An ad test — run by hand or by a start-a-business app — asks three questions in a row. Can you reach the people you believe are your market? Do they respond to the way you describe what you are making? And once they land on your page, will they act on a priced offer — starting with an email given after seeing the price?
Each link can fail on its own, which is what makes the test useful — and it is far better at answering no than at answering yes. Almost nobody clicking is real information. Clicks that produce nothing is real information. A healthy pile of email addresses is not proof that you have a business; it is permission to keep going. And an email given after seeing the price is the rung directly below a payment. The same test can be run further up that ladder — a pre-order, a deposit, a first payment — when you want evidence that is no longer leading. This article covers the bottom rungs, because they are the cheapest to buy; the ones above are the same instrument with a sharper ask.
That is still worth a few hundred dollars, because it is quantitative evidence about how a market responds to your idea, gathered in a week, before you spend months building. If you are doing this while keeping your job, that is the whole appeal: a week of evenings and a few hundred dollars buys the answer you would otherwise quit to find out. This article is the traffic half of the whole demand test; the page you are sending people to is the other half, and it should already be live before you read on.
Run by hand, the rest of this article is the job. Run with Draper, the mechanics compress: it works up the message variants and the ad creative alongside you, you approve them and set the budget, and it builds the campaign and puts it live on the platform your buyers are on. When the numbers come back, they land in the same conversation, and the next move — rewrite the promise, try a different angle, or stop — gets made with the results in front of you.
1. Pick the platform your buyer is on
Choose by where your buyer already spends time, not by which platform you find most tolerable.
Meta is the usual right answer. Most buyers are there, and its targeting is the most developed of any self-serve platform — which matters most for consumer products, where the audience is defined by interests and behaviour rather than by job title. Minimum daily budget is $1 for impression-based campaigns and $5 for anything optimised for clicks or conversions.
X works, and we have run this ourselves. Draper's own messaging test ran on X: several angles for the same product, one week, a few hundred dollars. There is no minimum campaign spend.
Reddit suits products whose market gathers around a problem rather than a demographic — a trade, a condition, a piece of software everyone hates. Minimum daily budget is $5 with a $25 lifetime minimum, and its CPMs generally run below Meta's. If you already know which rooms your customers are in, this is the platform where that knowledge converts directly into targeting.
If you cannot answer where your buyer is, stop here. That is an earlier question and this test cannot answer it — it can only spend money failing to.
2. Optimise for clicks, not conversions
Every platform will ask what you want it to optimise for, and the obvious answer is the thing you actually want: signups. At this budget that is the wrong answer, and it is the single most common way a founder's ad test comes back unreadable.
Meta's optimiser needs roughly 50 optimisation events per ad set per week to leave what it calls the learning phase. That threshold has not moved. Below it, the ad set sits in a state Meta labels "learning limited" — it never gathers enough signal to stabilise, so it keeps guessing about who to show your ad to. A few hundred dollars of signups will not come close to fifty in a week. Your entire budget gets spent while the machine is still working out what it is doing.
Clicks are different. They are cheap and frequent, and a few hundred dollars buys hundreds of them. Optimise for link clicks and the platform gets enough signal to behave predictably — and you can still count signups yourself, on your own page, because that number was never the platform's to give you.
Name the trade honestly: optimising for clicks means the platform goes looking for people who click, not people who buy. That costs you something real when you are acquiring customers. It costs you nothing here, because you are measuring rather than acquiring, and you are going to read the signup rate yourself.
3. Turn off everything that reshuffles the test
In February 2026 Meta merged its manual and Advantage+ campaign flows into one, with AI optimisation switched on by default across audience, placements and budget. Every one of those can still be toggled off. For a message test, one of them has to be.
Audience expansion must go off. Left on, the platform widens beyond the targeting you set, looking for cheaper results elsewhere. That is genuinely useful when you are scaling and fatal when you are testing, because the first of your three questions was can I reach the people I think are my market — and the platform has just answered it by going and finding different people.
Budget optimisation across ad sets must go off, or more simply, run one ad set. If you give the platform two, it will shift spend toward whichever looks cheaper early and decide your experiment for you before you have read it.
Placements can stay automatic. Narrowing them raises your costs, and you are not comparing placements — you are comparing one message against a target audience. Spend your restraint where it buys you something.
One campaign, one ad set, one ad, targeting exactly who you said you were targeting. Everything else is the platform having an opinion about an experiment it does not know you are running.
4. Write the ad and the page as one instrument
The ad and the landing page are not two things. They are one measurement with a join in the middle, and if they say different things you will not be able to tell which half failed.
