Searching and Writing

Specialised AI-based tools can help you search for scholarly literature. Typically you type a natural language query and receive a concise summary with references to journal articles. Examples of AI-based tools are Elicit, Consensus and SciSpace. Each tool draws on its own underlying database – with some tools, the summary is based only on abstracts or, where available, on full-text articles.

Natural language search means that you can type a question or sentence instead of extracting single keywords. The search is then semantic, so you can get hits on texts that have similar content even if they do not contain the exact same words as your question. One advantage is that compared to "traditional" searches, you are more likely to get hits where the keywords are used in the right context and have the right relationship to each other.

For example, a traditional search using keywords related to education and the EU may result in hits related to education in EU countries, education about how the EU works, or EU policies on education. A semantic search allows you greater control over the context.

If you are looking for publications containing a specific word, it may be more efficient to use a non-semantic search tool, such as the Library Search Tool or a Boolean search in an article database. The same applies if you are searching for a specific work or publications by a specific author.

Large Language Models (LLM) are used to generate well-formulated summaries, often with references to sources. The selection of the most important parts of an article can sometimes be wrong, so that the summary is misleading. You may need to have prior knowledge of the subject to be able to judge whether the answer is a good representation of the literature.

Please note:

• You are allowed to read summaries and translations of articles.
• You may not copy and reuse summaries as if they were your own text.
• If you plan to reference the work, first read it in its original form.

Unlike traditional article databases, most AI search services don't show the total number of hits. You may see "top 5", "top 10", or similar. Articles match the search to a greater or lesser degree (rather than either matching or not, as in keyword search).

This makes it harder to gauge how much literature exists on a topic, identify the newest work, or see at a glance what kinds of publication are available and whether they have been peer‑reviewed.

AI-supported search tools may also have filters, but they usually cannot be prompted like a chatbot. So typing "Show only peer-reviewed articles from the last 5 years" tends to work less well than selecting the appropriate filter options manually.

It is not possible to search for all existing research worldwide. No service has it all.

Several different AI search tools are connected to the Semantic Scholar database, such as Elicit, Consensus and SciSpace. They specialise in finding academic publications, whereas ChatGPT, Microsoft Copilot and Perplexity search the web.

AI search tools generally have better coverage of topics in engineering, medicine and the natural sciences than in law or the humanities. The results are often dominated by journal articles, with far fewer books, theses, and other publication types.

Check the guidelines for your course first. Rules differ, and where none are stated, you may use AI tools.

Please note:

• It is not appropriate to search with AI search tools if you are going to perform a reproducible search.
• If you are going to systematically go through all the research on a topic, AI search tools are not appropriate.
• If your course sets no restrictions on search strategies, feel free to mix AI tools with traditional databases and other platforms to suit your needs.