Editorial and research assistant for async expert interviews
Build a workflow called Editorial & Research Assistant (ERA) that helps a marketing team turn expert interviews and asynchronous contributions into a source-grounded article outline and first draft — without inventing facts or quotes. The workflow has two web front-ends and two JSON APIs behind them: An internal editorial workspace (React SPA) served at a space-authenticated route — the main UI the team works in. It shows a project dashboard (create from a content-type template, load a demo) and a per-project workspace with tabs that drive the whole pipeline: Brief, Questions, Contributors, Synthesis, Outline, Draft, Handoff, and Activity. It renders immediately and fetches its own data from the editor API. A contributor questionnaire (React SPA) served at a token-gated route (an unguessable per-contributor token in the URL). Contributors see the project brief and their assigned questions and answer each by typing (autosaved) or by voice — recording in-browser with live speech transcription, or uploading an audio file that gets cleaned into a transcript server-side. A backend API (space-authenticated) powering the editor: create projects, generate and approve an interview questionnaire, invite contributors, synthesize submitted responses, generate and approve an editorial outline, and generate a first draft that includes inline flags marking claims that need fact-checking or missing input. A contributor API (token-gated) powering the questionnaire: serve the brief and questions, save typed answers, accept audio, clean transcripts, and mark responses submitted. The end-to-end flow: create a project from a content-type template → generate and approve a questionnaire → invite contributors (via chat or a shareable link) → contributors respond by text or voice → synthesize the responses → generate and approve an outline → generate a first draft with fact-check/needs-input flags → hand off to a document and notify collaborators. Use a large language model for question generation, response synthesis, outline and draft creation, and transcript cleaning — I used Anthropic's Claude, but use whatever model provider you prefer. Use a team chat tool for contributor invites, reminders, and notifications — I used Slack, but any chat tool works. Use a document service for the final handoff — I used Google Docs/Drive, but any doc tool works (fall back to a labelled mock if none is connected). Persist project state, responses, and uploaded audio in durable workflow storage (e.g. a SQLite database plus audio files on a named volume); this workflow reads and writes that state. Tools used: Anthropic (Claude) API, Slack, Google Docs, Google Drive, browser Web Speech API, SQLite
What this prompt builds
Editorial & Research Assistant is a workflow that collects expert input via text or voice, then synthesizes responses into source-grounded outlines and first drafts. It reduces editorial admin, eliminates unnecessary meetings, and helps teams produce stronger content faster without requiring subject-matter experts to write. The workflow handles asynchronous contribution, transcript cleaning, synthesis, and draft generation with inline fact-check flags.
The problem
Content teams who rely on input from busy subject-matter experts struggle with fragmented, slow collection across Slack, email, documents, and voice notes. Editors spend hours chasing responses, transcribing recordings, and synthesizing raw contributions into usable structure. The workflow lets contributors respond by text or voice on their own schedule, then uses AI to synthesize their input into a source-grounded outline and first draft—complete with inline fact-check flags. It eliminates unnecessary sync meetings, reduces editorial admin, and helps teams produce stronger, more accurate content without asking experts to become writers.
Solution and impact
Contributors answer interview questions by typing or recording voice responses in a browser-based questionnaire, with autosave and live speech transcription. The workflow then synthesizes all submitted input into a factual outline and first draft, marking claims that need verification or areas requiring additional input. Editors save hours per article, maintain source accountability, and hand off polished drafts to collaborators—all without coordinating calendars or scheduling interviews.