A site's internal search log is a direct record of the actual words people use to describe what they want, arguably one of the most honest sources of vocabulary data available, since users type these terms with no awareness anyone is analyzing them. Lou Rosenfeld describes five approaches to analyzing this data: pattern analysis, which looks at trends in search behavior, vocabulary, and timing; failure analysis, which focuses on queries that return zero or poor results: these are top priority, since they represent confirmed, immediate findability failures; session analysis, which follows a single user's sequence of queries to see how their understanding evolves within one visit; audience analysis, which segments search behavior by user group; and goal-based analysis, which ties search performance directly to business KPIs.
Good search-log analysis balances two competing goals: precision, the percentage of returned results that are actually relevant, and recall, the percentage of all relevant content that the search actually surfaces. Learnings from search logs feed directly back into the rest of the IA work covered in this chapter, aligning controlled vocabulary and labels with users' real terms, identifying missing content, and informing where navigation or faceted search needs improvement.
A zero-result search query is one of the rare moments where a user has told you, in their own words, exactly what they wanted and exactly where you failed to give it to them.
"Renew membership" should be prioritized first. Failure analysis treats zero-result and poor-result queries as top priority, and at 400 searches a month, this query represents by far the largest volume of confirmed, immediate findability failure on the site, twenty times the volume of "code of ethics" and more than double "conference schedule."
The two zero-result queries fail for genuinely different reasons, which is exactly why failure analysis matters: it doesn't just count failures, it diagnoses their cause. "Renew membership" fails because of a labeling mismatch, the content exists, but under a name, "Annual Subscription Continuation," that doesn't match how members actually describe the action. "Code of ethics" fails because the content simply doesn't exist yet, no page has ever been written to answer that query.These require different fixes entirely. The labeling mismatch can likely be resolved quickly, by renaming the page, adding "renew membership" as a synonym in the site's controlled vocabulary, or both, without creating any new content. The missing content problem requires an entirely different kind of work: someone has to research, write, and publish a code of ethics page before any search or labeling fix could possibly help. Treating both as the same type of "zero-result problem" would miss that one is a same-day fix and the other is a content-development project.