Find the thing worth making

Why one evening is enough

Most people spend months building the wrong thing because they never spent one evening checking whether anyone wanted it. The check is cheap. You need a laptop, three free tools, an AI assistant and about three hours. At the end you will have one of three verdicts: go, narrow, or kill. All three are wins, because two of them save you months.

We run this check before every project at Hapngo, including this site. It has killed more of our ideas than it has approved, and that is the point.

Start with a problem, not a niche

"AI tools for teachers" is a niche. "Teachers spend Sunday evenings rewriting the same worksheet for three ability levels" is a problem. Niches are crowded and vague. Problems are specific and searchable. Write your idea as a problem sentence before you open a single tool. If you cannot write it, you have a topic, not an idea, and you should keep reading until you find the problem hiding inside it.

Hour one: does anyone search for it?

Search intent is the strongest demand signal you can get for free, because people type what they want and nobody is watching.

• Type the problem into Google three ways: as a question ("how do I..."), as a product ("best ... for ..."), and as a complaint ("... not working"). Look at what ranks. Forums and Reddit threads on page one mean the demand is real and under-served. Big brand pages mean demand is real and contested. Nothing relevant means nobody is looking, or you are using the wrong words.
• Open Google Trends and compare your problem term against two neighbours over five years. You want a line that is flat or rising. A spike that fell off a cliff is a fad. A slow rise over three years is the best pattern there is.
• Use the free tiers of Semrush or Ahrefs to get a rough monthly search volume and a keyword difficulty score. Free limits are tight (Semrush allows about ten queries a day without a paid plan), so spend them on your five most important phrases. A volume of 500 to 5,000 searches a month with low difficulty is the sweet spot for a solo creator. Millions of searches means you are up against companies with SEO teams.

Write down the numbers. Guesses evaporate; numbers survive to the next step.

Hour two: listen where people complain

Search tells you what people want. Communities tell you how badly they want it and what they have already tried.

• On Reddit, search the problem sentence and sort by "Top" over the past year. Read twenty threads. Note the exact words people use, the tools they mention, and the moment someone says "I would pay for this" or "I built a spreadsheet for this". A homemade spreadsheet is the single best demand signal on the internet. It means the pain is bad enough that someone did work to fix it.
• On Hacker News, use the search box and look at "Ask HN" and "Show HN" posts near your problem. The comments are blunt and technical. If three people have already shipped something similar and all of them are quiet now, ask why.
• Check YouTube comments under the top three videos on the topic. Comments asking "but how do I do X" are unanswered demand.

Copy the best complaints into a document, in their words. You will use these sentences as headlines later, and they will convert better than anything you invent.

Hour three: ask the AI the right questions

ChatGPT, Claude and Perplexity are useful here, but only if you ask them to attack the idea, not praise it. All three will happily tell you any idea is wonderful. Make them work.

Questions that produce something useful:

• "List the ten products people currently use to solve [problem]. For each, give the price and the most common complaint in reviews." Perplexity is best for this because it cites sources you can click.
• "What would have to be true for a solo creator to compete with those products? Be specific about what they cannot do."
• "Play a sceptical investor. Give me the three strongest reasons this idea fails within a year."
• "Which of these three sub-problems has the most people who would pay, and why?"

Cross-check every claim the assistant makes against what you found in hours one and two. If it names a competitor you have never seen in search results, look it up. If it invents a statistic, drop it. The assistant is a fast research partner, not a source.

What a GO looks like

You have a go when all four are true:

1. There is measurable search demand: at least a few hundred searches a month on phrases that describe the problem, with a trend line that is flat or rising.
2. People are already paying to solve it, badly. Existing products with public complaints are better than an empty field. An empty field usually means no money in it.
3. You found real people describing the problem in their own words, and at least one of them built a workaround.
4. You can name the first thing you would publish or sell, and describe who would use it in one sentence.

A narrow verdict is when demand exists but the problem is too wide. "Meal planning" is wide. "Meal planning for shift workers who cook once a week" is narrow, and narrow wins. Cut the problem down until the search volume gets small but the people get specific.

A kill is when there is no search demand, no community chatter, and the AI can only offer generic encouragement. Killing an idea in one evening costs you nothing. Killing it after six months of building costs you six months.

What to write down before you close the laptop

One page, nothing more:

• The problem sentence.
• Five search phrases with volume and difficulty.
• Ten quotes from real people.
• Three competitors with prices and their most common complaint.
• Your verdict and one line on why.

That page is the brief for everything that follows. Step two builds a name on it.

Checklist

• Written the idea as a problem sentence, not a niche
• Checked search demand and a five-year trend line
• Read twenty real complaints and copied the exact wording
• Made an AI assistant argue against the idea and checked its claims
• Recorded a go, narrow, or kill verdict on one page