AI7 min

Predictive UX: the death of the blank slate in modern apps

Predictive UX explained: smart defaults and suggestions that replace blank screens, when predictions backfire, and how to keep users in control of the guesses.

Saurabh YadavFounder, Design Studio by Saurabh
AIPredictive UXAnticipatory designSmart defaults UXProactive UX design
Share

There's a particular kind of tiredness that comes from a blank screen. An empty search box. A form with twelve fields. A new document with a blinking cursor. The app is waiting for you to tell it everything, as if it's never met you before.

Predictive UX is the opposite. The app makes a sensible guess about what you need, offers it, and lets you correct it. Your usual delivery address is already filled in. The calendar suggests a meeting time that works for everyone. The banking app shows the payee you pay every month at the top.

Done well, it feels like the product knows you. Done badly, it feels creepy, bossy or just wrong. Here's how to design predictive features that help, and how to know when to leave the blank slate alone.

What is predictive UX?

Predictive UX (also called anticipatory design) is designing an interface that uses what it knows, about the user, their past behaviour, the time, the place or similar users, to prepare the next step before they ask.

It's not new. Autocomplete, recently used files and "frequently bought together" have been around for years. What's changed is that AI models make predictions far more flexible. Instead of only suggesting a past value, an app can now draft a reply, pre-fill a whole form from a photo, or notice something unusual in your data and flag it.

The design questions stay the same, though: how confident is the guess, how much does a wrong guess cost, and how easily can the user say no?

Levels of prediction, from gentle to bold

It helps to think of predictive features as a ladder. Each rung saves more effort and asks for more trust.

Level What the app does Example Trust needed
1. Smart defaults Pre-selects the most likely option Last-used shipping address already selected Low
2. Suggestions Offers options, user picks Suggested replies, recent searches, likely payees Low to medium
3. Pre-filled drafts Prepares a full draft for review An expense report filled in from a receipt photo Medium
4. Proactive nudges Speaks up without being asked "Your card bill is due in two days" Medium to high
5. Automatic actions Acts on the user's behalf Auto-moving savings each week High

Most products should live on rungs one to three. Rung five needs explicit opt-in and a very clear way to stop it.

Tip

Start every predictive feature at the lowest rung that solves the problem. You can always move up once users show they trust it. Moving back down after a bad experience is much harder.

Where predictive UX helps most

Prediction is most valuable when the task is repetitive, the likely answer is clear and a wrong guess is cheap to fix.

  • Forms. Pre-fill what you already know. Every field you fill in for the user is one less chance to drop off. Ask only for what you can't work out yourself.
  • Search. Show recent and popular searches before the user types.
  • Repeat tasks. Reorders, regular payments, weekly reports.
  • Stressful moments. When users are anxious or rushed, removing even one decision matters.

From my work

On Pukaar.ai, the homepage reads a baby's sleep and feeds against age norms, so a parent doesn't have to work out whether something is normal at 2 a.m. And one of the things I'd design next is to start listening for a cry right away with a live waveform, instead of making a parent wait through a 10-second setup. That's predictive UX at its simplest: assume what they came to do, and get out of the way.

When prediction backfires

Here's where it goes wrong, and I've seen versions of each of these in products I've used.

Wrong guesses that cost a lot. Pre-selecting the wrong bank account for a transfer isn't a small mistake. For money, health or anything irreversible, suggest rather than pre-select, and always show a review step.

Creepiness. If the app reveals it knows something the user didn't expect it to know, trust drops fast. "We noticed you've been searching for pregnancy products" is a famous kind of failure. Ask yourself: would this user be surprised we know this?

Filter bubbles. Predicting from the past keeps showing users more of the same. Sometimes people want to discover something new, and a feed that only reflects their history gets boring or narrow.

Being bossy. Microsoft's Clippy is the classic example: help nobody asked for, offered at the wrong time. Proactive features need a high bar for interrupting.

Watch out

Don't use prediction to push users towards what's good for the business instead of what's good for them, like pre-ticking an add-on because "most people choose it". That's a dark pattern dressed up as personalisation. I've written more about this in designing ethically.

Design rules for predictive features

These are the rules I'd apply to any predictive feature, AI-powered or not.

  1. Make accepting and rejecting equally easy. One tap to accept, one tap to dismiss. Never bury the "no".
  2. Explain the guess, briefly. "Suggested because you pay this every month" turns a mystery into something users understand.
  3. Never auto-commit anything important. Prefill, then review, then confirm.
  4. Learn from rejection. If a user dismisses the same suggestion three times, stop showing it.
  5. Give an off switch. For proactive nudges and automatic actions, a clear setting to turn them off.
  6. Respect privacy. Only use data the user would expect you to use, and say what you use.

The cold start problem: what to show on day one

Predictive UX needs data, and new users don't have any. That's where the blank slate sneaks back in.

A few ways to handle it:

  • Ask one or two smart questions during onboarding instead of a long setup.
  • Use sensible defaults based on what most users do, and say so ("Popular with new users").
  • Show examples or templates instead of an empty screen.
  • Let predictions improve visibly over the first few uses.

This overlaps a lot with good empty states. My post on empty states and error messages shows how to make "nothing here yet" useful.

A predictive UX checklist

  • The task is repetitive and the likely answer is clear.
  • The prediction starts at the lowest useful rung.
  • Accepting and dismissing take the same effort.
  • Each suggestion has a short "why".
  • Nothing important happens without review and confirmation.
  • Repeated rejections reduce or stop the suggestion.
  • Users can turn proactive features off.
  • New users get helpful defaults, not a blank screen.

Work with us

Building an AI feature? Our AI product design agency designs copilots, agents and AI workflows that users trust.

FAQ

What is the difference between predictive UX and personalisation?

Personalisation changes what a user sees based on who they are, like recommended content. Predictive UX goes a step further and prepares the user's next action, like pre-filling a form or drafting a reply. Both rely on user data, so both need transparency and control.

Is predictive UX always powered by AI?

No. Many predictive patterns are simple rules: remember the last choice, show recent items, pre-select the most popular option. AI makes predictions more flexible, like drafting text or reading a photo, but a lot of the value comes from basic smart defaults that any team can ship.

How do I know if a predictive feature is working?

Track how often suggestions are accepted, edited or dismissed, and whether the task gets faster. A high dismissal rate means the guesses are off. Also watch for undo actions and complaints, which signal that predictions are crossing into unwanted territory.

Key takeaways

  • Predictive UX replaces blank screens with sensible guesses the user can accept or change.
  • Think in rungs, from smart defaults to automatic actions, and start low.
  • Predict freely where mistakes are cheap; suggest and confirm where they're costly.
  • Explain each guess briefly, and make dismissing as easy as accepting.
  • Plan for new users with good defaults and templates instead of empty screens.

What to do next

Predictive UX works best alongside interfaces that change shape as they learn. My post on adaptive interfaces covers how to do that without confusing people. For a real example of designing around a stressful moment, see the Pukaar.ai case study.

Want your product to feel like it already knows what users need? Tell me about your project and let's find the blank screens worth replacing.

Saurabh Yadav's Memoji

Written by

Saurabh Yadav

Founder of Design Studio by Saurabh, a UI/UX design studio in India. We design apps and websites that feel effortless, from research to shipped pixels. Most recently: Pukaar.ai, as seen on Shark Tank India, and the Demerara Bank app.

Keep reading

All articles →