How to Validate a Startup Idea With Public Data (Not Surveys)

Most validation advice tells you to interview people or run surveys. Both have the same flaw: once you ask, you've changed the answer. Friends are polite. Survey takers perform. The cleanest demand signal is the complaint someone posted in public before they knew you existed.

We count those complaints for a living. Here is the five-step method we use, with the base rates we've measured across 836 tracked gaps, so you know what a real signal looks like before you spend a weekend building.

Step 1: Count people, not posts

A thread with 40 comments can be three people arguing. Count distinct humans who independently described the same problem. One person posting the same complaint in four subreddits is one person.

This single correction kills most "everyone wants this" claims.

Step 2: Know the base rate before you celebrate

Here's the distribution across our full dataset, as of August 28, 2026:

Distinct people asking	Gaps	Share
1-3	766	92%
4-7	47	5.6%
8-11	12	1.4%
12+	11	1.3%

Read that again: 92% of demand signals never get past three people. If you've found eight strangers independently asking for the same thing, you're in the top 3% of everything we've ever tracked. The quiet home server ask hit 8 people in 30 days — that made it the loudest signal of its month, not background noise.

The practical rule: 2-3 people is a hypothesis. 5+ is worth a landing page. 10+ deserves a prototype.

Step 3: Check what already ships — and read the complaints anyway

Search for existing tools before building. But don't stop at "competitors exist, idea dead." The more useful question is whether people keep asking despite the options.

Private email storage is our clearest example: our scan detected 20 shipping products, and people are still publicly asking. Twenty tools that haven't closed the ask means the products solve a different problem than the one being described — usually trust, setup pain, or pricing. That's an opening, not a wall.

One honesty note: any competitor scan has a recall ceiling. A tool with no web presence won't show up. Treat "we found none" as "we found none," never as "none exist."

Step 4: Look for a trigger

Demand with a date on it converts faster than ambient wishing. The strongest trigger we track is product death: when TV Time announced changes, users with a decade of watch history started shopping for an exit within days. Price hikes, API shutdowns, and acquisitions do the same thing. A gap with a trigger has users who are already moving; a gap without one has users who might move someday.

Step 5: Decide with thresholds you set in advance

Write down your kill criteria before you look at the data, or you'll rationalize whatever you find. Ours, for what it's worth: fewer than 4 distinct people after 30 days of signal, no verbatim quote describing real pain, and no trigger — park it. Meeting two of three — build the smallest thing that answers the ask.

What this method won't do

It won't validate B2B enterprise pain (that lives in sales calls, not subreddits), anything outside English-speaking communities, or demand people feel but never post about. Public-complaint counting is a filter, not an oracle. It's just a far better first filter than asking your friends.

If you'd rather start from counted demand than hunt for it, the August snapshot lists what people asked for last month, and our niche census shows where competition is actually thin.

---

All counts from GapMine's live dataset, same numbers the site renders. Every linked gap shows its public source posts — check our work.