How to Run a Survey That Actually Tells You Something
A survey that actually tells you something asks about what people have already done, not what they think they’d do, keeps every question specific and neutral, and reaches the exact people who have the problem you’re solving. The moment you ask someone to predict their future behavior or agree with your idea, you’ve built a machine that produces the answers you were hoping for. That is worse than no data, because it feels like proof.
Most teen founders reach for a survey too early because it feels productive. You send a Google Form to 60 people, 40 respond, 85% say they’d “definitely use” your app, and you walk away certain. Then you build the thing and nobody shows up. The survey didn’t lie on purpose. You asked it the wrong questions, in the wrong order, of the wrong people.
When should you even run a survey?
Later than you think. A survey is good at one thing: measuring something you already roughly understand across a lot of people. It is terrible at discovery. If you don’t yet know why people struggle with a problem, a form full of multiple-choice options will only reflect the options you already guessed.
So run customer interviews first. Talk to eight or ten people, let them surprise you, and figure out the real shape of the problem in their words. Once you’ve heard the same complaint three times and you want to know how common it is, then a survey earns its place. The rule of thumb: interviews tell you what questions to ask, surveys tell you how many people would answer them the same way.
If you’re still at the “is this even a real problem” stage, you’re not ready for a survey. Start with how to validate a startup idea in high school instead.
Why do most startup surveys produce garbage?
Because they’re written to be reassured, not to learn. Three bugs show up in almost every first survey:
- Leading questions. “How much would you love an app that finally makes studying easy?” You’ve told the person the answer before they’ve read the options. Nobody picks “I’d hate it.”
- Hypothetical questions. “Would you use this?” and “Would you pay $5?” ask people to predict their own future. Humans are famously bad at this and famously generous when the prediction costs them nothing.
- Wrong audience. You post the form in your group chat and your school’s Discord. Those people like you. They’ll answer to be nice, and none of them are the customer you’d actually sell to.
The same disease behind bad interviews infects bad surveys: people lie to be kind. The fix is the same too. The Mom Test rule — ask about real, past behavior instead of your idea — works just as well on a form as it does in a conversation. And if your evidence so far is that your friends said they love it, read why “my friends love it” is not validation, because a survey of those same friends is just that trap with a spreadsheet attached.
What does a good survey question look like?
It asks about something that already happened, offers neutral options, and never hints at the “right” answer. Compare these directly:
| Bad question | Why it fails | Better version |
|---|---|---|
| ”Would you use an app that organizes homework?” | Hypothetical, leading | ”How did you keep track of your homework last week?" |
| "Would you pay $5/month for this?” | Predicts a future purchase | ”What tools do you currently pay for, if any? What do they cost?" |
| "How annoying is forgetting assignments?” | Assumes the pain, leads the answer | ”In the last month, how many assignments did you forget or turn in late?" |
| "Do you like studying with friends?” | Vague, no behavior | ”The last time you studied for a big test, did you study alone or with others?” |
See the pattern? The good versions all point at the past. The past already happened, so nobody can flatter you with it. “How did you track homework last week” forces a real, specific answer — “I use the Notes app and forgot two things” — that you did nothing to invent. That single honest answer is worth more than 50 people clicking “definitely.”
How to build a survey step by step
Keep it short. Every extra question loses you responses and tempts you to add junk. Here’s a build order that works with zero budget and a free Google Form or Tally:
- Write your one real question first. What is the single thing you need to know? “How do people currently deal with X?” Everything else supports that. If you can’t name it, you’re not ready to survey.
- Add a screening question up top. Something like “In the last 30 days, have you [done the thing my product is about]?” If they answer no, they’re not your audience — you can politely end their survey there and keep their noise out of your data.
- Ask about past behavior, not opinions. Frequency, last time, what they used, what it cost. Concrete and countable.
- Save any opinion questions for the end, and treat them as flavor, not proof. “What frustrates you most about X?” is a fine open-text closer. Just never let it be your evidence.
- Cut every question that won’t change a decision. If a question’s answer wouldn’t make you build differently, delete it. You do not need their grade level “just in case.”
- Test it on two people before sending. Watch them fill it out. Every confusing word or leading phrase will reveal itself in the first 90 seconds.
Aim for five to eight questions. A survey people finish beats a thorough one they abandon.
Where do you find people who aren’t your friends?
This is the part that decides whether your survey means anything. Your group chat is the wrong sample — those people are biased toward you. You want strangers who have the problem.
- Post in communities where your customer already hangs out — a subreddit for the hobby, a Discord server for the class, a Facebook group parents use. More on that in how to find where your customers already hang out online.
- If you’re stuck talking only to people you know, where to find people to interview applies to survey respondents too.
- Offer a tiny incentive if you can afford it. A $5 gift card raffle for one respondent out of 30 costs you a coffee and roughly doubles response quality, because people who want the gift card still have to have the problem to get past your screening question.
One warning: do not chase a big number for its own sake. Thirty honest responses from real customers beat 200 from classmates who’ll never buy. If you want to think clearly about how many responses is “enough,” the logic in how many customer interviews do you actually need carries straight over to surveys.
Reading the results without fooling yourself
When responses come in, your brain will hunt for the numbers that make you feel good. Fight it. Do the opposite: go looking for the answer that would kill your idea, and see if the data supports it.
Weight behavior over opinion, always. If 70% say the problem “really annoys” them but only 10% could name a time it actually cost them something in the last month, believe the 10%. Talk is cheap; the specific memory is the truth. And a survey never tells you someone will pay — for that you need people to actually put money or a real commitment down, which is what a presell or a fake door test measures. A survey narrows down who to test on. It doesn’t replace the test.
Running a real validation sprint — surveys, interviews, and small tests stacked together until you actually know something — is exactly what we walk you through in the first week of the batch0 program. If you’d rather do it with a cohort and a mentor checking your questions before you send them, apply here. It’s free to apply, and you only pay if you get in.
Your survey has one job: to tell you the truth even when it’s inconvenient. Write it so it can’t do anything else, send it to people who don’t care about your feelings, and read it looking for reasons you’re wrong. Do that and a survey stops being busywork and becomes the cheapest way you have to find out whether anyone actually wants what you’re about to spend a month building.