Researching and synthesizing with sources

This lesson is still being tested: it improves with learners' feedback. If something is unclear, it is the lesson that needs fixing, not you.

By the end of this lesson, you will be able to

ask a precise research question: the topic, the country, the period, the level;
ask for a sourced synthesis, where each point has its link;
check the sources and spot what is missing;
know what deep research modes are for, and what they cost.

Why it matters to you

You need to quickly understand a topic you don’t know: a procedure, a technique, a market, a project. AI with web search saves you hours, as long as you know what its sources are worth. This lesson builds on lesson 4 of module 2: we go from a piece of information to check to a complete synthesis.

Before you start


You need	Your assistant, free account, with web search (module 2, lesson 4)
Cost	€0 (deep research modes depend on the plan)
Duration	40 minutes
Reminder	Put nothing personal in a search. If your assistant’s memory is turned on, it may be used to word the search (OpenAI’s help): for a neutral search, use a conversation without memory (module 2, lesson 5)

Understand it in 3 minutes

A precise research request:

What you specify	Example
The question	The steps to set up an association
The scope	In France, today
The level	For someone who knows nothing about it
The sources you want	Official sites first, for example service-public.gouv.fr (in French)
The form	6 points at most, each with its link
The gray areas	“At the end, tell me what you could not verify”

One source per point. A synthesis with three links at the bottom of the page doesn’t say which link proves what. Ask for the link next to each point.

Deep research modes (read in the official help pages on 24/10/2026):

Claude, “Research”: several searches that follow one another, each guided by the previous one. Reserved for paid plans; web search must be available (depending on how Claude is displayed for you, you turn it on with the + button, or Claude triggers it by itself when useful); it uses up your usage limit faster than a normal conversation; if you have connected tools (Gmail, Google Calendar…), it can also search them without asking you each time.
ChatGPT, “deep research”: it first proposes a research plan, which you can edit, then returns a report with its sources. The number of searches depends on the plan; a counter shows what you have left.

These modes are for long questions that cross many sources. For a short question, normal web search is enough. In both cases, the report gets checked like everything else.

Do it together

1Step 1: The request

Search the internet: what are the steps to set up an association in France?
I want a synthesis in 6 points at most, for a person who knows nothing about it.
For each point, give the link of the page that proves it, favoring official sites such as service-public.gouv.fr.
At the end, tell me what you could not verify.

2Step 2: Check three sources

Pick the three points that matter most. For each, open the link and answer the three questions from lesson 4 of module 2 (the official pages are in French: your browser’s translation feature can help you read them, but the French text stays the reference):

Does the page exist, and does it talk about this point?
Does it say the same thing as the synthesis?
How old is it, and is it an official or reference site?

Write down what you find: ✓ confirmed, ≈ half right, ✗ wrong or not found.

3Step 3: Have it corrected

For each point marked ≈ or ✗:

Point [number] doesn't match the page cited: it says [what you read]. Correct this point based on the page, or remove it if it has no source.

4Step 4: The final version

Here is the result of my check, point by point: [1: confirmed / corrected / not verified], [2: ...]. Make the final version keeping these statuses: the confirmed or corrected points with their source, then a "To be checked" line for the points not verified. Don't present a point that was not verified as confirmed.

Your turn

A real question from your life that calls for research: a procedure, a big purchase, a trip, a work topic, a technique to learn.

Write your request with the six elements of the table.
Check at least three sources, with ✓, ≈ or ✗.
Have it corrected, then ask for the final version.
Keep the synthesis with today’s date: it will get old.

Check your work

It worked if:

each point of the synthesis has its link, next to it;
you opened and read at least three sources, and noted ✓, ≈ or ✗;
at least one source is official or a reference;
what remains uncertain is written in a “To be checked” line, instead of being presented as sure.

Common mistakes:

a question that is too broad (“tell me about associations”): the synthesis stays general;
links at the bottom of the page without knowing which proves what;
taking a forum or a blog for an official source;
launching a deep research for a simple question: it takes longer, and it wears down your limit.

Keep it and pick it up later

Store your research request in your library, with brackets for [the question], [the country] and [the level].
Keep your syntheses dated; for a topic that changes (rules, prices, procedures), check again before using one again.

Going further

optional

For a question where accuracy matters, ask for two syntheses with different sources (“official sites only”, then “press articles only”), and compare. It is a real cross-check only if the two sources are independent: an article may repeat the official press release, so check where each one got its information.
On ChatGPT, the help says you start deep research by typing /Deepresearch, or from the tools menu (+); on Claude, Research is turned on with the + button, once web search is available.

New words

4 words explained

Word	What it means
Sourced synthesis	A synthesis where each point has its link to the page that proves it
Scope	The limits of a search: the country, the period, the exact topic
Reference source	A site that is an authority on the topic: the government, an official body, a software’s publisher
Deep research	A mode where the assistant chains many searches before returning a report (“Research”, “deep research”)