The Problem

Writing tools help you say what you mean. But you can’t know how your writing lands until someone reads it — and by then it’s too late.

Most AI writing assistants correct grammar or polish prose. None of them tell you where a busy reader will stop reading, where a skeptic will push back, or where your opening fails to earn attention.

The Approach

Build a diverse synthetic audience. Not one AI reader — a panel of personas with different reading behaviors (skimming, skepticism, emotional response). Each one finds the exact passage where they get confused, lose interest, or misinterpret your intent.

This isn’t editing. It’s stress-testing.

What I Built

An Obsidian plugin that analyzes your active note against multiple reader personas in parallel:

• Core personas that always run — The Skimmer, The Scanner, The Busy Reader, The Hook Judge, The Challenger, The Voice Editor
• Dynamic personas selected per-analysis based on content — The Skeptic, Literal Reader, Domain Outsider, Emotional Reader, Cringe Detector, and more
• Custom personas — drop a YAML file to define your own

Each persona reports where they stopped and why. Click any finding to jump to and highlight the passage in your note. Re-analyze after edits to track what improved.

Tech Stack

• Obsidian Plugin: TypeScript, Obsidian API
• Chrome Extension: TypeScript, Chrome Extensions API
• LLM: Provider-agnostic — Anthropic, OpenAI, Google, or any OpenAI-compatible endpoint
• Distribution: Obsidian plugin available via BRAT or manual install

How It Fits

First Misread lives where the writing happens — inside Obsidian for long-form drafts, and in the browser for anything else. Draft → run the panel → find blind spots → redraft. It catches the gaps that self-editing can’t — because you can’t unsee your own intent.