What Does It Take to Research with AI? A Rapid Review of Competencies to Train LLM-Literate Researchers

Imagine the world of science as a massive, bustling library where researchers are the librarians. For decades, these librarians have had a strict rulebook: they must find books, read them carefully, check the facts, and write their own reports based on what they've learned. But recently, a new, incredibly fast, and chatty robot assistant has arrived in the library. This robot, powered by something called a "Large Language Model" (or LLM), can read millions of books in a second, summarize them, write drafts, and even help solve complex puzzles. It's like having a super-smart intern who never sleeps.

However, there's a catch. This robot is a bit of a trickster. Sometimes it tells the truth, but other times it confidently makes up facts (a glitch researchers call "hallucinating") or copies ideas without giving credit. Because of this, the big question isn't just "Can the robot help us?" but "What skills do the human librarians need to keep the robot in check?" If the librarians don't know enough about the books they are organizing, they might let the robot write the whole report, leading to a library full of lies. This paper dives into exactly what skills researchers need to use these AI robots responsibly without losing their own minds or the truth.

The Great AI Detective Hunt

A team of researchers decided to play detective to answer a burning question: What does it actually take to be a researcher who uses AI tools effectively, critically, and responsibly? They didn't just guess; they went on a rapid review mission, scanning 194 articles published between 2022 and 2025. From this mountain of paperwork, they carefully selected 40 articles that showed real-life examples of people using AI in their research. They then used a mix of human brains and a second AI tool to sift through these articles, looking for the specific "superpowers" or competencies that successful researchers seemed to have.

What they found was a list of eight essential skills. Think of these not as computer coding skills, but as the mental muscles a researcher needs to flex.

1. The Expert Eye (Domain Expertise and Oversight)
This was the most common skill found, appearing in 123 instances across the studies. Imagine the AI robot as a very fast, very confident tour guide who knows a little bit about everything but isn't an expert in anything. If you don't know the territory yourself, you might believe the guide when they point to a fake waterfall and say, "That's the real thing!" The paper suggests that researchers must be the true experts in their field. They need to know the subject matter well enough to spot when the AI is lying, making up sources, or giving a shallow answer. The robot can do the heavy lifting, but the human must be the one to say, "Wait, that doesn't sound right," and double-check the facts.

2. The Smart Switch (Metacognition and Decision Making)
Found in 55 instances, this skill is about knowing when to use the robot and when to turn it off. It's like deciding whether to use a power drill to hang a picture or just use a hammer. The researchers found that the best scientists don't let the AI take over the whole job. They decide which parts of the research are safe for the AI (like organizing notes) and which parts must stay human (like drawing the final conclusion). It's about being the captain of the ship, not just a passenger.

3. The Moral Compass (Ethics, Privacy, and Integrity)
With 53 mentions, this skill is about playing fair. Using AI isn't just about getting answers; it's about doing it without stealing, lying, or leaking secrets. Researchers need to know how to protect private data, how to admit when they used a robot to write part of their paper, and how to make sure they aren't accidentally plagiarizing. It's the difference between using a calculator to do math and cheating on a test by having someone else do the work for you.

4. The Art of Asking (Prompt Engineering)
This appeared 38 times. If you ask a robot a vague question like "Tell me about science," you'll get a vague, boring answer. But if you ask, "Explain the theory of relativity to a 12-year-old using a pizza analogy," you get something amazing. This skill is about learning how to talk to the AI. It's like being a director giving instructions to an actor; the better your script (the prompt), the better the performance. Researchers found that refining these questions over and over is key to getting useful results.

5. The Paper Trail (Reproducibility and Reporting)
Found in 29 instances, this is about keeping a diary. If you use an AI tool to help you find information, you need to write down exactly what you asked it, what version of the tool you used, and how you checked its work. Why? So that if someone else wants to repeat your experiment, they can see exactly how you did it. It's like leaving a breadcrumb trail so no one gets lost in the forest of your research.

6. The Blueprint Builder (Methodological and Experimental Design)
This skill, mentioned 20 times, is about planning the whole project before you even turn on the AI. It's like an architect drawing a building plan before laying a single brick. Researchers need to design their studies so that the AI fits in properly without breaking the structure. They have to define their goals and limits clearly so the robot doesn't wander off and build a house on a swamp.

7. The Tech Translator (AI Literacy and Technical Knowledge)
Interestingly, this was mentioned less often (only 16 times) and was flagged as a risk if missing. This doesn't mean researchers need to be software engineers who can build the AI from scratch. Instead, they need to understand how the robot thinks, what its limits are, and what kind of mistakes it tends to make. It's like knowing that a car needs gas and brakes, even if you don't know how to build the engine. Without this basic understanding, you might trust the car to drive itself when it's actually broken.

8. The Data Detective (Data Analysis and Interpretation)
Found in just 7 instances, this is about using the AI to help crunch numbers or find patterns in data. But the paper is clear: the human must still be the one to interpret what those numbers mean. The AI can find a pattern, but the human has to decide if that pattern is important or just a coincidence.

The Big Picture

The authors suggest that using AI in research isn't just about learning a new tool; it's about becoming a better, more critical thinker. The most important takeaway is that the human is still the boss. The AI is a powerful assistant, but it cannot replace the need for human judgment, expertise, and accountability. If a researcher doesn't have the "Expert Eye" or the "Moral Compass," the AI might lead them astray.

The paper concludes that schools and training programs need to teach these eight skills together. It's not enough to just show students how to type a prompt; they need to learn how to question the answers, how to stay ethical, and how to keep the human in the loop. In a world where robots can write essays and analyze data, the most valuable skill a researcher can have is the wisdom to know when to trust the robot and when to trust their own brain.