What a search system actually searches shapes everything about its quality. Indexing specific semantic elements, a title field versus a body field, lets a system weight matches differently depending on where a term appears. Search zones divide content into separate areas that can be indexed and searched with different rules. Latent semantic indexing goes further still, using statistical techniques to find conceptually related content even when the exact keywords don't match at all.
All of this optimization work has to balance two competing goals: precision, the percentage of returned results that are actually relevant, and recall, the percentage of all genuinely relevant content the search actually manages to surface. Pushing hard on one of these tends to cost you the other, a system tuned to guarantee every result is relevant will often miss some genuinely useful matches, and a system tuned to catch everything relevant will often let some irrelevant noise through with it.
A search system that returns three results, all perfectly relevant, hasn't necessarily done a better job than one that returns thirty, some of them noise, it depends entirely on whether the thing the user actually needed was inside those thirty or left out of the three.
1: A precision problem. Hundreds of loosely related results for a precise query means too much irrelevant material is getting through, the defining signature of low precision.
2: A recall problem. Relevant content genuinely exists but isn't being retrieved because of a terminology mismatch, exactly what a recall failure looks like, and a strong candidate for controlled vocabulary integration to fix.
3: A problem with what's being indexed. The search system is searching the wrong content entirely, navigation text instead of actual page content, which no amount of query refinement could fix, since the real content was never indexed in the first place.
4: A search zones problem. Zones being searched in isolation, rather than together or with appropriate cross-zone weighting, causes exactly this kind of incomplete retrieval for topics that legitimately span more than one zone.
Fixing failure 2 by broadening the system's ability to match different terminology to the same concept, for example, through latent semantic indexing or a controlled vocabulary, could reduce precision as a trade-off. The same broadening that helps someone find content described with different words can also start surfacing more loosely related results for other queries, since the system is now casting a wider net by design. This is exactly the precision-recall tension the concept section describes: the fix for one failure mode can quietly introduce the other, which is why tuning search quality is rarely a one-directional improvement.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.
Generate prioritized, actionable sprint tasks for improving search and findability based on audit findings. Breaks down search improvements into specific, estimable work items suitable for sprint planning.
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.
Casts the AI as: You're a technical content strategist specializing in metadata and SEO for documentation.
Names 4 specific outputs to produce, so the response comes back structured rather than a general summary.
Content type, Target audience, Common search queries and Current issues tell the AI what your specific situation is, not a generic one.
Before/after comparison with rationale