AI tools for research and a new way to use language models
Fecha del post: 13-05-2025
AI systems designed to assist us from the first dives to the final bibliography.
One of the missions of contemporary artificial intelligence is to help us find, sort and digest information, especially with the help of large language models. These systems have come at a time when we most need to manage knowledge that we produce and share en masse, but then struggle to embrace and consume. Its value lies in finding the ideas and data we need quickly, so that we can devote our effort and time to thinking or, in other words, start climbing the ladder a rung or two ahead.
AI-based systems academic research as well as trend studies in the business world. AI analytics tools can analyse thousands of papers to show us which authors collaborate with each other or how topics are grouped, creating an interactive, filterable map of the literature on demand. The Generative AI,the long-awaited one, can start from a research question and return useful sub-content as a synthesis or a contrast of approaches. The first shows us the terrain on the map, while the second suggests where we can move forward.
Practical tools
Starting with the more analytical ones and leaving the mixed or generative ones for last, we go through four practical research tools that integrate AI as a functionality, and one extra ball.
It is a tool based above all on the connection between authors, topics and articles, which shows us networks of citations and allows us to create the complete graph of the literature around a topic. As a starting point, Inciteful asks for the title or URL of a paper, but you can also simply search by your research topic. There is also the possibility to enter the data of two items, to show how they are connected to each other.
Figure 1. Screenshot in Inciteful: initial search screen and connection between papers.
Figure 2. Screenshot on Inciteful: network of nodes with articles and authors.
In Scite, AI integration is more obvious and practical: given a question, it creates a single summary answer by combining information from all references. The tool analyses the semantics of the papers to extract the nature of each quote: how many quotes support it ( check symbol), question it (question mark) or just mention it (slash). This allows us to do something as valuable as adding context to the impact metrics of an article in our bibliography.
Figure 3. Screenshot in Scite: initial search screen.
Figure 4. Screenshot in Scite: citation assessment of an article.
In addition to integrating the functionalities of the previous ones, it is a very complete digital product that not only allows you to navigate from paper to paper in the form of a visual network, but also makes it possible to set up alerts on a topic or an author you follow and create lists of papers. In addition, the system itself suggests which other papers you might be interested in, all in the style of a recommendation system like Spotify or Netflix. It also allows you to make public lists, as in Google Maps, and to work collaboratively with other users.
Figure 5. Screenshot on Research Rabbit: customised list of items.
It has the backing of the British government, Stanford University and NASA, and is based entirely on generative AI. Its flagship functionality is the ability to ask direct questions to a paper or a collection of articles, and finally get a targeted report with all the references. But actually, the most striking feature is the ability to improve the user's initial question: the tool instantly evaluates the quality of the question and makes suggestions to make it more accurate or interesting.
Figure 6. Screenshot in Elicit: suggestions for improvement for the initial question..
Extra ball: Consensus.
What started as a humble customised GPT within the Plus version of ChatGPT has turned into a full-fledged digital research product. Based on a question, attempt to synthesise the scientific consensus on that topic, indicating whether there is agreement or disagreement between studies. In a simple and visual way it shows how many support a statement, how many doubt it and which conclusions predominate, as well as providing a short report for quick guidance.
Figure 7. Screenshot on Consensus: impact metrics from a question
The depth button
In recent months, a new functionality has appeared on the platforms of large commercial language models focused on in-depth research. Specifically, it is a button with the same name, "in-depth research" or "deep research", which can already be found in ChatGPT, Plus version (with limited requests) or Pro, and in Gemini Advanced, although they promise that it will gradually be opened to free use and allow some tests without cost.
Figure 8. Screenshot in ChatGPT Plus: In-depth research button.
Figure 9. Screenshot in Gemini Advanced: Deep Research button.
This option, which must be activated before launching the prompt, works as a shortcut: the model generates a synthetic and organised report on the topic, gathering key information, data and context. Before starting the investigation, the system may ask additional questions to better focus the search.
Figure 10. Screenshot in ChatGPT Plus: questions to narrow down the research
It should be noted that, once these questions have been answered, the system initiates a process that may take much longer than a normal response. In particular, in ChatGPT Plus it can take up to 10 minutes. A progress bar indicates progress.
Figure 11. Screenshot in ChatGPT Plus: Research start and progress bar
What we get now is a comprehensive, considerably accurate report, including examples and links that can quickly put us on the track of what we are looking for.
Figure 12: Screenshot of ChatGPT Plus: research result (excerpt).
Closure
The tools designed to apply AI for research are neither infallible nor definitive, and may still be subject to errors and hallucinations, but research with AI is already a radically different process from research without it. Assisted search is, like almost everything else when it comes to AI, about not dismissing as imperfect what can be useful, spending some time trying out new uses that can save us many hours later on, and focusing on what it can do to keep our focus on the next steps.
Content prepared by Carmen Torrijos, expert in AI applied to language and communication. The contents and points of view reflected in this publication are the sole responsibility of the author.
