Should Sign-Up Come Before or After the Paywall?

Somewhere in your onboarding there is a screen asking for an email and a password. You have probably moved it at least once, because every article says something different and none of them show you a real flow.

It is one of the highest-friction things you can ask of anybody. Type an email, invent a password, go and dig out a verification mail. All before the app has done a single thing for them.

So I got tired of the advice and counted instead. Thirty captured onboarding flows, one question: where does the account go?

And here is the thing I had wrong, in print, on this page, for months. I used to write the rule as “first screen or after the paywall, never in between”. Ten apps put sign-up exactly in the middle, deliberately, and several of them do it well. If you moved yours out of the middle because an article told you to, nobody had checked.

There are three patterns, not two. I will call the sign-up screen a gate, because that is what it does: nothing happens behind it until the user makes an account.

What follows is all three patterns with the apps that run each one, the exact wording that makes the middle one work, and the single thing that makes it fail. On the data, I will be honest: there is none, and I will show you where I looked for it.

The energy model, and where it stops helping

Here is how I think about it, and this is a model rather than a measurement. Treat the user’s willingness to engage as a battery: they download the app fully charged, and every action you ask for drains it.

Sign-up drains a lot, because it is pure friction with no immediate reward, and when it goes wrong, a password rule or a verification email that never arrives, the session ends there.

The model is useful right up until it hardens into the rule I used to write. So here is what the flows actually show instead.

Pattern 1 · Gate at the door

The account is the first thing. Nothing happens until it exists. This is the majority pattern in the captured flows, and the biggest apps in the library all run it. Strava takes the account before the sport questions. Notion, Headspace and Pinterest all take it within the first few screens, before any personalisation.

Inside the pattern the interesting variation is who the screen is built for. SoundCloud puts Create an account above Log in, and gives it the heavier styling, so the screen is built for somebody who has never been there. Raycast inverts it and puts Log in first, which is the right call once your install base is bigger than your weekly new-user count.

Revolut is the variant worth stealing: a swipeable set of product cards runs above the buttons, so the gate is still first in the flow but not first in the eye.

The rest of the gallery gives up on the wording. Waking Up asks with “Welcome to Waking Up. Create an account to get started.” Suno with “Create a free Suno account to begin”. Both accurate. Both worth nothing to the person reading them.

Zocdoc is the one screen in the ten that answers the question the user is actually asking: “Sign up — It’s free, secure and easy. We will help you keep track of your health by managing your appointments.” One sentence, and the account has a purpose.

And one to be careful with. MyFitnessPal dresses the gate as the first question: the headline is “Welcome! Let’s customize MyFitnessPal for your goals” and the primary Continue button is, in fact, the account step. That works when the customisation it promises is genuinely what happens next.

When the next screen is more account setup, users who tapped Continue expecting a question and got a password field start reading every later framing as marketing, which is expensive across a twenty-screen flow.

Best for

Apps where the account is the product (social, collaborative, anything syncing across devices), apps with enough brand recognition that the ask is not presumptuous, and paid acquisition where the user already knows what they downloaded.

Pattern 2 · Gate at the save point

This is the position I used to tell people never to use. The questions come first, and the account arrives framed as protecting what the user just built. Ten apps in the library do it: Duolingo, Calm, Cal AI, Transit, Finimize, BitePal, Ahead, Recime, Vocabulary and QUITTR. It is not a mistake. It is a pattern with a failure mode.

Most of them use loss framing, and you can see the whole spectrum in the wording:

“Sign up to save progress”  ·  Ahead
“Create an account to save your progress”  ·  Calm
“Now let’s create account — Save your progress & reach your goals”  ·  BitePal
“Save your favourites, stats & settings · WHAT’S A TRANSIT ACCOUNT?”  ·  Transit
“Keep your data safe — Create an account so you never lose favorites, collections, and settings when you reinstall or switch devices”  ·  Vocabulary

Transit and Vocabulary are doing more work than Calm is. They name the specific things an account saves instead of using the word progress, and Transit adds a link explaining what a Transit account even is. Naming beats abstracting: favourites, stats and settings are easier to want than “your data”.

