AI writing has detectable patterns

AI-generated text contains structural, stylistic, and semantic patterns that can be identified by trained software. Detection is not magic, it actually works.

“AI detection is not magic. It can actually work.”

Pangram uses advanced detection

Pangram analyzes text using natural language processing and massive datasets of human and AI writing from models like ChatGPT, Gemini, Grok, Llama, and Claude.

Third-party verified accuracy

Researchers at the University of Chicago and University of Maryland have independently verified Pangram's accuracy, confirming it leads the industry.

Content transparency matters now

As AI-generated content spreads through academia, media, and business, distinguishing human from machine text builds trust between students and teachers and maintains journalism integrity.

Detects 26 AI models

Pangram benchmarks against 26 frontier models including GPT-5.6, Claude Opus 5, Gemini 3, and Grok 4.3, with accuracy ranging from 99.4% to 99.9%.

Combines AI and plagiarism checks

Pangram detects both AI-generated content and plagiarism in one tool, plus identifies AI-assisted edits and analyzes what percentage of text is AI-written.

Proprietary technology, not borrowed

Pangram uses proprietary technology developed over years of research, not open-source models. It relies on diverse datasets and hard negative mining for industry-leading false positive rates.

Outperforms trained humans

Independent studies show Pangram's AI detection outperforms trained human readers at identifying AI-generated content.

Catches humanized AI text

Pangram detects AI-generated text even after it has been processed by tools designed to evade detection, ensuring reliable identification.

Works in 20 plus languages

Pangram's AI detection works across more than 20 languages including English, Arabic, Chinese, French, German, Japanese, Spanish, and many others.

Founded by AI researchers

Pangram was founded in Brooklyn in 2023 by AI researchers from Tesla and Google with a mission to rebuild trust amid rising generative AI content.

Four-step detection process

Users paste or upload text, Pangram scans for AI content, displays detailed results with highlighted sections, and stores records for future access.

Integrates everywhere you work

Pangram works with Chrome, Google Docs, Canvas, Google Classroom, Moodle, and Brightspace, making it easy to use across multiple platforms.

Lowest false positive rate

Pangram's false positive rate is 1 in 10,000, compared to competitors like GPTZero at 2.4% and Originality.ai at 1.7% on product reviews.

Teachers trust the results

Pangram's documented low false-positive rate allows educators to treat AI flags as invitations to talk, never as presumptions of guilt.

“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.”

Trained on 1 million documents

Pangram's classifier was initially trained on approximately 1 million diverse documents of human-written text and AI-generated content from frontier models.

Continuously improves detection

Pangram undergoes iterative refinement using real-world feedback and consistently retrains models with latest AI-generated content to stay current.

Free plan with 20 daily checks

Pangram offers 20 free checks per day after signup, including AI assistance detection. Subscriptions unlock additional credits and plagiarism detection.

Your data stays private

All user data is encrypted and stored securely. Pangram never shares or sells data to third parties.