Search

LinkedIn Help says the search bar accepts natural, conversational language, and LinkedIn Engineering and LinkedIn research describe ways to read what a query means instead of only matching its words11011. LinkedIn Engineering wrote in August 2024 that post searches had grown longer and more natural, such as “how to ask for a raise?”, and that search sometimes returned no posts because none held every keyword, or returned posts with every keyword that did not answer the question10. The Posts search system retrieves candidate posts using both keywords and meaning, then ranks them for how well they answer the query and the time searchers spend reading them10. A February 2026 LinkedIn research paper says keyword-based retrieval struggles to capture natural language intent, which motivated rebuilding Job Search and People Search around large language models, which are AI text models, and a November 2025 LinkedIn research paper says its Semantic Job Search relies less on exact word matches between queries and job descriptions1112. A Microsoft patent from December 2021 describes a user typing a sentence, with the system maintaining context through a series of searches13. A December 2020 Microsoft patent describes rewriting search queries using related job functions, and a September 2026 Microsoft patent describes a system-generated modified query that users can refine with filters1415.

Search suggestions appear in a dropdown as a member types, and Microsoft patents describe methods for suggesting related searches and query refinements11617. A January 2026 Microsoft patent describes suggesting related searches with the goals of making suggestions natural, relevant, error-free and engaging; its example suggests “machine learning engineering” for “machine learning engineer”16. A July 2020 Microsoft patent describes finding a person by name when only the sound is known, by grouping names that sound alike and suggesting the spellings from the matching group18. An August 2019 Microsoft patent describes suggesting an extra category, such as title, company, industry, school or location, to add to a recruiter's search17. The September 2026 Microsoft patent also describes a panel of suggestions for improving a query, built from signs such as the user scrolling without selecting many results15.

LinkedIn Engineering and LinkedIn research describe how AI ranking of search results, which scores many options against the same query to return the best matches, is made efficient enough for Job Search and People Search at scale51112. LinkedIn Engineering wrote in February 2026 that this ranking needs only the initial understanding step of an AI text model and none of the text writing that chatbots do5. A February 2026 LinkedIn research paper describes making AI-based search feasible at hundreds of thousands of queries per second11. A November 2025 LinkedIn research paper on its Semantic Job Search says long job descriptions made ranking too slow12. The paper describes MixLM, which condenses job descriptions into stored numeric summaries ahead of time so the ranking model reads the query and summaries instead of every full description, enabling deployment across all traffic12.