The Unwritten Journey audits the implicit content, subtext, and blind spots in a response produced by an LLM (ChatGPT, Claude, Gemini, Mistral…).
A response can be convincing while never signaling what it leaves out: unstated presuppositions, implicit logical steps, or aspects of the topic left unaddressed.

The tool draws on Arthur Sarazin’s taxonomy of the implicit and detects 6 categories of unstated content: connotation, subtext, implicature, de facto, implementation detail, and pure omission.

Hosted in the European Union (Render, Frankfurt) and powered by the openai/gpt-oss-120b model via the Albert API (DINUM/Etalab), French sovereign AI infrastructure.

Feel free to share your comments or suggestions by email or on social media!

Made available under a Creative Commons CC BY 4.0 license and available on Github.