The team at LinkedIn launched a new job search experience where members can describe the job they want using free-form language instead of relying purely on keywords or filters.

Behind the scenes the system uses embedding-based retrieval (mapping text queries and job postings into a shared vector space) and large language model (LLM) techniques to interpret the user’s description and map it to relevant postings.

For example, instead of typing “software engineer Vancouver” the user might type “entry level backend engineer in Vancouver building APIs” and the system converts that description into its semantic representation, retrieves jobs whose embeddings align, and ranks them by relevance.

The ranking then uses typical supervised learning signals (clicks, applies) and may incorporate reranking via LLM-generated candidate suggestions (assumption: based on standard search + reranking pipelines [S1]).According to LinkedIn’s announcement, this shift is meant to reduce search friction and increase engagement by making job discovery closer to how people naturally express their aspirations.

While exact metrics weren’t publicly shared in the article, the goal is to move from keyword-based rigid search to a more flexible conversational experience and thereby convert more searches into quality applications.

Key insight: The approach works because it aligns with how people think (in sentences not rigid filters), leverages embeddings + NLP to bridge between user intent and job listings, and then uses behaviour data to continuously refine the ranking. The causal mechanism: less friction in expression → more relevant matches → higher engagement.