Which prompt to choose
Serious research with AI isn't a single question: it's a sequence. The first prompt opens up the field, the second digs in, the third checks. Skipping the phases is the reason many people get answers that are self-assured but wrong. Choose the starting point based on what you need.
- Understand a new topic from scratch: first prompt, which gives you the map and the right questions to ask yourself.
- Compare options or opinions: second prompt, which sets the viewpoints against each other.
- Verify whether a piece of information is reliable: third prompt, which separates facts, estimates and opinions.
How to do it
- Open the AI assistant. If you need to look up recent facts, turn on the mode that consults the web (usually an icon with a globe or the option "search the web"); without it, the AI answers with what it learned up to its training date and it can be old.
- Copy the right prompt.
- Replace the topic and send.
- On the answer, follow up: "go deeper on point 2", "find a contrary source", "what of all this is uncertain?".
The operational syntax to open research on a new topic:
I want to understand this topic from scratch: "how rent-to-own of a house works".
Don't give me a summary. Structure it like this:
1. The map: the 4-5 key concepts I need to master, one line each.
2. The right questions: what I should ask myself before deciding.
3. The common mistakes: what people get wrong when they tackle this topic unprepared.
4. What to verify with an expert: the aspects where general information isn't enough.
Flag with [TO VERIFY] every point that depends on laws or figures that may have changed.
The operational syntax to compare options:
I have to compare these options: "owned solar panels versus rental".
Build a table with: criterion, option A, option B.
Use as criteria the ones that really matter to whoever decides, not the brochure ones.
After the table, tell me: for which kind of person A is worth it, for which one B is worth it.
Flag where your information might be dated and should be verified today.
The operational syntax to verify a piece of information:
I read this statement: "[paste the statement]".
Help me assess it. Separate:
- What is an established and verifiable fact.
- What is an estimate or a forecast (and whose).
- What is opinion presented as fact.
If you're going only on your memory without being able to check the sources, tell me clearly.
After the answer, make the move that changes everything: ask "give me the strongest argument against what you just said". Forcing the AI to contradict itself exposes the overly smooth answers.
A concrete example
Sara had to decide whether to enroll her daughter in a coding course or a music course and was getting lost among contradictory articles. She used the first prompt to map the topic "cognitive development and extracurricular activities", then the second to compare the two options with her own criteria (time, cost, what it actually trains), and finally she asked for the contrary argument.
The AI, forced to contradict itself, admitted that the studies on "coding makes you smarter" are often sponsored. Sara didn't get the answer from the AI: she got a grid to decide herself. She chose based on what her daughter liked, with the clear conscience of having looked at the topic from several sides.
When it does NOT work (and how to fix it)
If the AI invents sources, studies or numbers
The most frequent and most dangerous fault. The models generate references that look credible but don't exist. The way out is twofold: turn on web search so the AI cites real pages, and add to the prompt "If you don't have a verifiable source for a number, write: I don't have a reliable source, instead of estimating". Then actually open the important links: if the page doesn't exist, the data doesn't exist.
If it always agrees with you
AIs tend to go along with you, a flaw the technicians call sycophancy. If you frame the question in a slanted way, you get confirmations. The fix is structural: always ask for the contrary argument, and frame the question neutrally ("assess the pros and cons of X", not "why X is a good idea").
If the answer is huge and you don't know where to start
It means you didn't impose a format. Add to the prompt "5 points maximum, one line each, then tell me which to go deeper on". Useful research is the one you can navigate, not the longest one.
A tip from someone who actually uses it
Keep the AI for mapping and the first orientation, never for the final word on decisions that matter. It's excellent at telling you which questions to ask and where to look; it's unreliable on the last mile of precise facts. Use it like a librarian who points you to the right shelves, then open the books yourself.
Frequently asked questions
Can AI replace a Google search?
To get oriented and understand the structure of a topic, often yes, and it's faster. For precise, recent or official facts, no: you need the direct source. The best combination is AI to understand what to look for, the search engine to confirm the details that matter.
How do I know whether a source it cites is real?
Open it. If the AI gives a link, click it; if it gives only the name of a study, search for it by title. A source you can't find in two minutes should be treated as nonexistent, even if it sounds authoritative.
If I turn on web search, can I trust everything?
No, and this is the misunderstanding to dismantle. Web search reduces the inventions but doesn't guarantee the quality: the AI can cite a real page that is nonetheless wrong, biased or old. It changes the source of the problem, not the obligation to verify. Your judgment on the source's credibility remains irreplaceable.