Deep Research is an AI-assisted workflow, and in some products a named feature, in which a language model autonomously plans a multi-step information-gathering process, executes web searches or database queries, synthesises findings across many sources, and produces a comprehensive written report — often with citations. It is distinguished from single-turn question answering by the iterative search-read-plan loop that continues until the model judges the topic sufficiently covered.

Think of it like commissioning a junior analyst who goes away for an hour, reads dozens of sources, resolves conflicts between them, and comes back with a structured briefing — rather than someone who answers off the top of their head.

OpenAI's Deep Research feature, released in early 2025 as part of ChatGPT Pro, and Gemini's equivalent perform hundreds of searches per session, reading full web pages and cross-referencing claims. A user asking for a competitive landscape analysis receives a document with dozens of cited sources, organised sections, and nuanced hedging of uncertain claims. Anthropic offers similar capabilities through multi-agent Claude configurations.

Deep Research is an application of RAG at scale: each search result is retrieved, chunked, and placed in the Long Context window for synthesis. Chain of Thought reasoning guides the model's decisions about what to search next. The workflow is susceptible to Hallucination when sources conflict or when the model fills gaps without explicit retrieval; citing sources mitigates but does not eliminate this risk. Foundation Models with large Context Windows can hold more retrieved text simultaneously, improving synthesis quality. Prompt Engineering determines how the model structures its search plan and how it resolves contradictions across sources.

You will encounter Deep Research as a product feature in AI assistants and as a pattern to implement when building autonomous research agents or competitive intelligence tools.