By Teresa Torres · producttalk.org · @ttorres on X · LinkedIn
Teresa Torres introduces a new series, Conversations with Claude, showing how she uses Claude Code for real work. Her first example is a content audit of Product Talk, which she delegated to Claude with a task file and keyword-research tools. Claude extracted roughly 100 keywords from her CDH book, checked search volumes and the site's rankings, and identified high-volume gaps where her content did not rank. It also suggested both content updates for ranking pages and entirely new articles. The piece matters because it shows an AI-assisted workflow for SEO and content strategy that many product writers neglect, while noting that a generic prompt alone would produce weak results.
01Key takeaways
- Start AI-assisted audits with clarifying questions and explicit decision rules, such as what counts as ranking well.
- Use high-volume generic terms as entry points, then frame them with your specific product perspective.
- Prioritize gaps around your central thesis, since they often yield the biggest audience opportunity.
- Give the model a task file and context files rather than relying on a bare prompt to get useful results.
- Treat AI output as a starting plan to discuss and refine, not a final decision.
02Key sections
- Why this series exists
- Torres explains she wants to show the practical why behind Claude Code, not just setup steps. She also discloses she has no compensation from Anthropic.
- Setting up the audit
- Claude asked clarifying questions about keyword scope, geography, what counts as well-covered content, and output format before starting work. Torres answered with specific rules for judging rankings.
- What the audit found
- Claude reported strong branded rankings but large gaps on central concepts like outcomes versus outputs, product roadmaps, and story mapping. It proposed a trojan-horse strategy of writing on broad, high-volume terms with a discovery angle.
- Why generic prompting falls short
- Torres stresses that simply asking Claude to audit a site produces weak output, and that the quality came from the context and iteration she set up beforehand.
03From the post
“During one of our recent Claude Code Office Hours, a participant asked, "Can we zoom out? Help me understand what I can use Claude Code for." This is a great question. I was so focused on how to help people set up Claude Code that I forgot to share the”
04Frameworks mentioned
Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.