The way you phrase a question shapes the evidence you find. Ask a leading question (one that assumes a direction of effect) and you will tend to surface confirmatory evidence. Ask a neutral question (one that opens the topic up rather than closing it down) and you will get a fuller, more balanced picture.

This is important because AI tools do not correct for the direction of your question. When you type a question into an AI research tool, you are not just searching — you are directing it. The words you choose, and the assumptions baked into them, shape what comes back.

The examples below show the same topic asked two ways. Notice how the leading version assumes an outcome before any evidence has been reviewed.

Look at the word that gives the game away. “Cause”, “reduce”, “worse”, “harm” — each one assumes an answer before any evidence has been considered. The neutral version replaces that assumption with an open question: what does the evidence actually say?

Leading questions are the most common problem, but they are not the only one. Here are five pitfalls that can undermine your search before it begins.

If you are unsure whether your question is neutral, try this reframe:

Instead of: “Does X help Y?”

Try: “What is the evidence on the relationship between X and Y?”

You can make it more specific by adding context: who the population is, what the setting is, or what outcome you care about. That specificity is what takes a decent question and makes it a good one. 

Putting it into practice: a worked example

Here is how you can move from a first instinct to a well-formed question.

First instinct: “Does social media cause anxiety in university students?”

What’s wrong: The word “cause” assumes a causal relationship that the evidence may not support. “Social media” is also vague — it could mean passive scrolling, active posting, or comparison behaviour, all of which the literature treats differently.

Apply the formula: “What is the evidence on the relationship between social media use and anxiety symptoms in university students?”

Add specificity: “What does the evidence say about the relationship between passive social media use and anxiety symptoms in undergraduate students aged 18–25?”

The final version is open, specific, and free of assumptions. It will surface a wider and more useful range of evidence than the original. The research studies that fit your exact research question may not exist; however, Consensus will give you the closest available research because it now knows what you are looking for.

Putting it into practice: a worked example

Here is how you can move from a first instinct to a well-formed question.

First instinct: “Does social media cause anxiety in university students?”

What’s wrong: The word “cause” assumes a causal relationship that the evidence may not support. “Social media” is also vague — it could mean passive scrolling, active posting, or comparison behaviour, all of which the literature treats differently.

Apply the formula: “What is the evidence on the relationship between social media use and anxiety symptoms in university students?”

Add specificity: “What does the evidence say about the relationship between passive social media use and anxiety symptoms in undergraduate students aged 18–25?”

The final version is open, specific, and free of assumptions. It will surface a wider and more useful range of evidence than the original. The research studies that fit your exact research question may not exist; however, Consensus will give you the closest available research because it now knows what you are looking for.

Before you search, ask yourself:

Once you can ask a neutral, well-framed question, you are ready to go further. The companion guide From Framework to Prompt introduces structured frameworks — PICO, SPIDER, PEO, and others — that help you build on the skills practised here.

The ability to ask a good question is not a small thing. It is where all good research begins.

AI does not think for you — it starts where you point it. The quality of your question is the quality of your research.