← All guides

AI + UX · 5 min read

AI loading states that don’t panic people

When AI takes time, show progress, partial results, and a way to stop — not a frozen spinner forever.

Explained like for kids

Waiting with no update feels like the app fell asleep. Good waiting is like a progress bar on a download: you see it’s working, what’s happening, and you can cancel.

Examples

Each example has a kid-friendly “why,” then a clear Don't and Do with a real UI mock so you can see the difference.

Frozen spinner vs streaming

Like for kids: A spinning wheel with silence is scary. Watching words appear is like watching someone write — you know they’re still there.

Don’t

Blank screen + endless spinner

People refresh, duplicate requests, or quit.

Wrong — endless spinner

Thinking…

Do

Stream tokens / show partial draft

Perceived speed goes up. Trust that work is happening.

Right — streaming draft

Here’s a clearer empty state: “No projects yet.

Words appear as they’re generated

Name the waiting step

Like for kids: “Working…” is vague. “Looking in your PDF…” tells you the helper is busy with the right job.

Don’t

Generic “Thinking…” for 40 seconds

No signal if it’s stuck or still useful.

Wrong — vague waiting

Thinking…

Do

Step labels: Reading → Drafting → Checking

Progress feels real. Easier to spot stalls.

Right — named steps
  1. ✓ Reading your PDF
  2. → Drafting summary
  3. Checking claims

Always allow stop

Like for kids: If a helper starts cleaning the wrong room, you should be able to yell “stop!”

Don’t

No cancel while generating

Wastes tokens and patience. Feels trapped.

Wrong — can’t cancel
Generating long report…

No way to stop

Do

Stop button that actually halts the run

Control stays with the human. Safer iteration.

Right — Stop is available
Generating long report…Stop

Honest time hints

Like for kids: Saying “2 seconds left” when it takes 2 minutes is a lie. Better to say “about a minute” or say nothing.

Don’t

Fake precise countdown that jumps around

Broken estimates destroy credibility.

Wrong — fake countdown

2 seconds left

Jumps to 45 seconds… then 3…

Do

Rough ranges or step-based progress only

Honesty beats fake precision.

Right — honest range

Usually about a minute for long PDFs

Step 2 of 3 · Drafting

Remember these

  • Prefer streaming / partial results over a blank wait.
  • Name the step (“Reading your file…”) when you can.
  • Always offer Stop / Cancel on long runs.
  • Estimate time only when you can be roughly right.

Streaming and progress patterns for AI features. Rewritten in plain language for learning — not a reprint of the originals.