Guides
How to test a product idea fast: a website in one weekend
Test demand for a product idea before you build it: a landing page, a small ad budget and clear go/pivot/kill rules. Costs, limits, legal basics.
Building a product takes months, but finding out whether people need it can be much faster. Over a weekend you build a simple page with one offer and one button, and over the next week or two you bring a few hundred people to it and watch what they do. By his own estimate, Buffer founder Joel Gascoigne lost about a year and a half on ideas he built without that kind of check.[1]
AI and ready-made templates have sped up page building: it can now take hours rather than weeks. There is one catch: a beautiful page proves nothing on its own. Only the actions of people count, and you decide in advance which actions you will treat as the answer.
Below is a short plan for Friday, Saturday and Sunday, a comparison of ways to build the page, and simple rules for reading the numbers. No technical background is needed.
We sell website templates, but a template isn’t for everyone. Where another route is better, we’ll say so.
Last verified: October 26, 2026. Prices and terms of services change often. This article was checked against the services’ official pages and primary sources on that date, and prices are as of that date. Important claims carry source numbers, and the list of sources is at the end of the page.
Short version: the weekend plan
- Friday evening (1–2 hours). Write down your hypothesis in numbers and a decision rule. For example: “If at least 15 of 300 visitors leave a request after seeing the price, we move on. If 6 or fewer do, we close the idea.” The numbers are illustrative; choose your own. Then state your offer in one sentence: who it’s for, what the pain is, what result it delivers and what it costs.
- Saturday (about a day). Build a page with one button, a price and an honest “not launched yet” note. Connect a form and a counter.
- Sunday (2–3 hours). Bring the first visitors: from ads, communities and personal messages.
- The next 1–2 weeks. Count more than email addresses. Count costlier actions: choosing a plan, booking a call, paying a deposit.
- At the end. Make the decision by the Friday rule: build, change the offer or close the idea.
Launching the test over a weekend is realistic, but the answer comes a little later: people need time to reach the page. And remember: the page measures interest, it doesn’t guarantee success.
What a test page will show
A page shows how strangers behave. That is more reliable than the opinion of friends and colleagues. But not all actions carry the same weight. Buffer’s founder put it this way: picking a plan before leaving an email says more than an email alone, and after that he wrote to the people who signed up and talked to them.[1] Alberto Savoia, the author of the “pretotyping” method, writes in his book that a real signal appears when a person has skin in the game: time, reputation or money.[2]
| What the visitor did | How serious a signal it is |
|---|---|
| Clicked a button | Weak |
| Left an email address | Weak to medium |
| Chose a plan or a price | Medium |
| Booked a call | Strong |
| Paid a deposit | Strongest |
The table is our summary. Well-known examples: Buffer first built a page with a description and a button, then a page with plans, and only after that the product itself. Over seven weeks it collected 120 sign-ups.[1] Instead of a product, Dropbox showed a demo video, and its waiting list, by the founder’s account, grew from 5,000 to 75,000 people overnight.[3] These are the founders’ own accounts; there is no independent verification.
What a page won’t show: whether people will use the product, pay for it again and again, and whether you will be able to build it as promised. That comes later. We found no confirmed stories where an idea was tested with an AI-built page and the demand numbers were published: authors mostly write about speed, for example four hours for a page with Claude Code.[4] You will collect the numbers for your own decision yourself, and that is the point of the test.
The weekend plan
The weekend goes into launching: by Sunday you will have a page and the first visitors. The answer comes later, because bringing people in and waiting for the numbers can’t be done faster. In the accounts we found, it took about a week for Tim Ferriss, who was testing book titles,[5] and two weeks for GrowthMentor.[6] The time by day below is our estimate.
Friday evening: a hypothesis, a rule and an offer
- The hypothesis. Write it in the form “at least X% of people in group Y will take action Z.”[2] For example: “At least 5% of small-shop owners who land on the page will click ‘I want it for $29 a month.’” Action Z should cost the person something: leave a request after seeing the price, book a call, pay a deposit. A like costs nothing.
- The closing rule. Write down the reverse right away: “Of 300 visits, 6 requests or fewer (2%) — we close the idea; 15 or more (5%) — we move on.” The numbers are illustrative: there are no universal thresholds for “the idea is confirmed.” Write them down in advance; otherwise, once you see the result, you will quietly move the threshold in a convenient direction.
