Demand test
Smoke Test a Startup Idea — Measure Demand Before You Build
A smoke test puts the whole idea in front of strangers — one page, one promise, one action — before the product exists. It answers “will anyone reach for this concept?” It does not answer “can we engineer it?” and it is not a fake-door button inside a product people already use.
Last updated: September 25, 2026
Direct answer
What is a smoke test for a startup idea?
A smoke test is a whole-concept demand experiment: you describe the outcome, show a single call-to-action, and watch whether qualified people take a concrete next step — usually a waitlist, a reply, or a small deposit — without shipping the product. Eric Ries (The Lean Startup, 2011) used smoke tests to separate “can we build it?” from “should we?” Giff Constable's later write-up at smoketestmethod.com is the operational definition this page follows.
Key takeaways
What to remember
- Write the bet and the fail line before any visitor arrives. A rate you invent after a weak week is not a result.
- One audience, one promise, one CTA. Extra options measure design preference, not demand.
- Traffic quality beats volume. Wrong visitors produce false negatives — and false confidence if they click out of curiosity.
- A waitlist is honest only if you say the product is not ready and you follow up. Silence after a signup is the unethical version.
- Interviews explain the pain. A smoke test counts whether strangers will reach. Run both if you can; do not substitute one for the other.
Don't mix these up
Smoke test vs fake-door test
Founders (and a lot of blog posts) use the two names as synonyms. They are related demand tests, but they sit in different places. Lean Analytics (Croll & Yoskovitz, 2013) describes the painted-door / fake-door pattern as a control that is not wired up yet — typically a button or menu item inside something people already use. A smoke test, in the Ries / Constable sense, is the whole concept on a page that exists only to run the experiment.
Smoke test — the whole idea
No product yet. You publish a landing page that names who it is for, what outcome you promise, and one action. The visitor has never used you. The question is: will they reach for this concept?
Fake door — one feature, existing product
People already use something you shipped. You add a button or menu for a feature that is not built, and you measure clicks from users who already trust you. The question is: will they reach for this addition?
Read the fake-door playbook →Before you build the page
Write one falsifiable hypothesis
If you cannot finish the sentence below, you are not ready to buy traffic or post in a community. The hypothesis names the person, the recent pain, the action, the outcome, the timebox, and the channel. Missing any one of those turns the test into a vibe.
People who [role] and recently [pain] will [action] for [named outcome] within [timebox] when they arrive from [one channel].
Example shape, not a result: “Independent tutors who invoiced late last month will join a waitlist for a scheduling-and-invoicing tool within 14 days when they arrive from a tutoring Facebook group.” If that sentence feels fake, rewrite the role or the pain — do not decorate the page.
The page
A thin, honest page structure
You are building an instrument, not a brand site. Five parts. Anything else is usually a way to hide a weak promise.
- 1
Headline = who + pain + outcome
One sentence a stranger can reject. “Invoicing that does not eat Sunday night, for independent tutors” is a promise. “The future of education ops” is a mood.
- 2
Three to five lines of what happens next
What they get if they act, when you will reply, and what you will not do. Do not invent a feature grid you cannot deliver this month.
- 3
One call-to-action
Waitlist, “reply with your last invoice story,” or a small deposit if you can actually take money with a refund path. Two buttons split the signal.
- 4
An honest footer
Say this is early access or a waitlist. Name the person behind the page. Collect the minimum contact detail you will actually use.
- 5
Leave off the brochure
No fake testimonials, no press logos you do not have, no chat widget that turns the test into unrecorded conversations. If you need social proof you do not have, you are not ready to pretend.
Traffic
Send people you can name
There is no universal visitor count that “proves” an idea. Constable's smoke-test write-up is about a small experiment on a critical assumption, not a traffic trophy. Pick one channel you can actually reach, record the source, and decide in advance how many qualified visitors you will wait for before you read the rate.
A community you already participate in
One honest post in a group where the ICP already talks. Spam is not a channel. If you cannot describe why you belong there, pick a different source.
A short outreach list
Twenty to fifty named people who match the role. Personal notes, not a blast. This is slow and often cleaner than a random ad click.
Search or a small paid burst
Useful when the query is specific and you can discard junk traffic. Paid clicks from the wrong job title are not “more data.”
Your existing list — with a caveat
Friends and newsletter readers who already like you inflate signups. Use them to debug the page, then read the result from strangers separately.
Pass / fail
Set the thresholds before anyone visits
A smoke test without a pre-written bar becomes a story you tell yourself. Write the five lines below on the same day you write the headline. Do not borrow someone else's dashboard screenshot as your pass rate — channels, offers, and audiences are not interchangeable, and this page will not invent a universal conversion number.
- Qualified visitor: who counts (role + source). Everyone else is discarded, not averaged in.
- Conversion event: the one action you will count (waitlist, reply, deposit). Clicks on the logo do not count.
- Minimum sample: how many qualified visitors you will wait for before you read the rate.
- Pass line and fail line: two numbers you can live with. “Interesting” is not a line.
- Timebox: the calendar date you will decide, even if the sample is ugly.
