ChatGPT Deep Research
Deep Research is ChatGPT's autonomous research mode. Rather than answering a question from training data or a single web search, it behaves like an analyst — browsing iteratively, reading dozens to hundreds of sources, and synthesising findings into a structured, referenced report. Tasks that would take a human researcher hours can be completed in 5 to 30 minutes.
What Deep Research Is
When you invoke Deep Research, ChatGPT launches an autonomous research loop rather than generating an immediate response. The system searches the web, follows links, reads full documents including PDFs and images, evaluates source quality, and iteratively refines its search strategy based on what it finds. At the end of this process it produces a structured, analyst-grade report — typically with citations, a summary, and organised sections covering the topic in depth.
You can attach your own files or spreadsheets to provide additional context that the research agent incorporates alongside web sources. This is particularly useful when you want the report to be framed around your specific data, competitive context, or domain constraints rather than purely public information.
How It Works
Deep Research is powered by a specialised model in the GPT-5 series, optimised for multi-step information gathering rather than single-turn conversation. The process:
- Query planning: The model analyses your research question and identifies the key sub-questions it needs to answer.
- Iterative browsing: It performs multiple web searches, reads pages and PDFs, and adjusts its search strategy based on what it finds — following threads that prove productive and abandoning dead ends.
- Image and document reading: Unlike simple web search, it can extract information from images, charts, and PDF documents encountered during research.
- Synthesis: After gathering sufficient evidence, it writes a structured report with sections, a summary, and source citations.
The full process typically takes 5 to 30 minutes depending on the complexity of the query. You can watch it work in real time in the ChatGPT interface.
February 2026 Additions
OpenAI expanded Deep Research capabilities in February 2026 with two significant additions:
- MCP server connections: Deep Research can now connect to MCP (Model Context Protocol) servers and integrated apps, allowing it to pull from proprietary or authenticated data sources rather than only public web content.
- Source restriction: You can instruct Deep Research to restrict its searches to trusted or authenticated sources — useful for compliance-sensitive research where you cannot rely on unverified public pages.
Plan Limits
| Plan | Full Research Queries/30 days | Lightweight Queries/30 days |
|---|---|---|
| Free | — | 5 lightweight (o4-mini based) |
| Plus / Team / Enterprise | 10 | 15 |
| Pro | 125 | 125 |
Full research queries use the most capable model and perform the deepest analysis. Lightweight queries (powered by o4-mini) are faster and consume less quota — appropriate for narrower questions or when you need quick structured summaries rather than exhaustive research.
Best Use Cases
Strong use cases
- Market research and competitive analysis
- Academic literature reviews
- Investment memos and due diligence
- Technical landscape surveys (e.g., "compare vector databases in 2025")
- Regulatory and compliance landscape overviews
Limitations
- Not real-time — cannot track live prices or breaking news
- Takes time — not suitable for quick factual lookups
- Can misread complex charts without structured underlying data
- May not have access to paywalled or login-required sources
Checklist
- How does Deep Research differ from a standard ChatGPT web search?
- What model powers lightweight Deep Research queries, and why is it different from full queries?
- How many full research queries does the Plus plan allow per 30 days?
- What were the two major additions to Deep Research in February 2026?
- Name two scenarios where Deep Research would not be the right tool to use.