Deep research
 Compiled by
Updated
See also
Introduction - What is AI "deep research"?Deep research (sometimes called "deepsearch," "iterative deep research," or "research mode") refers to an advanced capability in modern large language models (LLMs) where the AI tool or system does not give an instant answer from its trained knowledge alone. Instead, it breaks a user’s question into sub-questions or research tasks. Other features characteristic of deep research tools include:
Examples of this capability today (December 2025):
(See OpenAI Deep Research, Google’s Gemini Deep Research and Perplexity’s Deep Research (renamed Perplexity Labs). Some features of deep research
AI-powered deep research systems, such as those developed by OpenAI, Google’s Gemini, or Perplexity, automate these processes, significantly reducing research time (e.g., from days to minutes) while maintaining accuracy. These systems can take 5–30 minutes per query (not seconds) because they are literally performing human-level literature review and investigative work. Examples of deep, reasoning models
OpenAI's deep research modelIn 2025, Nature published a story entitled, OpenAI’s "deep research" tool: is it useful for scientists? reporting that OpenAI's deep research model produces pages-long reports that are well-cited and might help perform literature reviews. As it turned out, the pay-for-access platform called ‘OAI deep research’, synthesizes information from hundreds of websites into cited reports, and follows a similar path as Google, called ‘Deep Research’, released in December 2024. Both act as personal assistants doing the equivalent of hours of work in tens of minutes. While scientists were impressed with OpenAI's deep research tool, others were less enthusiastic. “If a human did this I would be like: ‘This needs a lot of work’,” Kyle Kabasares, a data scientist at the Bay Area Environmental Research Institute, said.... See librarians' points of viewSee Aaron Tay's August 8th blogpost on Academic Deep Research.
More on deep researchDeep research in medicine increasingly relies on artificial intelligence (AI) and semantic searching technologies to navigate the vast and rapidly growing body of biomedical literature. Traditional keyword-based searching often misses relevant studies because it cannot account for synonyms, abbreviations, or the nuanced ways clinical concepts are expressed. Semantic search, powered by AI and natural language processing (NLP), addresses this gap by interpreting the intent and contextual meaning of search queries, linking related terms such as heart attack, myocardial infarction, and acute coronary syndrome. This capability allows researchers and clinicians to uncover connections across diverse sources, identify hidden patterns, and integrate findings more efficiently. As a result, semantic searching enhances deep research by enabling more comprehensive evidence discovery, supporting systematic reviews, precision medicine, and informed clinical decision-making. How do I know it's deep research?Deep research is prominently located in some interfaces, and a default option for emerging models. Academic libraries have been a target for new AI technologies since the worldwide emergence of GenAI. “Deep Research”, “Research Plan”, and “Research Report”, are deliberate terms, and language appealing to academics. The word “deep” (with “research”) pops up everywhere in AI: DeepSeek, DeepMind (Google), ChatGPT Deep Research (OpenAI) – DeepPiXEL, DeepNeural AI, DeepX, etc. Any (all?) deep search using AI returns a mix of credible and not credible sources from the web including articles, websites, and blogs. It's not the random sources that must be evaluated that is the issue, but that you cede control to the bots, limiting your autonomy. For all their flaws, search engines allow you to explore, which involves exploring the scope and constraints of your research question, discovering new language, and following unique pathways and ideas. Deep research creates a research plan, but it's not a feasible plan for academics. It’s a list of how a bot will search; in contrast, creating a research plan is key to learning, an exploration of your question and not a search for answers. Established pathways for research allow for the possibility that you won't find anything, or that your research question will change and evolve. Impact on biomedicineAccording to Wang et al (2025) scientific research can only be compared and reproduced by strictly following fixed methods, which are relatively easy for AI to imitate and learn. Basically, current AI can speed up the retrieval and summarization of literature on one hand, and facilitate the comparison and discussion of results on the other. For dry lab research, especially text-based ones, such as systematic review, meta-analysis, bibliometric study, AI may replace them efficiently. With the dawn of a new era, the future development and face of academia is full of curiosity. Undoubtedly, the crucial part of research will continue to be creative originality and critical thinking. Current AI products that combine deep thinking and deep search with real-time web and database access, focus on peer-reviewed research, offering custom AI workflows. Emerging platforms in the biomedical space are Elicit.com and Undermind.ai. Examples of GenAI "Deep Research" features
References
Disclaimer
|
