IdeaLens: Detecting AI Ideas in Long-form Writing
arXiv cs.CL · 6 Oct 2026, 03:43 AEST
IdeaLens is a new detector that identifies whether the ideas in a document originated from a human or AI. It focuses on idea provenance rather than authorship, using outlines to minimize word-level overlap with the text.
Key facts
- We introduce IdeaLens, a detector that identifies whether a document's ideas came from a human or AI (idea provenance), regardless of who wrote its words.
- In a controlled study, IdeaLens's AI flag rate drops from 95% to 7% as models write from increasingly detailed human plans, while Pangram 4 still flags 92%; from AI-derived plans, IdeaLens stays above 96%.
- Conversely, on a new dataset of 50 stories that human authors wrote from AI-generated plans, IdeaLens flags 68% of the stories as AI, compared to 8% for Pangram 4.
- On a comprehensive suite of 19 existing detection benchmarks, we show that IdeaLens maintains strong detection rates at low false positive rates, suggesting that ideas themselves provide a powerful discriminative signal.
Read aloud
IdeaLens is a new detector that identifies whether the ideas in a document originated from a human or A I. It focuses on idea provenance rather than authorship, using outlines to minimize word-level overlap with the text. We introduce IdeaLens, a detector that identifies whether a document's ideas came from a human or A I idea provenance, regardless of who wrote its words. In a controlled study, IdeaLens's A I flag rate drops from ninety five percent to seven percent as models write from increasingly detailed human plans, while Pangram four still flags ninety two percent; from A I-derived plans, IdeaLens stays above ninety six percent. Conversely, on a new dataset of fifty stories that human authors wrote from A I-generated plans, IdeaLens flags sixty eight percent of the stories as A I, compared to eight percent for Pangram four. On a comprehensive suite of nineteen existing detection benchmarks, we show that IdeaLens maintains strong detection rates at low false positive rates, suggesting that ideas themselves provide a powerful discriminative signal.
Sources in this cluster
- IdeaLens: Detecting AI Ideas in Long-form Writing
arXiv cs.CL · 6 Oct 2026, 03:43 AEST