Lesson 11 of 12 · 2 min read

What makes onboarding in an AI app feel personal?

Asking only what the AI will use, and showing it being used. A few questions about someone’s real constraints, like budget, time, and equipment, then the result built in front of them, turns a form into the first moment the product works for that person.

Diagram: three answers, budget, time, and equipment, shape a first meal plan that is ready right away.
Budget, time, and equipment, then a plan. It never asked for a job title.

Five questions about a life, not a profile

In a LinkedIn post, the designer Josim Uddin shows the first run of Nutry, an AI nutrition app: five short steps instead of a long form. Then the app builds the meal plan while the person watches.

Nutry’s five steps, and my guess at what each answer changes
  1. Body statshow much the plan has to provide
  2. Target pacehow hard the plan pushes
  3. Budgetwhich foods it can choose
  4. Cooking timewhich meals fit a weeknight
  5. Kitchen toolswhich recipes are possible at all
  6. The planbuilt while the person watches

Three of the five are about a life rather than a body: money, evenings, and what’s in the kitchen. Those are the answers a generic plan gets wrong.

Every question has to earn its place

Lucy Learns, my dog-training app, turns the dog trainer’s handout into today’s session, so the first screen already fits this dog.

Three Lucy Learns screens: the splash with a black Labrador and her person, today’s session, and a cleared level.
From my work Lucy Learns. The opening screen, today’s session, and a cleared level. Read how I built Lucy Learns
A question that personalizes, and one that only collects

Collects

  • Asked because the form has a field
  • Nothing visible changes
  • Feels like paperwork

Personalizes

  • Asked because the answer is used
  • The result changes because of it
  • Feels like being listened to

Show the work at the end

I think the screen where Nutry builds the plan does the most work: it shows the answers going somewhere. Whether the plan fits is a separate test, and Josim’s post reports none.

If an answer doesn’t change what the person sees, don’t ask the question.

Check yourself

Which onboarding question in your product changes nothing the person sees?

Cut it, or move it to the moment its answer is actually used. Asked there, it explains itself.

How does a new person know their answers were used?

If the result appears with no sign of the answers in it, the questions felt like a toll. Show the connection.

The idea: Josim Uddin, onboarding for Nutry, an AI nutrition app. The example: How I built Lucy Learns, on this site.