- The offer in one sentence. Who it’s for, what the pain is, what the result is, what it costs. For example: “For freelance bookkeepers: we automatically collect client documents into one folder. $29 a month.” It’s better to show the price right away: that way you test both interest and willingness to pay. Buffer also added a pricing page before it built the product.[1]
- Conversations with people, if you have time. Five to ten short conversations with people from your target group will spare you a bad headline and a silly price. Ask about the past, not about your idea.[7] If you run out of time, set up the conversations for the duration of the test.
Saturday: a page with one button
First choose what to build the page with (the comparison is in the next section). The page doesn’t have to look expensive, but it must be clear and honest. For a first try, set aside a whole day. The page needs:
- A headline and a subheadline: the problem or the result for a specific group of people.
- Three “how it works” points and the price.
- One button. Extra options blur the data.
- An honest note: “Not launched yet. Leave your email and we’ll write when it’s ready.” Why this matters is covered in the legal section.
- A form with a clear consent to receive emails.
- One question after the form: “What do you use now?” or “How much would you be willing to pay?”
- An event counter: views, button clicks, form submissions. Plausible (from $9 a month)[8] or PostHog (free up to a million events a month)[9] will do. Add UTM tags to links from ads: they show where each visitor came from.
Use only real reviews and numbers. Check the page on your phone: many visitors from ads will arrive on one.
Sunday: the first visitors
You need visitors from your target group, not just anyone:
- Paid search. At GrowthMentor, €418 of Google ads over two weeks produced 75 sign-ups, at an average of €0.94 per click.[6]
- Testing headlines with ads. Tim Ferriss tested book titles this way: under $200 and about a week.[5]
- Communities and personal messages. Free, but follow each platform’s rules.
Don’t count on cheap clicks. According to LocaliQ, the average cost of a click in search advertising (Google Ads and Microsoft Ads) across all industries in 2026 is $5.42.[10] For $200 that is about 37 clicks (our calculation), so choose narrow queries and add free channels. If few people came over Sunday, that’s normal: the test will continue through the week.
The next 1–2 weeks: count and decide
Count not visitors but actions: clicked the button, left an email, answered the question, chose a plan, booked a call, paid. Then compare the result with the Friday rule and choose: build, change the offer or close the idea. The main thing is not to move the thresholds after you have seen the numbers.
What to build the page with: a site builder, an AI builder or a template
There is no universally best way. Choose based on what will happen to the page after the test.
| No-code site builder | AI builder | Template and an AI agent | |
|---|---|---|---|
| Examples | Carrd, Framer, Webflow | Lovable, Bolt, Base44 | A ready-made Astro site, edits through Claude Code |
| Starting price as of October 26, 2026 | Carrd Pro Standard: $19 a year [11]; Framer: from $10 a month billed yearly [12]; Webflow: from $15 a month billed yearly [13] | Lovable Pro: $25 a month [14]; Bolt Pro: $18 a month billed annually [15]; Base44 Starter: $16 a month billed annually [16] | Template: $79–$149 one-time, and an AI subscription (Claude Pro: $17 a month billed annually or $20 monthly [17]). Hosting can be free [18] |
| Main risk | Lock-in to the platform, changing prices | Pages that all look alike, security of forms and databases [19] | You need a terminal; you connect the form and the counter yourself |
| Skills needed | Low | Low | Higher |
Each cell says when a price assumes yearly billing; paying monthly usually costs more. Prices change often (Webflow raised its Basic plan in May 2026), so check the pricing pages before buying.
When to choose what
- You need a page by evening, and a terminal is unfamiliar. Take Carrd: its Pro Standard plan is the cheapest paid option we checked, but it is built for a single page.
- You need a prototype with user sign-in or a database. An AI builder will do, but check security separately.
- The page will later become a real site, and you will often change it through AI. A template fits here: the code stays with you and the design is already done. If you are doing this for the first time and in a hurry, it is simpler to start with a builder.
Where AI helps and where it doesn’t
AI speeds up work on the page, but the decisions your test result depends on stay with you.
- It helps. Draft the page, suggest five headline options from notes of customer conversations, make targeted edits, find template leftovers on the page (other people’s names, placeholders), group survey answers. Ask it to change only what’s needed, for example: “Change only the headline, subheadline and button text; don’t touch the layout.”
