Making a Content Library AI-Ready
Disciplines: Content strategy · Information architecture · AI/RAG retrieval · Team enablement
TurboTax’s help content library was written for people. As AI assistants became the front door to customer support, that same content had to work for machines too—and a lot of it didn’t. I led the effort to make the library AI-ready, driving the structural and editorial changes that let AI assistants generate trustworthy responses.
Sometimes the strategies that make excellent content for a human reader, like expandable sections and info-rich tables, are horrible for AI retrieval. An AI agent might pull half an answer or miss information entirely.
The challenge I faced: make a large, mature content library retrievable and trustworthy for AI without breaking the human experience.
This is the work I’m most excited to keep doing: sitting at the intersection of content craft and AI systems. I took content design fundamentals (clarity, structure, information architecture) and pointed them at a brand-new problem.
My role
- Driving the standard into practice—taking an AI-readiness standard and making teachable, applicable, and enforced across thousands of articles.
- Coaching the team—leading a team of writers through new technical structural changes most had never encountered.
- Closing the loop—partnering with data science and machine learning engineering to help define how we’d measure the chatbot’s accuracy.
My approach
1. I turned a standard into something a team could actually apply. I co-authored our AI-readiness playbook and then it needed was someone to operationalize it at scale. I built the standards into the training and workflow for the annual content refresh—including building a NotebookLM companion—so every writer applied the AI-readiness lens to every article they touched.
2. I helped define “good.” Making the content retrievable is only the first half the job; you have to ensure the AI is actually doing the retrieving. I partnered with data science and engineering, advocating for robust taxonomy utilization and contributing to the golden dataset used for evaluations.
3. I scaled the thinking beyond my team. I presented AI-content best practices to designers groups across the company to pressure-test the standards’ relevance. Defaulting to collaboration guaranteed the standards didn’t stay siloed on one team.
My impact
- ~90% of the help library reworked through an AI-readiness lens in a single season, on a path to full coverage.
- A durable practice, not a one-off. AI-readiness became part of how the team writes new content so the library stays AI-ready instead of needing another rescue.
- A measurement loop. Developing a rubric to monitor the library’s AI readiness score over time.
Elisse Miller 