Industry leading research to identify AI generated text.

Our models are trained to classify text as human- or AI-written. We take in a document or piece of text and return a prediction of the likelihood that it was generated by an LLM.

Need inspiration? Try an example.
We're focused on delivering the
most accurate results.

Pangram's text classifier was developed by researchers with deep experience building AI models at companies like Tesla and Google. We set ourselves apart with a strong commitment to exceptional overall accuracy.

Read more about our performance in our Technical White Paper
Leading performance
Our text classifier exhibits a 9x lower error rate in a comprehensive evaluation against other leading AI-generated text detectors, winning on both false positive rate and false negative rate.
Evaluated on ten categories of text such as reviews, news, emails, and essays, our text classifier performs at 99% or higher accuracy across all categories, with less than 1% false positive rate.
Evaluated on the output of eight leading language models, our text classifier outperforms competing AI detectors with over 99% recall.
Fewer Documents
Enterprise ready robustness
Pangram's text classifier correctly categorizes ChatGPT, GPT-4, LLaMA, Bard, Claude, Mistral, and many more available public LLMs.
We catch AI-generated text that has been paraphrased with a tool such as Quillbot or Undetectable AI.
Our classifier is robust to cowritten text, minor changes, and artificially added typos.
Multi-domain Efficacy
Our models are trained on a diverse set of text content across different writing tasks.
Our classifier significantly outperforms other detectors across all domains evaluated, including news, email, and reviews.
We tune our models to incorporate new LLMs immediately as they become publicly available.