Search Smarter: AI as a Research Assistant

Literature research is an important part of any scientific endeavour. The rise of artificial intelligence has opened new ways to make the search for relevant literature more efficient. But which AI tools are suitable for scientific research and how do they differ from traditional databases? In this article, we explore how you can integrate AI into the literature search process and what you need to consider when doing so.

What is Artificial Intelligence & Generative AI?

Artificial intelligence, or AI for short, refers to systems or machines that perform tasks that would normally require human intelligence. This can involve a variety of activities, e.g. data analysis, image recognition or text creation.

How does ChatGPT work?

ChatGPT belongs to generative AI, a specialised area of artificial intelligence. It is based on a Large Language Model (LLM) and was trained with a large amount of text data from various internet sources, including Wikipedia, news sites and, in some cases, portals with scientific publications. The underlying text corpus comprises several hundred billion words. These texts were used to train the neural network to learn how language is structured, i.e. which words typically follow each other. The aim of the training was to generate human-like and natural-sounding texts.

How can I use AI for literature research?

ChatGPT & Co are not suitable for literature searches, as they often generate incorrect references. You can use ChatGPT & Co for other tasks in literature research. AI can help with brainstorming, preparing and carrying out literature research - but also with understanding the text. However, it is important to take a critical look at the results generated by the AI and compare them with the current state of research and your own specialist knowledge.

Finding a topic

The first step in any research is to choose a suitable topic. AI tools such as ChatGPT, Google Gemini or Microsoft's Copilot can provide valuable support here.

A possible prompt could be: ‘I would like to write a scientific paper on [...]. Can you help me find a topic?’

Research question

Once you have found a topic, the next step is to formulate a precise research question. AI tools like ChatGPT can help you generate several possible research questions.

 Possible prompt: ‘Develop three research questions for the topic [...].’

Prepare the literature search

Once you have formulated a precise research question, the next step is the systematic literature search. ChatGPT and the like can help by generating search terms, creating search strings or recommending suitable databases for the search. Here it is important to identify synonyms, generic terms, sub-terms and related terms in order to carry out an effective search in databases. By combining AI-generated search terms and traditional search methods such as the use of Boolean operators, you can optimise your search results.

Prompt: ‘Create a literature search for the research question [...]. Please identify key terms related to the topic. Find synonyms and related terms and present them in a table. Truncate the terms in the next step. Create meaningful search strings for the literature search in databases.’

Although these workflows can be well supported by AI, the technical control of the terms always lies with the user. You can now insert the created search string into the search in databases such as Web of Science or Scopus. Pay attention to the help page of the individual databases, as certain search operators may not be supported.

Under this link you can read the chat history with these prompts in ChatGPT.

Conduct a literature search

If you feel like you are stuck in traditional databases, then you should use specialised AI tools for literature research to search for more scientific publications. Specialised AI tools for literature searches, such as Scite, Semantic Scholar or Consensus, combine the search in scientific databases with AI techniques such as natural language processing (NLP) or semantic search. This enables these tools to better understand the context of a search query and process large amounts of data more quickly. These tools access scientific databases and take citable sources into account. They can not only identify relevant articles, but also provide concise summaries and visual representations.

Find similar literature

To deepen your literature search, you can use the so-called snowball effect. There are also suitable AI tools for this step, such as Connected Papers. Connected Papers is a literature mapping tool that helps you to explore scientific papers and their connections to other papers. It generates a visual map of scientific articles based on a ‘seed paper’ and shows their relationships to each other. These relationships are determined by similarities in the citations and content of the papers.

We have only presented a small selection of tools here as examples. You can find more AI tools and their functions in this overview.

Challenges: What you need to look out for in AI-supported research

Unclear database

The database of many AI tools is often unclear and the providers disclose little to no information about it. The models often refer to publicly accessible content, especially open access content, while licenced or fee-based scientific databases are generally not taken into account. In addition, most of the data is available in English.

Danger of misinformation (hallucinations)

Another problem with the use of AI tools is the risk of hallucinations - false or invented information that can appear to be reliable information. AI models often use abstracts and metadata such as titles and authors to generate content. However, these abstracts can only provide superficial information, and important details or contextual aspects are lost.

Therefore, it is crucial to always check citations and references to ensure that the information provided is correct. If an AI tool does not display full texts, access to the relevant sources should be ensured via library access or interlibrary loan.

Costs and subscriptions

Many AI tools only offer a limited basic version of their services free of charge. Full access to advanced functions and reliable data usually requires a paid subscription. Users should weigh up whether the services offered are worth the price and which functions are really necessary for their own research.

Data protection

Another critical point is data protection, especially for tools that are based in the USA and are therefore subject to the data protection regulations there. If registration is required to use a tool, this should be taken into account with regard to the handling of personal data and the associated risks.

Conclusion

AI tools are a valuable addition to traditional literature research. They can speed up the research process and open up new perspectives. Nevertheless, traditional literature research in scientific databases remains indispensable, as these provide access to reliable, citable and quality-assured sources.

A conscious and critical use of both methods enables well-founded and comprehensive research. Always check that the sources you find are correct and citable and remember good scientific practice.

Additional resources

Bucher, Holzweißig, Schwarzer, Holzweißig, Kai, Schwarzer, Markus, & Verlag Franz Vahlen. (2024). Künstliche Intelligenz und wissenschaftliches Arbeiten : ChatGPT & Co.: der Turbo für ein erfolgreiches Studium. Also available as E-Book: https://permalink.obvsg.at/tug/AC17101632

Taskcards AI-Tools for Literature Search

Parts of the text were revised using DeepL Write and checked and edited by the author. The English translation was created with the help of DeepL Translate.