Startup Rebuilds Search: From Keyword Chaos to Semantic AI Ranking

AI · 6 min read

Users had been complaining that search returned syntactically similar but irrelevant listings; keyword matching favored duplicate or keyword-stuffed titles. Conversion from search to view dropped and search-plus-filter funnels were confusing for less experienced users. The product team prioritized relevance over completeness and set out to rebuild ranking with a semantic-first approach.

Engineers implemented vector embeddings for titles, descriptions, and tags, while designers introduced a simple query intent selector (Buy vs. Research vs. Compare) and dynamic suggestion chips that provided quick disambiguations. Results were grouped into 'Best match', 'Nearby alternatives', and 'Recently added' to help users scan options. The UI also clearly displayed why a result matched by highlighting semantic tags rather than raw keywords.

Post-launch, click-through for the top result rose 28%, and conversion from search sessions increased 18%. Search refinement time decreased, and customer satisfaction surveys reported clearer relevance. The team emphasized that semantic ranking is only half the solution—UX changes that explain why results were chosen and provide immediate corrective actions are equally important.