Signal strength still ranks the same way across tests: a deposit beats an email, an email beats a click. Compare yourself to the bar you wrote — not to a blog that published a percentage without naming the channel.
Ethics
An honest waitlist is still a waitlist
The test is unethical when you imply the product is for sale today and then disappear. It is a normal early-stage experiment when you say what happens next and you do it.
Say the product is not ready
“Join the waitlist” and “get early access” are honest. “Start your free trial” is not, if there is no trial.
Follow up inside a few days
A short email that asks about their last incident is the minimum. Collecting addresses you will never write to is just a list, not a test.
Do not charge unless you can refund or deliver
A deposit with a clear refund is a pre-order, which is a stronger test — and a different legal posture. If you cannot deliver something useful soon, do not take the card.
Collect the minimum
Email (or a reply thread) is enough for most smoke tests. Do not harvest phone numbers, company size, and a nine-field form “for later.”
Which test
When to smoke-test vs interview vs pre-sell vs fake-door
Pick the cheapest test that answers the question you actually have. Running the wrong one feels productive and teaches the wrong thing.
Use a smoke test when the idea does not exist yet
The expensive unknown is demand or messaging. You can name an ICP and a channel. You want a count, not another compliment.
Use interviews first when you cannot name the pain in their words
If the headline is still yours, not theirs, a page will measure copy you invented. Run the Mom Test script, then come back.
Use a fake-door when a product already has users
You are testing one addition, not the company. Put the button where people already work. See the fake-door guide — do not rebuild the marketing site to test a menu item.
Use a pre-sale when demand looks real and you can take money honestly
A charged card is stronger than a waitlist. Only do this with a delivery promise and a refund path. Otherwise you have invented a worse smoke test.
Common mistakes
What usually ruins the test
Declaring a pass rate after the fact
If the fail line moves every time the graph looks sad, you are not testing. You are coping. Write the bar first.
Building a five-section marketing site
Pricing tables, feature grids, and three CTAs hide a weak promise. The page should be boring enough that only the offer can save it.
Reading friend traffic as market demand
Warm clicks are a debug tool. The decision uses strangers who match the ICP. Mix the two and you will ship a product for your group chat.
Never talking to the people who signed up
A waitlist without follow-up interviews tells you a headline was interesting. It does not tell you the problem is the one you plan to build.
Treating a smoke test as a fake door, or the reverse
A landing page for a company that does not exist is a smoke test. A dead button in an app people already open is a fake door. Use the matching playbook.
Sources
Where these ideas come from
- Eric Ries, The Lean Startup (Crown Business, 2011) — Smoke tests as early demand experiments — separate “can we build it?” from “should we?”
- Giff Constable, “What is a Smoke Test?” — The operational definition: a small experiment on a critical assumption before you build.
- Alistair Croll & Benjamin Yoskovitz, Lean Analytics (2013) — Painted-door / fake-door as a feature-level control that is not wired up yet — not the same surface as a concept landing page.
- Rob Fitzpatrick, The Mom Test (2013) — How to talk to the people who converted (or bounced) without leading them into a polite yes.
With Yibud
Use the report to decide whether a smoke test is the right next move
Yibud is a free, no-signup validator. Scores are deterministic; optional AI only polishes wording. If distribution or “will anyone reach for this?” is the weak dimension, a smoke test is often the cheapest experiment. Analyze the idea first so you are not testing a random headline.
Analyze my idea →FAQ
Questions founders actually ask
Is a smoke test the same as a fake-door test?
No. A smoke test is a whole-concept landing page plus one CTA, run before the product exists. A fake-door test is a feature-level button or menu inside something people already use. They share a family — both measure a reach — but they answer different questions. Use the fake-door guide for the in-product version.
What conversion rate counts as a pass?
The rate you wrote down before traffic started. There is no universal pass number on this page, because channel, offer, and audience change the base rate. If you did not write a bar, you do not have a result yet — you have a story.
Do I need paid ads?
No. Paid ads are one channel. A community post or a named outreach list can be enough if the people match the ICP and you record the source.
Is a waitlist unethical?
It is unethical if you imply the product is ready and vanish. It is honest if you say this is a waitlist or early access, collect the minimum, and follow up.
Can I ask for payment on the smoke-test page?
Yes, if you can deliver something useful soon or refund cleanly — that is a pre-sale, and it is a stronger signal. If you cannot, keep the CTA at waitlist plus a fast interview.
How is this different from a normal marketing landing page?
A marketing page often assumes the product exists and tries to convert an ongoing audience. A smoke test is an experiment with a written bet, a timebox, and a stop rule. The page is the instrument.
What if nobody converts?
First check whether the visitors matched the ICP and whether the promise was clear. If the right people still ignore it, treat that as evidence against this bet. Rewrite the promise or the segment before you build.
How does Yibud fit in?
The analyzer highlights which assumption is most expensive. When that assumption is demand, this guide is how to run the count. When the assumption is still “what words do they use?”, start with interviews instead.
Know which risk to test first
Yibud scores the weak assumptions with a deterministic engine, then you pick the experiment — a conversation, a smoke test, or something cheaper.
Generate a Yibud report →