Split two hours into five blocks: write a brief (15 minutes), run two AI deep research tools on it at once (30), collect demand and complaint signals yourself (30), check the claims your decision rests on (25), and write a one-page memo that says go, stop or test (20).

This is for founders deciding whether an app idea deserves the next month of their time. The running example is GroomBook, an imagined booking app for independent dog groomers in the UK. Tool features are as of October 2026.

See it in actionTwo hours of research in five blocks
Spend 15 minutes on a brief, 30 running two research tools, 30 on your own signals, 25 checking claims and 20 on a go, stop or test memo. Illustrative example. Use the playback controls or the thin timeline at the top to explore.

In short:

  • Agents read wide and fast. You judge what matters.
  • Two tools on the same brief show which findings overlap.
  • Trends, Keyword Planner and app reviews give signals agents miss.
  • "Test" is the most common honest outcome, and that's fine.

Write the brief around one decision

Research without a decision turns into reading. Write one sentence first: "I am deciding whether to spend four weekends building a booking app for independent dog groomers in the UK."

That sentence fixes the customer, the place, the job and the cost of being wrong. Add five questions under it:

  • What do these groomers use today?
  • What do they pay for it?
  • What do they complain about?
  • Who are the 5 to 10 direct competitors?
  • What changed in the last two years?

End with output rules: "Cite a source for every number. Mark inferences. Put competitors in a table with pricing page URLs. Say what you could not find."

Run two deep research tools in parallel

Most AI assistants now have a mode that searches the web for several minutes and writes a cited report. Claude's Research feature needs a paid plan and web search turned on. Gemini Deep Research has daily limits that depend on your plan. ChatGPT and Perplexity offer similar modes.

Paste the same brief into two of them. Different tools search and weigh sources differently, so overlap tells you something. A competitor in both reports is probably real. A market size in only one is a claim to check.

Each run can take several minutes or longer. Start the next block while they work. Our comparison of deep research tools for market research covers each one in more detail.

Collect the signals agents miss

Some of the best evidence isn't written up in articles. Collect four signals yourself in 30 minutes.

  1. Check the trend in Google Trends for two or three phrases, such as "dog grooming software". It shows direction on a 0 to 100 scale, not counts.
  2. Check rough monthly searches for the same phrases in Google Ads Keyword Planner. Ranges like 100 to 1K are enough.
  3. Copy ten phrases from recent 2 and 3-star reviews of the two biggest rival apps.
  4. Open each rival's pricing page, and compare it with a year-old copy in the Wayback Machine.

The reviews matter most. A line like "no-shows kill my Saturdays and this app can't take deposits" is a requirement in the customer's own words.

Check only the claims that change the decision

When the reports arrive, pull out the five or six claims that would change what you do. Put them in a claim table and check each source.

For GroomBook, the pricing claim holds and the big market size gets dropped.

Claim Source given Primary source? Keep?
Rival A charges per staff seat Rival's pricing page Yes Keep
30% of groomers use paper diaries Blog citing a 2019 survey No Re-check
Market will reach $X billion Report seller's press release No Drop

The rows are illustrative. Three rules make this fast. Open every cited page and find the exact claim. Prefer the source closest to the fact. Drop "global market" figures and size the niche bottom-up instead.

Write a memo that ends in one word

Ask an assistant to draft a one-page memo from your claim table and notes, then edit it. Use six lines:

  1. Decide: go, stop or test.
  2. Name the customer in one sentence.
  3. List three verified facts for the idea.
  4. State the strongest reason it might fail.
  5. Name the riskiest assumption.
  6. Plan the cheapest test of that assumption, doable in one week.

Two hours of reading can show a market exists. It can't show that people will pay you. For that you need contact with real customers, such as five interviews or a landing page test.

Agents read wide, you judge

AI agents are good at finding competitors, summarising pricing pages and grouping complaints. They are weak at very recent numbers, anything behind a login, and telling loud complaints from common ones.

Keep both roles. Let the agent read 50 pages. You decide which five facts matter.

How GrowApps research differs

When you start a new app in GrowApps, it asks about your relationship to the idea, the problem today and the win you want.

It then runs several rounds of web research on apps like yours and shows a research board with numbered links to the sources it read (how GrowApps works). That research decides what the first version includes. Use this two-hour workflow first, to decide whether to build at all.

Start with the one-sentence decision

Write your decision sentence now: who, where, what job, and what it costs you to be wrong. Add the five questions and paste the brief into one research tool before you close this tab. Then plan the test with how to validate an app idea in a weekend.

Frequently asked questions

Can AI do market research on its own?

It can do the reading quickly. It can't yet replace opening sources, spotting stale numbers and talking to customers. Treat an agent's report as a well-organised first draft with links to check.

How long should market research take for a small app?

Two hours is enough to decide whether to test. Then spend a week on the test itself. More reading rarely beats a week of contact with real customers.

Sources and further reading