How can you refine a query if DeepResearch returns a report that is too broad or, conversely, too narrow in scope?

To refine a query when DeepResearch returns a report that is too broad or too narrow, start by analyzing the initial results to identify gaps in specificity or relevance. If the report is too broad, add constraints or keywords to narrow the scope. For example, if searching for “machine learning in healthcare” returns overly general results, refine the query to “machine learning for early-stage cancer detection using MRI datasets 2020-2023.” This adds specificity by defining the application (cancer detection), data type (MRI), and time frame. Similarly, use technical filters like programming languages (e.g., “Python-based”), frameworks (e.g., “TensorFlow”), or domain-specific terms (e.g., “HIPAA compliance”) to exclude irrelevant results. If the report lacks depth, prioritize peer-reviewed sources or preprints via platform filters.

If the report is too narrow, broaden the query by removing overly restrictive terms or expanding keywords. For instance, a search for “Python Flask REST API authentication JWT 2023” might exclude relevant methods like OAuth or SAML. Simplify it to “web API authentication methods 2023” and include synonyms (e.g., “token-based,” “session-based”) to capture a wider range of techniques. Use wildcards (e.g., “cloud*” to include “AWS,” “Azure”) or hierarchical terms (e.g., “neural networks” instead of “transformers”). If the topic is time-sensitive, remove date filters to include foundational research. For example, “optimizing SQL queries” could benefit from older but widely cited papers on indexing strategies.

Finally, iterate incrementally. Start with a base query, review the first page of results, and adjust keywords or filters based on gaps. For example, if a search for “blockchain scalability” returns theoretical papers but lacks implementation details, add “layer-2 solutions” or “sharding in Ethereum.” Use DeepResearch’s advanced features like keyword highlighting or related-search suggestions to identify patterns. Document successful refinements (e.g., “adding ‘real-time’ improved IoT sensor results by 40%”) to build a reusable strategy. This approach balances precision and flexibility, ensuring the tool adapts to the problem rather than forcing the problem to fit the tool.