deep-research

To generate a multi-source research report with web citations, you need an iterative research process that decomposes topics into angles, runs parallel searches across multiple engines, and synthesizes findings with confidence scoring and gap analysis. This approach triangulates evidence and explicitly identifies limitations for decision-ready outputs.

Recency filtering in research planning requires an explicit interview phase to capture the user's time window cutoff before execution begins. The research plan is then designed around that specific date boundary, ensuring all parallel searches retrieve only sources published within the required timeframe.

Balancing polarized topics during multi-angle investigation involves running a second research round specifically targeted at finding first-party success cases. This angle balancing reduces the skew typically caused by general or negative media coverage, providing a more neutral synthesis.

Multi-source synthesis manages context limits by employing a final synthesis agent that reads research outputs directly from disk. This prevents re-passing full content through the context window, allowing the agent to compile cited findings, confidence scores, and structured debates without hitting token limits.

Yes, parallel searches can run across Perplexity, Tavily, Exa, Gemini, and native web tools simultaneously. The orchestration process decomposes your topic into independent angles and queries these multiple sources in parallel to gather comprehensive evidence before synthesizing the results.

Gap analysis and confidence scoring are needed when research reports must support high-stakes decisions or presentations. They provide explicit limitations of the evidence and rate the reliability of findings, ensuring readers understand exactly where coverage is insufficient and how much trust to place in the conclusions.