Which tool to choose
- Research on public, current facts: an assistant with built-in web search (ChatGPT or Gemini) searches, reads multiple pages and synthesizes on its own, citing the sources.
- Research on your own materials (PDFs, articles, notes): NotebookLM works only on what you load and anchors each sentence to the exact point in the document.
- Real-time news and trends, especially from X: Grok is useful for what's moving now, to be used as a signal to verify, not as a definitive source.
How to do it
- Write the research question, not a vague topic. "How has the average rent price in Bologna changed over the last five years" works; "Bologna rents" doesn't.
- Define depth, sources and format in a single instruction. The operational syntax:
Do research on [question]. Consult multiple independent sources, prioritize official and recent data. Give me back: a ten-line synthesis, the three main figures with the source and link for each, and the points on which the sources disagree. Don't include anything you can't attribute to a source.
- Let it work, then read the sources first. Open the links: a source that doesn't exist or that doesn't say what the AI claims is the sign that the synthesis should be taken with a grain of salt.
- Ask for the disagreements, not just the summary. The most useful part of research is where the sources diverge: that's where you understand the topic is open and that a single smooth synthesis would hide the complexity.
- Have it dig into the weak points. The operational syntax:
On point [X] the synthesis is generic. Dig deeper with specific sources and give me the precise figures, always with the link.
A concrete example
Elena has to decide whether to open a bike rental business in her city. She asks the AI for research on local tourist demand and on the competitors. The AI consults the municipality's website, two industry portals and the competitors' reviews, and produces a synthesis with the figures and the links. Elena opens the sources: the figure on tourists comes from the municipality, true; the number of competitors was underestimated, because the AI had counted only the top results of a map. Elena corrects it by checking in person. The research saved her an afternoon, but she makes the decision on the data she verified.
When it does NOT work (and how to fix it)
If it cites sources that don't exist or don't say what it claims
The AI can invent a plausible link or attribute to a source a figure that isn't there. Remedy: open every source you'll use to decide. If a link doesn't open or the page doesn't contain the cited figure, discard that statement: an unverified source is worth zero.
If the synthesis is generic and doesn't really answer the question
This often comes down to a question that's too broad. Remedy: narrow it ("not rents in general, but two-room apartments in the center") and ask for precise figures instead of trends. A specific question produces specific research.
If you can't find how to enable web search
Interfaces change and not all plans include it. Remedy: look for a globe-shaped icon, a magnifying glass or the entry "search the web." If it isn't there, do the search yourself on the search engines and paste the pages for the AI to synthesize, asking it to base itself only on those.
A tip from someone who actually uses it
Always ask what the AI couldn't find. Honest research has gaps: paywalled sources, non-public data, missing years. An AI that presents a synthesis without admitting gaps is filling the holes with estimates disguised as facts. Knowing what's missing is worth as much as knowing what's there.
Frequently asked questions
Can I trust a synthesis made by the AI for an important decision?
The structure and the gathering, yes; the figures, only after verifying them at the source. The AI is an excellent first pass of research, not the last word. For a decision that carries weight, you always check the key data in person.
How many sources does the AI actually consult?
It depends on the tool and the question, but rarely as many as you would consult for a serious topic. It tends to stop at the top results. Remedy: ask it explicitly to look for sources beyond the top results and to prioritize official ones over aggregators.
Is it true that the AI does research better than a person?
Faster, not better. It gathers and summarizes at a speed no one matches, but it doesn't tell an authoritative source from a poor one the way an expert eye does, and it doesn't know what it doesn't know. The speed is its, the judgment on quality stays yours.