Good labels rarely come from a single source. Analyzing existing content, page titles, subheadings, summaries, surfaces the vocabulary already in use. Competitive analysis reveals the conventions an entire domain has already converged on. User research, particularly open card sorting, captures the actual language real users apply when they group and name content themselves. Domain-specific controlled vocabularies or thesauri, where they already exist, bring a level of consistency that's hard to match from scratch. Subject-matter experts, technical support staff, and content authors all carry vocabulary a design team wouldn't otherwise have access to. And search-log analytics, the terms people actually type when looking for something, are some of the most honest evidence available, since nobody phrases a search query to be polite or on-brand.
No single source is sufficient on its own. Combining several, and cross-checking the terms they surface against each other, is what actually produces a labeling system grounded in reality rather than internal assumption.
The team that already works inside an organization is often the worst-positioned group to guess what outsiders will call something, which is exactly why label discovery has to reach beyond the room the decision is being made in.
The research-heavy combination is correct. This domain has exactly the characteristics that call for direct engagement with real users rather than assumption: contested vocabulary, real stakes if terminology lands wrong, and a diverse audience whose members may not all share the same relationship to clinical versus community language. An open card sort captures how actual community members group and name these concepts themselves, user research surfaces their lived vocabulary directly, and subject-matter expert consultation helps the team understand the difference between clinically accurate and community-accepted terms, a distinction that matters enormously here and that internal guesswork alone could easily get wrong.
Relying on competitive analysis alone would be particularly risky in this specific domain. Other mental health platforms may themselves be using outdated or clinically outmoded terminology, and simply adopting whatever labels are already common elsewhere risks propagating language the target community has already moved past or actively finds stigmatizing. Competitive analysis remains a useful input for understanding general conventions, but it shouldn't be the primary or sole source of labels here, it needs to be checked against direct community research, not treated as sufficient on its own.This is a case where the general principle from the concept section, combine multiple sources rather than relying on one, matters more than usual, precisely because the cost of a wrong label in this domain isn't just user confusion, but the risk of genuinely alienating or hurting the people the platform exists to serve.
Fill in your own details below; the prompt updates as you type. When it's ready, copy it into Claude or whatever AI tool you use.
Synthesizes survey responses into IA-relevant themes with counts, quotes, and recommended actions.