- It doesn’t help. Choosing who you’re offering what to. Replacing customer conversations: an AI answer to the question “is this a good idea?” is not data about your buyers. Bringing people in and making the decision from the numbers. Checking the legal side.
- Be careful with people’s data. Veracode, a company that sells code-checking tools, has been testing code from AI models since 2025. In its July 2026 report, generated code failed security checks in nearly 44% of cases, about the same as a year earlier.[19] For a page with no sign-in and no database the risk is lower, but check forms and access keys before launch.
How to tell whether the idea is working
You need a rule written in advance: which action you count as the answer, how many visitors you need, and where the threshold is.
Benchmarks from sources are not your target. The median landing-page conversion rate is 6.6%, and for SaaS it is 3.8% (Unbounce report, 2024 edition, based on more than 41,000 landing pages). The company sells a page builder, and its methodology doesn’t say exactly how a conversion is counted.[20] Only the same actions from the same traffic source can be compared: a sign-up from an ad and a call after an email to acquaintances are different things.
Small numbers deceive. If 5 of 100 visitors left a request, that is 5%, but the true value could be 2% or 11% (our calculation: a 95% Wilson interval). For 15 requests out of 300 the range is 3.1–8.1%, and for 6 out of 300 it is 0.9–4.3%. That is why, in the Friday example, 15 out of 300 is confidently above the closing threshold (2%), and 6 out of 300 is confidently below the “move on” threshold (5%). But 5 out of 100 decides nothing. If you get 7 to 14 requests out of 300, that is a gray zone: you need more visitors or a different offer.
Comparing two versions of a page is even harder: to tell a 2% conversion rate from a 5% one, you need about 590 visitors for each (our calculation at a 5% significance level and 80% power). So one page is enough for a first test.
The decision is one of three options (this is our framework, not a standard):
- Build. The thresholds were reached on costly actions (payment, pre-order, call), not only by email, and the same pain keeps coming up in conversations.
- Change the offer. There is a response from another group of people or to a different offer.
- Close or postpone. After the agreed number of visits (for example, 300), nobody picks a paid plan or books a call.
The enthusiasm of friends and relatives, likes, free sign-ups and the words “I’d definitely buy this” do not count as an answer to the question “is the idea needed?”
Legal minimum: honesty and consent
A test page collects people’s data, and sometimes money, so ordinary rules apply to it. This is not legal advice: laws depend on the country, and we have not examined the rules for digital products and subscriptions.
- An honest status. If the product doesn’t exist yet, say so plainly: “Not launched yet.” Don’t invent reviews. A “buy” button that leads to an empty page (this kind of test is called a “fake door”) is honest if the person is told after the click that the product isn’t available yet, and no money is taken. This is our conclusion, not a legal rule.
- Prepayment. Taking money before the product is ready is possible, but it comes with obligations. In the US, the FTC rule for online orders of goods requires honest delivery times and a choice between waiting and a refund; for digital products, check separately.[21] In the EU, a buyer usually has 14 days to withdraw, with exceptions.[22] Take payments through a payment service: Stripe recommends having enough cash on hand to cover refunds and disputes.[23] On Cloudflare’s free plan, you can’t collect credit card details on your own site (clause 2.2.1(h)).[24]
- Mailing. In the EU you need clear consent, with no pre-ticked boxes.[25] In the UK, marketing emails to individuals usually require consent, unless the “soft opt-in” conditions are met.[26] In the US, the CAN-SPAM Act requires honest subject lines, a valid physical postal address and a working unsubscribe.[27]
Three mistakes that are easy to avoid
- Taking an email address as demand. People leave it out of curiosity and simply because it’s free. Show the price, add a question after the form and an invitation to a call, and count the people who reached a costly action.
- Moving the threshold once the numbers are known. You wanted 5%, you got 4%, and suddenly “that’s not bad either.” Write down the threshold and the check date on Friday and show the note to someone else.
- Trusting nice percentages on small traffic. 2 requests out of 20 is 10%, but the true value could be as low as 3% or as high as 30% (our calculation). Decide in advance how many visitors you are waiting for, and don’t stop the test on the day a figure pleases you.
Summary
A page built over a weekend doesn’t prove success, but it lets you get an answer sooner than you would after spending months on a product. The main points:
- Over the weekend you launch the test, and the answer comes over the next 1–2 weeks.
