Start where people complain, not where people pitch
Demand is visible in public, but not in the places built for promotion. Launch sites, pitch decks and trend reports tell you what founders and analysts want to be true. The places where somebody is stuck at 11pm and writes it down tell you what is actually true, because there is nothing to gain by writing it.
That is the whole reason this platform reads issue trackers and forum threads rather than press releases: a complaint written to nobody in particular is the least strategic sentence on the internet, and therefore the most reliable one.
Where demand shows up in the open
Five sources cover most of what a small team can realistically read. Each has a bias, and knowing the bias is how you avoid mistaking one community's taste for a market.
- Reddit and topic forums — where people ask for alternatives and describe workarounds. Biased towards consumers and towards whoever posts most.
- GitHub issues — where paying and technical users describe a missing feature precisely, often with the cost of not having it. The clearest evidence, and the narrowest audience.
- App-store reviews — where non-technical users say what broke and whether they left. The best source for a local market, because reviews are written in the local language.
- Hacker News and practitioner threads — where people compare tools and say what they gave up on. Biased towards early adopters, so treat volume with suspicion and specifics with respect.
- Support and community channels of an existing product — where the badly-served demand of a market leader is written down for free, by the people already paying for it.
The phrases that mark real demand
Demand has a grammar. "Is there anything that…", "alternative to…", "I ended up writing a script to…", "we pay $X a month for this and it still cannot…", "I gave up and went back to…". Each of these is somebody reporting a failed purchase decision, which is far stronger evidence than somebody saying a category is interesting.
The strongest single marker is a named price attached to a named frustration. It resolves the willingness-to-pay question in one sentence, and it is the marker most people skip past while collecting mentions.
Group the phrasings, then count people — not posts
The same problem is written thirty different ways, and each way looks small on its own. Grouping them is the step that turns scattered noise into a measurable market, and it is the step manual research usually gets wrong, because a person searching keywords finds the phrasing they thought of and misses the twenty-nine they did not.
Once grouped, count distinct people rather than posts. Fifty replies under one thread is one person's problem plus an audience; five posts by five strangers in five places is a market. This platform does the grouping with embeddings and reports both numbers for exactly this reason.
Check who already serves it before you get excited
Most gaps that look empty are simply unfamiliar. Before treating a demand as unserved, find the three products that already address it, read their pricing pages, and read their one- and two-star reviews. If those reviews complain about the same thing your evidence complains about, the opportunity is real and you now know its shape. If the reviews are about something else entirely, you have found a different opportunity than the one you were looking for — which is usually good news.
Check the direction, not only the level
A market with a thousand mentions that has been shrinking for a year is worse than a market with two hundred that has doubled in three months, and the raw count cannot tell them apart. Look at the last few months separately from the total. Anything you build takes months to reach a customer, so the number that matters is where the demand will be then, not where it has been.
The shortcut
Everything above is what this platform does continuously across Reddit, GitHub, Hacker News, app-store reviews and Vietnamese forums. The screener lets you sort the result by demand, by pain, by momentum or by the gap between demand and competition, and every row opens onto the posts the numbers came from — so the research is checkable rather than taken on trust.