Prompt Fan-out
A query expansion strategy used in AI search to improve retrieval coverage. It decomposes the user's original question into multiple mutually independent sub-queries that can be executed in parallel, retrieves from different data sources separately, and then integrates the results into a comprehensive answer. Compared with query rewriting and query expansion, it emphasizes parallel decomposition and multi-source integration.
Timeline (last 90 days)
Prompt fan-out technology refers to starting from a seed question and automatically generating a set of semantically related sub-questions from different angles, usually based on the semantic expansion capability of large language models or aggregated analysis of search suggestions and community Q&A; the generated prompts should be manually screened to ensure they are both comprehensive and actionable.
A single user question may trigger five to ten searches, and which specific sub-queries are chosen is decided by the system, may differ from run to run, and users typically cannot see the full list of sub-queries.
A method for choosing tracking prompts from the Fan-out perspective: start from buyer questions (10 to 15 initially), expand each question into sub-queries (the five dimensions of attributes, comparisons, evidence, use cases, and constraints), turn quality sub-queries into natural-language prompts, group them by category, and fix the list for continuous tracking.
The limitations of Fan-out include: you cannot see each sub-query (generators and manual expansion give possible sub-queries, not the exact list the system actually runs), patterns change over time, and covering sub-queries does not force a mention.
All Facts (4)
Prompt fan-out technology refers to starting from a seed question and automatically generating a set of semantically related sub-questions from different angles, usually based on the semantic expansion capability of large language models or aggregated analysis of search suggestions and community Q&A; the generated prompts should be manually screened to ensure they are both comprehensive and actionable.
50%UnverifiedA single user question may trigger five to ten searches, and which specific sub-queries are chosen is decided by the system, may differ from run to run, and users typically cannot see the full list of sub-queries.
50%UnverifiedA method for choosing tracking prompts from the Fan-out perspective: start from buyer questions (10 to 15 initially), expand each question into sub-queries (the five dimensions of attributes, comparisons, evidence, use cases, and constraints), turn quality sub-queries into natural-language prompts, group them by category, and fix the list for continuous tracking.
50%UnverifiedThe limitations of Fan-out include: you cannot see each sub-query (generators and manual expansion give possible sub-queries, not the exact list the system actually runs), patterns change over time, and covering sub-queries does not force a mention.
50%