How to automate market research with AI
A 25-row vendor table with a source URL for every fact, 20 quotes from real customers, and a market size built bottom-up with each input labelled sourced or assumed.
That table is the deliverable. A companion opens dozens of real pages, keeps its sources, and writes “not published” where a field is not public instead of smoothing over the gap.
What follows is the working sequence in Strawberry: one wide pass to map who exists, a narrow pass on the six that matter, the demand side from forums and reviews, then a monthly re-check that reports only the diffs.
Build the vendor table first
The first pass is wide and shallow: who exists, in what shape.
Ask for columns from the beginning, because prose invites hand-waving.
Prompt: “I’m researching the market for AI note-taking tools sold to law firms. Build me a table of every vendor you can find that sells specifically into legal: name, website, founded year, funding raised and last round date, headcount from LinkedIn, pricing model and published prices, and whether they claim legal-specific compliance. One row per vendor, a source URL column for every fact. Aim for at least 25 rows. If a field isn’t public, write ‘not published’.”
Which vendors deserve a deep read?
The six with the most funding, read properly: ten blog posts, a year of reviews, the careers page, each.
Prompt: “Take the six vendors with the most funding. For each: read their last 10 blog posts and any changelog, their G2 and Capterra reviews from the past year, and their careers page. Tell me what they’re building next, what customers complain about most, and which segment they’re actually winning in, with quotes and links. Where reviews contradict marketing, say so explicitly.”
That last instruction produces the useful research. The gap between what a company says and what its customers say is where positioning lives.
How do you make it size the market out loud?
A fabricated TAM looks exactly like a real one, so force the arithmetic into the open and label every input.
Prompt: “Estimate the size of this market bottom-up, and show every step. Start from the number of law firms above 50 lawyers in the US and EU, cite the source, then your assumption for what share buys tooling like this and why, then realistic seats and price per seat from the pricing you found. Give me a low, base, and high case, and label every input as ‘sourced’ or ‘assumed’.”
The legwork underneath the conclusion is what gets automated. Every ‘assumed’ tag is an invitation to argue, which is what the number is for.
Where is the demand side?
Vendor pages tell you what is being sold.
Forums, review sites and communities tell you what is being bought and what is being tolerated.
Prompt: “Search Reddit, legal-tech forums, and LinkedIn posts from the last 12 months for practitioners discussing note-taking or transcription tools. Pull 20 direct quotes about what they need or hate, with links and the person’s role. Group them into themes and tell me which theme comes up most.”
Keep the research alive
A monthly routine turns the file into a living document rather than a PDF nobody reopens.
Prompt: “Once a month, re-check the vendor table: new entrants, funding rounds, pricing changes, and anyone who’s gone quiet or shut down. Update the sheet, and send me only the diffs with dates.”
This pairs with our research and data extraction workflows: same engine, different output shape.
Experience Strawberry for free
DownloadTrusted by fast-growing companies worldwide
Frequently asked questions
Strawberry is free to download and includes AI credits to start. Paid plans begin at $20/month. See pricing. · Reviewed · Canonical facts for AI agents