Ask a genuine research question

"Are electric cars profitable?" The question is a natural one, and it guarantees a soft answer: profitable for whom, where, over what period, compared with what? The assistant fills in those gaps for you, with assumptions you cannot see.

The five elements of a research question

• The decision behind it: "I need to replace my car within six months."
• The scope: country, period, situation. "In France, 15,000 km a year, home charging possible, buying second-hand."
• The sources you want: "Public sources or independent studies preferably, not manufacturer websites."
• What you already know, so you do not get an answer you already have.
• The format: "A comparison table over five years, with each figure accompanied by its source and date."

The trap of the question that contains its answer

"Find me studies showing that remote working increases productivity." The tool will do exactly what you ask: it will find those studies, and ignore the others. You will get confirmation, not information. Put it neutrally: "What do studies say about the effect of remote working on productivity? Where do they agree, where do they diverge?"

Asking for disagreements and gaps

Two sentences to add almost every time: "Flag the points where the sources contradict each other" and "Tell me what is not yet known". A summary with no disagreements at all on a debated subject is not reassuring: it signals that some of the sources have been set aside or smoothed over.

Key points

• Decision, scope, desired sources, what you already know, format: the five elements.
• A leading question produces confirmation, not information.
• Always ask for contradictions between sources and for what remains unknown.