From a topic to a question you can answer

AI for research and reports · Lesson 2 / 24

A topic is not a question

"The influence of social media on teenagers" is a territory, not a question. Territories have no answers, so work built on one turns into a retelling of everything. Research starts where you have a question that could be answered no. If no imaginable result could disappoint you, you are writing a briefing note, not a study.

Four narrowings

Who exactly. Not "teenagers" but "14-16 year olds in large cities".
What exactly. Not "social media" but "daily minutes spent on short-form video".
Related to what. Not "influence" but "association with self-esteem on a scale you name".
Where and when. Country and period, or you will blend data that cannot be compared.

Testing the question

1. Can the answer be no?              mandatory
2. Object, property, relation named?  mandatory
3. Are the data or texts reachable?   otherwise it is elegant and dead
4. Does it fit the length?            one question per 20 pages, not five
5. Does anyone care?                  who needs the answer and why

A prompt that works

Here is my topic: [topic] and format: [term paper, 25 pages].
Give me 7 candidate research questions. For each state:
- what a negative answer would look like,
- what data would be needed to answer it,
- why the question might turn out too broad.
Do not rank them — I want the spread, not your favourite.

The ban on ranking is deliberate. Ask for "the best question" and the model hands you the blandest one. A spread of options is worth more than a single "right" answer.

Insight. A good question almost always sounds more modest than the topic. The feeling that it is too narrow is usually the sign that you have finally reached something researchable.

Common mistake. Phrasing the question with the answer already baked in: "why does social media harm teenagers" assumes harm is established. That is a thesis wearing a question mark.

Pro tip. Write two sentences: "if it turns out that ..., I will conclude A" and "if it turns out that ..., I will conclude B". If you cannot write both, you do not yet have a question.

Cheat sheet

A topic is a territory; a question has an answer.
A negative answer must be possible.
Narrow by object, property, relation, place and period.
Ask AI for a spread of questions, never the best one.

1. What marks a genuine research question?

2. What is wrong with "why does social media harm teenagers"?

3. Why ask the model for a spread instead of the best question?

Task — checked by AI

Take the topic of your next piece of work. Formulate 5 candidate research questions, narrowing the topic by object, property, relation, place and period. For each one, write what a negative answer would sound like and what data or texts would be needed to answer it at all. Then choose one question and explain why you discarded the other four.

← Back

🔒 Answer the question correctly to move on to the next lesson.