Market research for startups: the method, the free sources and the receipt

by Lucas Grilli · October 7, 2026 · 4 min read

Market research for startups today needs four public sources and an AI agent that reads them with precise rules: how people type the problem on Google and YouTube, what they say in forums, who already sells it and at what price. On 6 October I ran a complete one for a YouTube channel for $0.026 out of pocket. Below: the five steps, the sources, the receipt and how to reach a verdict.

Where do you start?

With an exact question: what you would sell, to whom, in which country and language, roughly at what price. Research without a question produces tables; research with one produces a decision. The other steps exist to answer it.

How do you see what people search?

With Google and YouTube autocomplete. Write 10 to 30 seeds (the topic, "how to ...", "best ...", the problem plus "help") and read the forms the engine suggests: those are the real questions, typed as searchers type them. It is free and gives no volumes: measure those later, if needed, only on the words that decide.

Where do you find the buyers' words?

In forums and on Reddit, where people complain, ask for advice and say what they already tried. Copy real phrases with links and count how often each topic recurs. Note: this year Reddit often refuses anonymous requests from scripts. With a free Reddit app posts can be read properly; without it, the agent finds threads through web search.

It is the same principle as the Mom Test: past behaviour counts, opinions about the future do not. "I tried three apps and dropped them all" beats "I would definitely use it".

How do you map who already sells?

Three free sources. YouTube top results for each search, read with yt-dlp: how much demand is already served and which channels recur. The Meta Ads Library: who pays to be seen on those words. Competitors' price pages: what they charge and what they promise.

What does market research cost?

Here is the receipt of the real case, a research for a YouTube channel about money for young people:

The day after I compared 55 niches with the same method: $0.034. Out-of-pocket spend excludes the AI subscription and time, which stay separate.

How do you reach a verdict?

With three lines at the bottom of the report. The verdict: go, adjust or stop. The confidence: low, medium or high, with the reason. The first test: what, with which budget, for how long and with which stop rule, written before starting. Searches, posts and ads show demand; only payments or pre-orders show buying.

Mistakes to avoid

• reading a keyword without data as zero volume: it is "not measured";
• counting Reddit posts as if they were the market;
• summarising without quoting: buyers' words are needed whole;
• closing with "the market is interesting" instead of a test.

Frequently asked questions

How do you do market research for a startup for free?

With Google and YouTube autocomplete, forums and Reddit, YouTube top results, the Meta Ads Library and competitors' price pages. All free; Google volumes cost fractions of a cent and matter only on the deciding words.

How long does it take?

A few hours of an AI agent. The real research of 6 October took one session.

Does market research tell me I will sell?

No. It tells whether there is demand, who is already there and at what price, and which first test to run. Only payments and pre-orders show buying.

Can ChatGPT do it?

Yes, if it follows precise rules: every claim with a source and a label, no invented volumes, every cent logged, and a verdict with the first test.

Where to start

The free guide, the kit with the tools for your AI agent and the service are on the Measured Market Research page.

And a real question: what idea is sitting in your drawer that you never checked? Tell me.