Deep research: Understanding the restrictions of the most recent powerful AI tool for scientific writers

In February 2024, I wrote AJN off the charts blog post titled “Using AI and Technology in Complex Research: Tips for Researchers and Students.” Since then, the sector of artificial intelligence has evolved rapidly. It is apparent to all of us watching this field develop that the businesses hosting and developing large language models (LLM) will ultimately give attention to scientific research. In the previous post, I explained that there is no such thing as a single software that enables conducting research or literature reviews using artificial intelligence. However, implementing recent features in AI software platforms, reminiscent of deep research capabilities, can confuse us and make us think otherwise. The purpose of this blog post is to introduce the concept of ​​deep research tools while also providing guidance to users who wish to find out about these emerging tools.

What is deep research?

Photo: Marcus Winkler/Unsplash

Deep research is a term utilized in LLM software platforms that enables users to enter a prompt to initiate an in-depth process that involves searching, analyzing, and synthesizing “hundreds of online sources to create a comprehensive report at the level of research analysis” (OpenAI, 2025). Using this tool also takes time under consideration as responses should not immediate and the time to result may vary depending on the subject and complexity of the research query. While such AI results could seem revolutionary, there are some limitations that have to be shared, especially with regards to publishing standards. While research reports produced by deep research tools are typically excellent, I actually have yet to come across a report that meets the editorial standards required for publication in a peer-reviewed journal.

One of the major limitations is that LLMs remain unreliable without human supervision to examine accuracy and make sure that references meet the standard standards of peer-reviewed academic publications. The second limitation is that (currently) deep research tools only search one scientific database – Semantic Scholar or in other cases PubMed. While each are excellent academic databases, searching just one or two databases severely limits your search since it excludes articles that may very well be indexed in other academic databases. In addition to other limitations that will probably be discussed below, a research report created entirely using deep research tools counting on a single database won’t be a publication-ready document. It is significant to notice that as these tools develop, we are going to likely see in-depth research using various academic databases that may potentially address this challenge. So how should we use this recent tool in academic research spaces?

Incorporating deep research tools into the research workflow

Is there a spot for deep research tools? The answer is yes, but with reservations. In my previous post in February, I showed learn how to make the literature review process easier by combining different tools to avoid hallucinations and using AI to refine and make clear – more as a support system or research assistant, and fewer as the final word “do it all” tool. I’d say the identical about leveraging the deep research capabilities which might be increasingly being integrated into AI software platforms. This tool does job of providing researchers and students with a general overview of their field of research.

As an example of an application where deep research tools could also be useful, consider a subject reminiscent of genetics and genomics in nursing education. A student or novice researcher could also be fascinated by this area. The first step is to search out out what has already been done on this topic. A literature review is often useful here as a research tool. However, a literature review often requires students and/or researchers to have a powerful interest in a given topic. I see the LLM tools for deep research as a strategy to start the method – a pre-literature review, when you will. For example, a user might use the next prompt:

“I am interested in the field of genetics and genomics in nursing education. I would like to gain an overview of what has already been published on this topic, while exposing gaps so that I can develop my research in this area.”

Obtaining a full report on this topic based on the prompt can function a wonderful precursor to undertaking a more formal literature review. The student and/or researcher can then apply the techniques described in my February 2024 blog post to leverage AI as a support system in conducting a rigorous literature review. Deep research tools can save researchers useful time to find topics that not only interest them, but around which they’ll construct a crucial area of ​​research.

Use deep research tools with caution

As with all AI/LLM products available on the market, researchers should exercise caution when interpreting results obtained from in-depth research reports. Before embarking on more formal research processes, it is incredibly essential to confirm the accuracy of all sources within the report. The report can’t be used as a stand-alone literature review and will not be suitable for publication. Recognizing the worth and pitfalls of those tools requires a high level of clinical reasoning and human supervision.

Used accurately, these developing tools can save time and help users refine potential topic ideas while eliminating others. Students and researchers alike undertake extensive literature reviews and discover that what interests them has already been written. The time it takes to vary the situation and begin over could be exhausting and result in incomplete work. Using deep research tools with appropriate care can alleviate these common academic challenges for some research purposes. Other limitations remain even after these glaring limitations are resolved. For example, no amount of database searching can replace a human researcher.

Bottom line: Proceed with caution and watch the trends on this area.

Justin Fontenot DNP, RN, NEA-BC, FAADN, Associate Professor, Tulane University School of Medicine, Nursing Program.