We review a lot of AI product intake flows. Most have the same pathology: too much onboarding, too little proof.
Here's the pattern that actually converts first-session users into second-session users.
Three questions, three clicks
The first minute of an AI product has exactly one job: reduce the user's uncertainty about whether this will work for them.
That breaks down into three questions the user is silently asking:
- "What is this for?" — can I confirm the domain in one screen?
- "What do I have to give it?" — what's the minimum input?
- "What does it give me back?" — what's the first visible output?
If your onboarding takes longer than three clicks to answer all three, you're losing users faster than your model is impressing them.
The anti-pattern: the feature tour
New user signs up. You greet them. You show them the left sidebar. "This is where your projects live." You show them the top bar. "This is where you switch models." You show them the keyboard shortcuts.
By the time the user tries their first prompt, they've been told about eighteen features and produced zero outputs. The product hasn't done anything for them yet. They're mentally comparing it to ChatGPT, which took one prompt to prove itself.
The feature tour teaches the product. The user came to be helped by the product.
The pattern that works
Click 1: one input field, one clear purpose. "Paste your CRM export" or "Describe the email you want written" or "Upload the PDF you need summarised." No model picker. No advanced options. No "skip this step." Just the thing the user came to do.
Click 2: the output. Real output. Not a preview, not a sample, not "here's what our model can do." The user's actual input has been processed; their actual result is on screen.
Click 3: one next action. Most users will either want to iterate ("try again with different instructions") or export ("send this to my team"). Pick one. Surface it prominently. Everything else — the sidebar, the settings, the templates — can live in the background until session two.
Where founders get this wrong
The usual failure mode isn't bad design — it's bad prioritisation. The product team wants to show off the model's capabilities, because they know how hard the model was to build.
But new users don't care how hard the model was to build. They care whether the model just saved them twenty minutes on an email draft. If it did, they come back. If it didn't, they churn before they ever see the features the team was so proud of.
The three-click rule forces a cold question: what is the shortest possible path from sign-up to "oh, that worked"? That path is the onboarding. Everything else is the product.
A useful diagnostic
Grab a timer. Have a friend who hasn't seen your product sign up and try their first real task. Watch silently. Stop the timer when they produce one output they consider useful.
Under 60 seconds: you have a funnel. 60-180 seconds: you have a product that needs conversion design work. Over 180 seconds: you don't have an onboarding — you have a tutorial, and most users don't finish tutorials.
The strongest AI products we see in the subscription queue are treating onboarding as the first product surface, not a side quest. Design it like the homepage — because, for new users, it is.