Every idea feels obvious to the person who had it. That feeling is not evidence. Validation is the work of finding out, cheaply, whether anyone else shares it, and it goes in a specific order: the cheapest test that could prove you wrong comes first.
Why building first is the expensive way
Building an app takes weeks and money even with an AI assistant doing the typing. At the end you learn one thing: whether people use it. You could have learned that in an afternoon, and you would still have the weeks.
The tests below cost, in order: nothing, a few conversations, one afternoon, one day, and one real transaction. Most ideas die at test two, which is the point. The ones that survive to test five deserve your weeks.
Test 1: say it in one sentence
Fill in this sentence without using the words "platform", "solution", "seamless" or "AI":
It helps [a specific kind of person] who struggle with [a specific thing that happens] by [what changes].
Good: "It helps independent hairdressers who lose bookings to missed phone calls by letting clients book from Instagram."
Not yet: "It helps small businesses be more efficient with AI."
If you cannot fill the blanks, you do not yet have an idea; you have a direction. Everything else in this list will give you mush, because you will be asking mushy questions. Cost: ten minutes.
Test 2: find five people and ask about the last time
Find five people who match the first blank. Not friends, not family: people who would actually pay. Ask them to tell you about the last time the problem happened.
Ask exactly this:
- When did this last happen to you?
- What did you do about it?
- What did it cost you, in money, hours or customers?
- What have you tried already?
- What would have to be true for you to change how you do it?
Never ask "would you use an app that…". People say yes to be nice. The last-time question cannot be answered politely, because either it happened or it did not.
Cost: a few hours over a week. This kills more ideas than any other test, which makes it the most valuable one.
Test 3: do it by hand for one person
Before you automate anything, be the machine yourself, once, for one real person.
If the idea is "AI sorts your invoices", ask one business owner to forward you their invoices this month, sort them yourself, and send the result back the way your app would. If the idea is a meal planner, plan someone's week by hand and send it as a message.
You learn three things a prototype would not tell you:
- Whether the output is actually useful. People often accept a result politely and never act on it.
- What the edge cases are. The invoice in a foreign currency, the receipt that is a photo of a photo. Those cases will be most of your work later.
- What they do with it next. Often the thing they do next is the real product.
Cost: one afternoon. Do this even if you are certain. Especially if you are certain.
Test 4: a page that describes it, and a button
Make a single page that explains the promise, and a button that says "Get early access". When someone clicks, ask for an email address.
Send that link to the people from test 2 and to one or two places where your audience already gathers. A local business group, a subreddit, a professional forum. Be honest that it does not exist yet.
Watch the ratio: out of everyone who read the page, how many left an address? A handful out of a hundred is normal and means the promise is understood. Zero out of fifty means the promise is not landing, and rewriting the page is cheaper than building anything.
Cost: an afternoon. An AI assistant will build this page for you in minutes; see how to install Claude Code.
Test 5: ask for money before it exists
The only test that proves demand is someone paying. Offer the thing you did by hand in test 3 as a paid service, at a real price, to the people who left an address.
If three people pay for a manual version, you have a business with a delivery problem, which software solves. If nobody pays but everyone is enthusiastic, you have a hobby, and that is a legitimate thing to have as long as you know it.
Cost: one uncomfortable conversation.
What each test does not prove
| Test | Proves | Does not prove |
|---|---|---|
| One sentence | You know who it is for | That they care |
| Five interviews | The problem is real and costly | That your solution is the one they want |
| Do it by hand | The output is useful and what it takes to produce | That it works at scale or unattended |
| Landing page | The promise is understood | That anyone will pay |
| Pre-sale | Someone will pay this price for this result | That they will keep paying |
Four traps
- Asking people who love you. Friends validate the person, not the idea.
- "Everyone" as an audience. If it is for everyone, you cannot find five of them to interview, and you will never write an advert.
- Falling for the feature. You get excited about the clever part. Customers buy the boring outcome.
- Building to avoid the conversations. Building is comfortable and feels like progress. Rejection is the actual work.
Where the technical question fits
Validation tells you whether the problem is worth solving. It does not tell you whether the solution is buildable, what it would cost to run, or whether the clever part needs AI at all. Those are separate questions, and getting them wrong is how validated ideas still fail.
Run the free check on this site with the sentence from test 1 and you will get a read on the second half: where AI genuinely fits, what the risks are, and a first step. Then check whether it needs AI at all, because that decision changes the cost of everything that follows.