Semantic session search
Semantic session search is querying a session archive by meaning rather than metadata: "sessions where users struggled to apply a coupon" instead of filters over URLs and dates. It converts the archive from a store you retrieve from into a corpus you ask questions of.
How it works
The standard architecture embeds derived representations of sessions (summaries, digests, or windowed segments) into a vector space and retrieves by similarity to the embedded query, usually with metadata filters running alongside for the constraints that are genuinely structural (date ranges, user segments). The derivation layer decides what is findable: embedding raw event JSON finds little, embedding behavioral digests finds behavior, and embedding summaries finds whatever the summarizer thought worth mentioning, so search quality is largely a restatement of session summarization quality. The retrieval loop closes into analysis: a semantic query defines an ad hoc segment, and the segment feeds the same watching and measuring workflows that predefined funnels do, which is how questions nobody anticipated at instrumentation time still get answered.
Where you encounter it
AI replay platforms, rrweb AI's query interface, and internal tools built over session embeddings.