Summarize with AI
A fake door test puts a button, menu item or pricing card for a feature you have not built onto your live site, then counts how many people click it. Whoever clicks lands on a short page that says plainly the feature does not exist yet and asks one question about what they need. It tells you in a couple of weeks whether strangers want the thing, before your team spends a quarter building it.
This matters most to a small AI team, because the founder is usually the one most sure a feature is wanted. You know your product and nobody else does, so the features you can picture most clearly are the ones you would use yourself. A fake door swaps your certainty for the behaviour of people who have never met you. It is also not the same job as comparing two versions of a page, which our post on testing a landing page with low traffic covers. Here you are testing demand for something that is not there.
A Team card that read Coming soon
Picture an AI tool that cuts recorded webinars into short clips for B2B marketers. Upload an hour of video, and it returns twelve captioned clips sized for LinkedIn, each with a suggested post. The pricing page has two cards, Free and Pro. The founder and his cofounder are debating Team workspaces: a shared clip library, comments on each clip, and one brand kit for the whole team. The estimate is six weeks for two engineers, roughly $40,000 of salary and most of a quarter.
The case for building it rests on demo calls. The kind of line a marketer says on those calls goes, a team plan would be great, and the founder has heard some version of it maybe five times. Each time he writes it in the call notes. His cofounder has already broken the feature into tickets.
So instead of starting, they add a third card to the pricing page. It reads Team, with a grey tag that says Coming soon, three lines on what it would include, and a button labelled Join the Team list. Over two weeks the pricing page gets 2,300 visits. The Pro button gets 161 clicks. The Team button gets 11. Four of those eleven people answer the question on the next page and leave a work email. Nobody writes the first ticket.
Where the door goes: nav, pricing card or inside the product
Where you put the door decides whose answer you get, so pick the place before you design the button.
A nav item collects curious visitors. People click menu labels to find out what a company is, so a Teams link in the header will get clicks from people who want to learn, not buy. Treat nav clicks as interest at most.
A pricing card collects people already thinking about money. Pricing visitors are comparing plans, so a click on a Team card sits next to a real choice: the Pro button one card to the left. That gives you a baseline on the same page. Sites that keep Pricing in the main nav, like the one we designed for SATHI, an affiliate platform for Shopify brands, already send buyers to that page, which makes it the natural place for a door about a paid plan.
A button inside the product collects people who already use it. If the clip tool added a Share with team button on its export screen, the clicks would come from existing users at the exact moment they wanted the feature. That is the strongest signal of the three, and the one that fits features rather than plans.
Name the door in the buyer's words, not yours
Before the pricing card went up, the founder's draft label was Workspaces. That is the word the team uses in standup. A marketer reading the pricing page has no idea what a workspace is in this product, so a low click count on Workspaces would only prove that strangers do not know the word.
This is the most common way a fake door gives a false no. The test measures the label as much as the demand. Write the button in the words a buyer would type into a search box or say on a call: share clips with your team, one brand kit for everyone. If you cannot describe the feature in words a stranger understands, the door will not get a fair test, and the eventual feature page will have the same problem.
The page after the click says it is not built
Most fake door advice spends its time on the button. The page after the click is where the test is either honest or not. Some teams show a spinner and an error, or a checkout that fails at the last step, so the visitor does not notice the trick. That is lying to the people most interested in your product, and it ruins the data too, because you cannot tell who gave up from who would have paid.
The honest version says it first. For the clip tool, the page after Join the Team list would read like this:
- Heading: Team workspaces are not built yet.
- Line one: We are deciding whether to build a shared library, comments and one brand kit for teams. Your answer decides it.
- One question: How many people on your team would open these clips? Options: just me, 2 to 5, 6 or more.
- One field: work email, with a checkbox to hear when it ships or when the team decides not to build it.
- A way back: a button that returns them to the Pro card, so the visit is not wasted.
Here is the claim worth arguing about. The people who still answer after reading not built yet are a better signal than the click. The click is cheap and partly curiosity. Choosing 6 or more and typing a work email after being told the feature does not exist is a small act of wanting it. Count that.
What counts as a signal, written down before launch
Write the decision rule before the door goes live, and share it with your cofounder. If you set the bar after the numbers arrive, you will set it wherever the numbers landed.
For the clip tool, a fair rule reads: build Team workspaces if at least 2 percent of pricing visitors click the Team button and at least ten of them leave a work email with a team of 2 or more. Kill it if fewer than one in two hundred visitors click. Anything in between earns five calls with the people who answered, before any code.
The result, 11 clicks from 2,300 visits, is under one in two hundred. Four answers, all from teams of two to five, is under the bar. The comparison with Pro matters too. About fifteen times as many people clicked the plan that exists. On the evidence, the people who want a team plan are mostly the people on demo calls, which means the line on those calls was politeness, or a need the founder could meet with a shared export link in an afternoon.
Low traffic does not rule a fake door out. You are measuring a big yes or no, not a small lift between two versions, so a few thousand visits can answer it. With a few hundred, run it longer, or put the door inside the product where every click comes from a user.
Eleven clicks is the good outcome
A team that sees eleven clicks often reads it as a failed test and goes back to planning the build. It is the best result they could have had. They paid two weeks of a pricing card instead of six weeks of two engineers. If you are deciding what goes into a first version, the same thinking applies to what to cut from your MVP: a feature nobody asks for, unprompted, is a feature to cut.
The cost of skipping the test is time, and time is what an early team has least of. Across all sectors, the median gap from seed to Series A is around 616 days, and roughly 70 percent of seed companies never raise a Series A. A quarter spent on a feature four people asked for is a real share of that window, spent on the wrong thing.
Ethics are part of the cost too. A fake door left up for months turns into a broken promise, and the people who signed up remember. Take the card down when the test ends. Email everyone who answered with the decision, including a no, and say what you are building instead. Never take a card number or a payment for something that does not exist.
Run one this week
- Pick one feature your team is about to build and write its cost in weeks of engineering.
- Choose the door: a pricing card for a new plan, a button inside the product for a feature, a nav item only for interest.
- Write the label in the words a buyer would use, and show it to one person outside the company. Ask what they think it does.
- Build the page after the click: not built yet in the heading, one question, one email field, a way back.
- Write the decision rule with numbers: the click rate that means build, the rate that means kill, and what happens in between.
- Set up two events in your analytics: door clicked and question answered. Note the clicks on the nearest real button as a baseline.
- Run it for two weeks, or until a few thousand visitors have seen the door. Then take it down and email every person who answered.
When the answer is yes, the feature still needs a page a stranger can read. If you would rather hire that work out, our ranking of landing page design agencies for B2B AI SaaS companies compares studios on public evidence.
A fake door is a few small pieces of page design: one label in the buyer's words, one card that sits beside a real choice, and one honest page behind it. Studio Maydit designs and builds pages like these in Framer, Webflow and custom code, on a fixed scope of three to four weeks, so the pricing page and the pages around it can change quickly when a test says they should. If you have a feature on the roadmap and want a second read on where its door should go and what the page after the click should say, book a 30 minute call.
