0% Originality.ai 0.1% RoBERTa 69.1% False positive rate — Résumé 01 Product review 02 Blog post 03 News article 04 Novel excerpt 05 Restaurant review 06 Résumé Pangram 0.5% GPTZero 2.4% Originality.ai 1.7% RoBERTa 30.6% False positive rate — Product review FAQs AI Detection FAQs Looking for answers? Explore common questions and get the information you need, all in one place. 001 What is AI content detection? 002 How does Pangram work? 003 How accurate is Pangram? 004 What is Pangram's false positive rate? 005 Is Pangram's AI detector free? 006 Is my data safe if I check text with Pangram? 007 What languages does Pangram detect AI content in? 008 What AI models does Pangram detect? 009 What signals does Pangram use to detect AI text? 010 How are schools using Pangram? 011 How are universities using Pangram? 012 Why do other AI detectors sometimes give false positives? 013 Why do other AI detectors fail to flag AI writing? 014 Why is Pangram the best AI detector? The Pangram blog More from our AI experts All About False Positives in AI Detectors A deep dive into how Pangram achieves industry-leading false positive rates in AI detection, with a false positive rate of 1 in 10,000. What is AI watermarking? How AI text watermarks like SynthID and Claude’s work, which providers use them, how they can be beaten, and what they mean for AI detection. Pangram's Performance on a Colluding Agent Swarm A brief report on how Pangram 4 performs when given a collection of items written by AI agents using an obscure web forum to help each other cheat on their assigned task. Does Pangram Detect Fable 5.1? Can Pangram 4 detect Anthropic's newest model, Fable 5. 1? AI Helps Write Nearly 50% of Press Releases Pangram scanned 178,672 PR Newswire press releases and found AI-generated text in 34% of them over the past year, climbing to 48% by June 2026. No, LLMs don’t just mimic human text LLMs are trained on human text, so why can AI detectors still tell the difference? A look at how post-training pushes models like ChatGPT, Claude, and Gemini away from the human writing distribution. Pangram shows 0 false positives on diverse student essays We evaluated Pangram 4 on 25,996 human-written student essays from the PERSUADE 2. 0 corpus and found zero false positives in every demographic subgroup. Pangram's Model Market Share How Pangram's models can show trends in AI provider market share. Pangram 4 Migration Guide All you need to know to migrate to Pangram's latest flagship model!