Use the same promise in both, and where you can, the same words — the ad's main line and the page's headline should be recognisably the same sentence. A visitor who clicks on one claim and lands on another has been given a reason to leave that has nothing to do with your idea.
This is also why the ad should not be clever. Curiosity is very good at buying clicks and very bad at buying information: an ad that makes people click because they cannot tell what it means will fill your page with visitors who never wanted the thing. You will read that as a page problem. It was an ad problem.
State the promise plainly, to the person you named, and let the page finish the sentence.
5. Set a low budget and run one message
A few hundred dollars, about a week. Not because that is optimal, but because it is enough to see a large difference and it is an amount you can afford to lose.
Run one message at a time. Splitting a small budget three ways returns three numbers, each too small to compare with the other two, and you finish with less certainty than you started with. Run one, read it, change the promise on the page, run it again.
Our own test is a fair illustration of the limits here. We ran several angles for the same product at once, and it was only readable because the gap turned out to be enormous — idea-validation messaging brought signups in at around $2 while competitor-analysis and social-media-research framings of the identical product cost north of $10. A fivefold spread survives a small sample. A difference of a few percentage points would have been completely invisible, and we would have drawn a confident conclusion from noise.
What will your budget actually buy?
It opens on three messages. Put your own budget and cost per click in, then set messages to 1 and watch the smallest gap you could trust shrink — that is the whole argument for running them one at a time.
A typical first test
At $1 a click, that is 400 visitors — 133 per message.
13 vs 27 signups
The closest two results this test could tell apart, out of 133 visitors each. Anything nearer — 13 against 20 — is noise at this volume.
That is a landslide, not a preference. A test this size can rule out a message that is far behind, but it cannot pick between two that are close.
Before you turn it on
Then leave it alone. Every edit resets the platform's learning and blends two tests into one number nobody can read. A week is not long.
6. Read the three answers separately
At the end you have three results, not one, and collapsing them into "it worked" or "it didn't" throws away most of what you paid for.
The three answers, read separately
Do they see it and click?
A weak result here means either the audience is wrong or the words are wrong for that audience — change one of them, not both, or you will not know which fixed it. A strong result means your description reaches the people you aimed at. The first link holds.
Do the clicks become email addresses?
Weak means the promise did not survive contact with the page: the page either broke the ad's promise or repeated it with less conviction once a price was attached. This is the most fixable of the three failures. Strong means the promise holds up when someone reads the detail and sees the number.
What do the addresses actually mean?
Very few is the clearest signal this test produces, and it is worth acting on. A healthy number is interest, not a verdict — an address given after seeing the price is worth more, because it is the rung directly below a payment. But it is still not a payment, and nobody has paid you yet.
That last row is the one founders misread. Email signups are not evidence that your idea is good. An address is cheap to give — it costs a stranger nothing to express mild interest in something that does not exist yet. A healthy number means the promise landed on the people you aimed at, which is genuinely useful and is not the same as a business.
The asymmetry is the point. This test says no with far more authority than it says yes. Very few clicks, or clicks that produce almost no signups, is real quantitative evidence that the idea needs rethinking — and it arrived in a week for a few hundred dollars rather than after a year of building. A good result does not give you certainty. It gives you a reason to take the next step, which is what you were buying.
When the numbers won't tell you anything
Ads are the fastest way to get an answer, and there are situations where they will hand you a number that means nothing. None of these is a reason to skip ahead to building — each is a reason to get the answer somewhere else first.
When you cannot name the audience yet. Targeting is the first link in the chain. If you cannot describe who to put the ad in front of, the test starts broken and no amount of budget repairs it. Settle whether that group is real first; it costs nothing.
When your buyer is not reachable by ad targeting. Ten named enterprise accounts, or a market defined by something no platform knows about its users. The answer exists, it is just not for sale — it is in the rooms those people are already in.
When the budget is genuinely too small. Below a certain volume every result is noise, and $50 that produces four clicks will still feel like a verdict when you read it at midnight. Wait and save rather than buy a number you will misinterpret.
Related questions
The page has to exist before the traffic does: what to put on a website built to test an idea. The whole sequence, from defining the promise to reading the result against your own economics, is here. And if the answer comes back yes, the acquisition problem starts again from nothing: how solo founders find their first 100 customers.
Draper runs this whole sequence with you — working out what you are making, sharpening how it is described, building the brand and the page around it, and putting the ads live on the platform your customers are actually on. Try Draper free →
