A prompt is input text that gives a language model instructions, questions, or contextual material. In search, it can combine a user query with retrieved documents and directions for producing a grounded summary. section 15.2.2 section 2.5.2
Prompts supply context for interpreting tokens: surrounding words influence how each word part is understood. A search workflow first retrieves relevant results, then constructs a prompt containing the query, those results, and an instruction such as “Be concise.” section 2.5.2 section 15.2.2
Effective search prompts make the requested output and evidence explicit:
- Include retrieved results with source URLs or citation markers so the generated answer can be checked. section 15.2.2
- Treat the prompt as an influence on the answer, not as a guarantee of factual accuracy: models can hallucinate or reflect wording and bias. section 15.2.2
Search summarization prompt
A compact prompt can provide retrieved web results followed by: “Query: What is a large language model? Be concise.” The model then summarizes the results and attaches citation markers.
