Inline annotations track AI writing changes and origins
Who Asked for This? Inline Annotations as Authoring Transactions for Provenance in Agentic Authoring
Human-Computer Interaction
Summary
When people write using AI helpers, each part of the text can come from many different AI requests, but it’s usually unclear who or what caused each change. The authors built a system called Reactant that lets writers add special notes inside their documents to mark which AI request made which change. This system keeps a verified record tracking every addition, deletion, or edit at the word level, making it easy to see the history of changes. The authors tested Reactant on their own paper and with colleagues who used it for months, showing it helps understand the past work and reuse AI skills.
What this means in practice
- •For technical writers: Track and display exactly which AI request generated each change inside collaborative documents to improve editing transparency.
- •For software developers: Build AI-assisted writing tools that record and verify detailed edit histories tied to specific commands and AI skills for better debugging and reuse.
Authors
Chang Xiao
Abstract
Writing with AI agents turns a paragraph into the outcome of many requests, yet the finished document rarely explains which request produced which change. We introduce Reactant, an interaction paradigm in which authors place typed inline annotations in their original documents. A verified transaction protocol records each request, skill identity, and the correspondences that identify additions, deletions, and transformations. The kernel validates the witness against the recorded states to establish word-level longitudinal lineage. We demonstrate Reactant through this paper's revision history, and report four months of three colleagues' self-directed use. Their uses include conversational inquiry into history and deriving reusable skills from recurring requests, illustrating how the transaction record serves as an extensible substrate for agentic authoring.