Searching with AI

How do AI search tools work?

In AI-supported information retrieval services, you ask your question in natural language rather than constructing a search string using keywords and Boolean operators (AND, OR, NOT). The service translates your query into a keyword search and combines it with a semantic search, where the meaning of your question is matched with the meaning of articles rather than individual words. This allows the service to identify relevant articles even when they use different terminology from the words you entered.

The results are presented both as a list of publications and as an AI-generated summary with clear references to the retrieved articles. The summary is generated by a language model that answers your question based on the articles found.

Want to learn more?

How AI-powered scholarly search services work

AI-supported information retrieval versus traditional search services and AI chatbots

When can you use AI search tools?

AI search tools work well as a complement to traditional databases when you:

• Want a quick overview or initial introduction to a topic.

• Are looking for search terms and relevant concepts to use in further searching.

• Want to supplement searches in other databases; semantic search can identify articles on the same topic even when different terminology is used.

Traditional search services may be preferable when you:

• Conduct a systematic search and need assurance that the search is comprehensive and that relevant articles have not been excluded through a hidden AI selection process.

• Know exactly which terms or subject headings should be used.

• Need to search a specific database or its full-text content.

• Are looking for a specific publication.

What can you find with AI-powered search services?

AI-powered search services are usually limited to broad, article-focused collections of scholarly literature. Many of them build their indexes on open collections such as Semantic Scholar or OpenAlex, which creates a degree of overlap between services. AI tools provided by major publishers instead search their own proprietary databases.

Most services search only metadata, titles, abstracts, and sometimes subject headings, rather than full-text articles. As a result, AI-generated summaries are typically based on abstracts rather than complete articles. Coverage of books, humanities research, and national-language material is often limited. For this type of content, as well as specialist databases, statistical resources, and digitized collections, you should use the library’s traditional search services, where the scope and coverage of the data are usually clearly described.

Searching effectively with AI

How you formulate your question affects the results you receive. Here are some tips for improving your searches:

• Write in natural language, much as you would when speaking to a colleague, supervisor, or fellow student. For example, “What factors influence students’ motivation in online learning?” is better than “student motivation online learning.”

• Provide context. The service needs sufficient background information to interpret your question correctly. Be specific and include details that help the tool understand what you are looking for.

• Adapt your query to the tool. For some search tools, particularly those that answer questions directly, small changes in wording can make a significant difference. Try different formulations and refine your query based on the results. For more advanced, agent-based searches, each search session may take several minutes, so it is worth investing time in formulating a well-considered question from the start. These tools also often ask follow-up questions before beginning the search.

• Reuse new terminology you discover. Keywords and concepts that appear in the search results and summaries can help you construct more effective search strategies in traditional databases.

• Try several different services. Search tools use different data sources and interpret questions differently. Some services (such as Undermind, Elicit, and Asta) offer more advanced, agent-based analyses (“deep research”), where the search process unfolds in multiple stages.

Can you trust AI services?

As with any use of AI, you remain responsible for evaluating the results. AI-generated content is not a finished product that replaces your own reading, thinking, and writing, but it can help you discover relevant material and gain an overview of a topic. Critically assess the results and keep the following points in mind:

• Review the summaries carefully. AI systems may misinterpret or oversimplify what articles actually say. Well-written answers with clear citations can appear trustworthy, but you should always read the underlying article before citing it.

• Results may vary. AI search tools are not entirely consistent, and the same query may produce different results at different times. In contexts where reproducibility is essential, such as systematic reviews, this is a significant limitation.

• The selection process is not transparent. These services typically show only a subset of the available results, and users cannot verify what has been excluded. It is rarely possible to understand in detail why a particular document has been ranked as relevant.

• The services evolve rapidly. New features are introduced, underlying language models are replaced, and indexes are expanded. Do not expect a service to function the same way over time.

Examples of AI Search Services

There are many AI services focused on scholarly literature searching, and both the market and the services themselves are evolving rapidly. Below is a selection of tools. Our traditional databases can be found in our Databases A–Z list.

Services Available Through the Library

Keenious

Keenious is an AI-powered search service that combines keyword search and semantic search to identify relevant articles. The service focuses primarily on identifying themes and suggesting additional searches rather than generating lengthy summaries. It can be used directly within Microsoft Word and Google Docs.

Other services

All the following services offer some form of free access, although the scope varies. Some require users to create an account.

Asta

Asta is an AI tool developed by the Allen Institute for AI that retrieves articles, generates research reports, and analyses data. The service is built on Semantic Scholar and clearly shows how it analyses the results.

Connected Papers

Connected Papers visualises relationships between scholarly articles through citation and reference analysis. The tool is primarily used to gain an overview of a research field and discover influential or related studies.

Consensus

Consensus is an AI-powered search service that searches, ranks, and summarises scientific research based on natural-language questions. It can also generate literature reviews and show the extent to which the research is in agreement on a particular issue.

Elicit

Elicit helps users find, summarise, and organise scholarly articles. Results are presented in tables that allow users to compare aspects such as methodology, sample characteristics, and findings across studies. It primarily draws on Semantic Scholar, OpenAlex, and Crossref.

ResearchRabbit

ResearchRabbit focuses on discovering related research through citation and reference networks. Results are visualised in diagrams that illustrate connections between articles and research areas. It can be integrated with Zotero and Mendeley.

SciSpace

SciSpace is a comprehensive AI-supported research tool offering features for literature search, summery generation, PDF analysis, and writing assistance. The tool can generate literature reviews and combines semantic search with external databases.

Undermind

Undermind is an AI tool for in-depth literature search in which the search process develops iteratively based on previous results. The service can identify research gaps, generate reports, and assess how comprehensively the search covers a research field.