IdeaLens detects AI-generated ideas by analyzing document outlines

Pangram: Pangram is a prose provenance detector that identifies whether text was written by AI or humans based on the wording and style of the document. It was used in the development of IdeaLens to generate silver labels for training data and serves as a baseline comparison in evaluations. Unlike idea-focused detectors, it primarily flags content based on linguistic patterns rather than the underlying ideas. IdeaLens: IdeaLens is a detector designed to determine the provenance of ideas in long-form writing by analyzing a document's outline of key points rather than its full text or wording. It was developed to address scenarios where AI contributes ideas that humans then write up in their own words. The tool was introduced in a research paper examining its performance across controlled studies and benchmarks for distinguishing human versus AI ideation. Idea Provenance: Emerging policies on AI use increasingly focus on who originated the ideas in a document rather than solely on who wrote the words. Detection Limitations: Detectors that analyze only wording often flag AI-polished human work while missing cases where humans write up AI-generated ideas themselves. Outline-Based Analysis: Representing documents as outlines of discourse roles and paraphrased content allows detectors to isolate idea-level signals from surface-level text.