Run a First Market Study With AI in One Hour
By Anis Hammouche·October 24, 2026
A first market study with AI can take an hour if you ask yourself three questions: who buys, how many of them there are, and who already sells to them. AI helps you structure the research and summarise what you find. You take the figures from official sources and check them yourself.
What AI does in an hour, and what it does not
When you are starting out, you need three answers: are there companies that buy what you want to sell, how many are there, and what do those who already sell it do today? An hour is enough to ask these questions properly, as long as you do not expect the tool to do all the work.
AI quickly drafts a study structure, suggests customer segments, lists the competitors it knows about and points you to sources to check. But it can give a figure that sounds plausible without any source behind it. A figure without a source is worth nothing in a study, and it can lead you to a bad pricing or targeting decision.
Step 1: frame the market with three questions
Start by writing your offer in one sentence, then ask AI to help you break down the market. The prompt below is a starting point:
I want to launch [your offer in one sentence] for [type of client].
Help me:
1. describe three possible customer segments, with their likely size
and the problem they are trying to solve;
2. list the types of competitors (direct, indirect, in-house solutions);
3. suggest five figures I should find in an official source
to test my hypothesis, without inventing any value.
For each figure, indicate the type of source where to look.
The last instruction matters most. It turns the AI's answer into a checklist, not a list of facts.
Step 2: find the figures in official sources
The strongest starting point is the Sirene register, which lists companies and their establishments. INSEE states that Sirene data is free, and it makes it available in three ways: the Annuaire des Entreprises for online search, an API updated every day, and files you can download from data.gouv.fr. INSEE's page on the Sirene register describes these access options. For a search without code, the Annuaire lets you filter by activity, location and headcount band.
| Information you need | Source to check | Limit to know |
|---|---|---|
| Number of companies by sector and area | Sirene register, Annuaire des Entreprises | The stock files also include closed companies: filter for active ones |
| Indicators by municipality | INSEE full file database | Use the most recent vintage shown on the page and note its date |
| Competitors and prices | Competitors' official sites, pricing pages | Note the date you checked, prices change quickly |
A count of companies gives you a range, nothing more. It does not tell you how much they buy. To estimate what a segment could bring in, you will need an explicit assumption, written down and tested with clients.
Step 3: check what the AI told you
Take each claim the AI made and ask one simple question: where does this figure come from? If the answer is vague, or if the tool cannot give you a link to a source you can open, remove the claim from your study.
Three checks cover most of it:
- the source exists and you opened it yourself;
- the figure really says what your study makes it say (one category can cover very different activities);
- the date of the data is stated and still relevant to your project.
If you use a research tool that cites its sources, click each link before copying it. A link that leads nowhere should make you doubt the rest of the answer.
Step 4: write the conclusion on one page
Write one page, not a fifty-page report. It should fit four blocks: your target in one sentence, the approximate size with its source, the competitors with what you observed, and your three hypotheses to test. On the last line, write what would change your mind: a price that is too low, a segment with no need, a free competitor.
This page is a starting point. It proves nothing. Read it again a week later, once you have talked to clients, and correct what no longer holds.
What the study does not tell you
Even a well-made market study does not prove that anyone will buy. It tells you where to look. Proof comes from talking to people who have the problem. I suggest starting with five twenty-minute conversations: they often teach you more than a week of desk research.
Ask each person three questions: what is the concrete problem, how do they deal with it today, what does it cost them in time or money. Do not ask whether they would buy your product. The answer is often yes, and it proves nothing.
Frequently asked questions
Can you trust the figures an AI assistant gives you? No, not without a source. A model can produce a plausible figure that does not match any real data. A figure only enters your study if you can open its source and check its date.
Which free sources should you start with? The INSEE Sirene register, the Annuaire des Entreprises and INSEE's statistical publications. They cover the number of companies and establishments, not their revenue or their purchasing budget.
How long does a useful study take? An hour is enough to frame the question and find the sources. After that, you need time to talk to a few people concerned, at your own pace. The first hour tells you what to check, not what to conclude.
Can AI replace client interviews? No. It can help you prepare the interview guide and summarise your notes, but the answers come from the people you meet.
Stuck on a specific case? Write to me through the contact page.