- Write down the hypothesis and thresholds before launch.
- Count costly actions, not just email addresses.
- Don’t trust small samples: 5 requests out of 100 decide nothing.
- Be honest with people: a “not launched yet” note, consent to emails, payment through a payment service.
Want to skip the setup?
If you choose the ready-made template route: ShipThemes templates are standard Astro projects with every page designed, so you can cut one down to a single offer page with Claude Code or Cursor and publish it on any static host. You connect the form and the event counter yourself, as described above. Browse the template catalog. And if you need a page by evening without a terminal, take Carrd: for that case it fits better.
We covered how to publish the page in How to publish an Astro site on Cloudflare. How to avoid the generic “AI look” is in How to avoid AI slop on a website built with AI. How to change a template with AI is in How to customize an Astro template with Claude Code.
Frequently asked questions
How much does it cost to test an idea?
From nothing to a few hundred dollars. In the published examples, GrowthMentor spent €418 on ads[6] and Tim Ferriss tested book titles for less than $200.[5] You can build a page on Carrd for $19 a year,[11] and a static site can be hosted on Cloudflare for free.[18]
Can you test an idea in a weekend?
You can launch the test: one author built a page with Claude Code in about four hours.[4] The answer comes later, though: attracting people and counting takes from one to two weeks.[5, 6]
How many visitors or requests do you need?
There is no single number. A rough guide is a few hundred visitors: 15 requests out of 300 visits is 5%, and the true value lies between 3.1% and 8.1% (our calculation). On smaller numbers, conclusions are shaky.
Is one page enough to test an idea?
No. A page tests interest. To test willingness to pay, you need a pre-order, or at least a choice of plan and a conversation with the person.[1]
Can you take pre-orders before the product exists?
Often yes, but with obligations: honest timelines, a choice between waiting and a refund, and the right to withdraw.[21, 22] More in the legal minimum section. This is not legal advice.
What should you choose: a site builder, AI or a template?
There is no single answer. Carrd and similar builders are for speed and no code. An AI builder is for a prototype with logic, but with a security check. A template with an AI agent is for when the page will later become a real site and the code should stay with you.
Sources
Pages checked on October 26, 2026.
- Joel Gascoigne (Buffer), “How to Successfully Validate Your Idea with a Landing Page MVP,” Medium, November 30, 2013; Buffer, “Idea to Paying Customers in 7 Weeks: How We Did It,” February 16, 2011 (the founder’s own accounts)
- Alberto Savoia, The Right It (HarperOne, 2019); Pretotyping.org: methodology
- Eric Ries, “How DropBox Started As A Minimal Viable Product,” TechCrunch, October 19, 2011 (quoting founder Drew Houston)
- Sebastian Morino, The Roadmap, June 28, 2026 (the author’s own account)
- Boing Boing, how Tim Ferriss tested book titles with ads, October 25, 2010
- Foti Panagiotakopoulos (founder of GrowthMentor), “How to validate a startup idea using a pre-launch landing page and less than $500 in PPC spend,” April 22, 2020 (the author’s report on his own experiment)
- Michael Lynch, summary of the book “The Mom Test”
- Plausible, pricing
- PostHog, pricing
- LocaliQ, Search advertising benchmarks (updated June 1, 2026) (an advertising company’s data)
- Carrd, Pro plan pricing
- Framer, pricing
- Webflow, pricing
- Lovable Docs, Subscription plans
- Bolt, pricing
- Base44, pricing
- Claude, pricing
- Cloudflare Docs, Static assets: billing and limitations
- Veracode, 2026 GenAI Code Security Report press release (July 28, 2026); 2025 study (data from a company that sells code-checking tools)
- Unbounce, SaaS conversion benchmark (2024 edition) and methodology (data from a company that sells a page builder)
- FTC, Business guide to the Mail, Internet or Telephone Order Merchandise Rule
- Your Europe, returning goods bought at a distance
- Stripe Support, accepting payments for pre-orders
- Cloudflare, Self-Serve Subscription Agreement (updated September 12, 2025), clause 2.2.1(h)
- GDPR.eu, consent requirements
- ICO (UK), PECR rules for electronic mail marketing
- FTC, CAN-SPAM Act: compliance guide for business