Two of the ten skip loss framing altogether, and they are the interesting ones. Finimize uses the name it already collected, “Sam, finish your setup”, so the account reads as an unfinished task rather than a new demand. QUITTR trades loss for crowd: “Become a QUITTR — Join over 1,000,000 users!”, with Skip For Now underneath.

Both work for the same reason the loss lines work when they work, they point at something that has already happened.

The failure mode is the sentence being untrue. Save your progress after two taps invites the obvious reply, what progress?, and the loss framing curdles into a sales line. A first-run Transit user with no favourites is being sold the preservation of nothing.

Duolingo is the exception in the whole library: it asks what you want to learn, gives you a first lesson, and only then wants a profile. The decline is offered in the same breath. Cal AI does the same thing in the user’s own words, “Save your progress · Would you like to sign in later? Skip”.

That is the version of this pattern that works, and the whole thing turns on the decline being remembered. A LATER that reappears on the next open is not a decline, it is a nag.

Best for

Apps where the onboarding produces something. If the user answered ten questions, built a plan, generated a story or customised a character, the ask has a subject. If your onboarding is three taps, this pattern has nothing to protect and you are better off at the door.

Pattern 3 · No gate at all

The option the old version of this article did not mention. Guest mode, a skip, or an account created silently and claimed later.

And here is the reason it travels badly. Eight of the ten examples in the library are commerce apps: Klarna, Walmart, Target, Best Buy, Etsy, On, 7-Eleven and Bolt Food. In all of them, a purchase can complete without anybody knowing who you are. Read the category before you borrow the pattern.

The best of them argue instead of blocking. Best Buy lists five ticked account perks and then offers Continue as guest anyway, unhidden. Walmart compresses the same case into one sentence, “Sign in to track orders and earn Walmart Cash, check in for faster pickup and get personalized shopping recommendations”, with Continue as guest under it.

Klarna is the most generous version in the gallery, because it names the state the user is actually in rather than the benefits they are missing: “In a hurry? No problem, you can register later.” All three accept that they might lose the argument. That is a checkout convention, and it does not transfer to an app that stores personal state from minute one.

The one that does transfer is Nibble.

Nobody is ever blocked, and the later ask is about protecting something concrete. The trade is real though: if the user changes device before registering, everything is gone, and you have to explain the loss of a state you never asked them to protect. That is the price of the pattern, not a bug in it.

Whichever pattern you land on, word the exit deliberately. The skips transcribed across the library run from the bare, Bolt Food’s Skip, through Finimize’s Skip this step and Numo’s Skip onboarding, to Udemy’s Save & exit.

Udemy’s is the only one of the four that tells the user nothing is being thrown away, which is the fear that makes people tap the gate instead of the exit.

What about “after the paywall”?

It is a genuine fourth option and it is the one I have shipped. It is also, in thirty captured flows, one that nobody in the library does. Not one of the three patterns is “ask after payment”.

The captured flows lean the other way, and Spark Mail is the clearest example. The account already exists by the time the paywall appears, and the paywall counts it as step one of the trial.

Deepstash and Vocabulary do the same trick on their own trial timelines, Vocabulary strikes through Install the app as step one, Deepstash uses the profile you just created. If you are going to charge for the gate, at least spend it twice.

Asking after payment has one real advantage, the user reaches the paywall with nothing spent, and one real cost: you need anonymous sessions. Everything created during onboarding has to persist on the device and attach to an account later. That is engineering work, and it is why the pattern is rare.

What about the data?

There isn’t any. RevenueCat’s State of Subscription Apps 2026, which covers more than 115,000 apps, reports nothing on sign-up placement, not a conversion delta, not a drop-off rate, nothing. Neither does my own teardown library, which records design decisions and their stated logic, not outcomes.

An earlier version of this article told you that moving sign-up out of the middle of the funnel improves paywall conversion. I could not source that, so it is gone.

So if somebody quotes you a percentage for what a sign-up gate costs, ask them which dataset. There is no public one. And yes, I know that asking makes me the difficult person in the thread. Ask anyway.

What you can do is test it on your own traffic. Run one placement, then the other, for long enough to clear a weekly cycle, and compare trial starts and Day 1 retention, meaning how many people come back the next day. Not trial starts on their own, which a softer gate will always win.