Google Health · Coded Material 3 prototype · Self-initiated · 2026

I redesigned Google Health around the questions people open it to answer.

After Google replaced Fitbit with Google Health, daily status became harder to find inside a dense activity feed. In one week, I audited the shipped experience, defined five redesign principles, and built a coded Material 3 prototype.

Personal project, not affiliated with Google. Every "before" is my screenshot; every claim, a cited law or a labeled hypothesis.

▶ tap the working prototype
shipped
Before: Google Health's shipped Today feed
redesign
← drag to compare →
0★
Play-Store rating after the forced May 2026 rollout: "hundreds and hundreds" of one-star reviews; "Google Health Ruined Fitbit" hit 600+ upvotes.
The backlash
0 → 2
Choices per detected activity, after the "smarter detection" chip cloud collapses to one binary question, asked once, not after every walk.
The core fix
0
Fixes on Google's own public roadmap, published one day after full rollout. The UX case isn't my opinion. Google conceded it in writing.
The permission slip
Case snapshot
Role

Solo: product audit, IA, interaction, visual, motion, prototyping, frontend

Timeline

July 2026 · ~1 week of evenings

Skills

HTML/CSS/JS · Material 3 Expressive · Claude Code

Scope

Full app shell: 4 tabs, activity detail, Log flow, detection, system states

Why this app

Google force-replaced the Fitbit app with Google Health in May 2026. It was review-bombed; Google shipped a 39-item fix roadmap within a day. I use it daily. The frustration, and every screenshot here, is first-hand.

The honest frame

No access to Google's research or metrics, so nothing here claims impact. The redesign leans on the two things nobody can argue with: Google's own design system and named UX laws. Anything unmeasurable is labeled a hypothesis. Assume smart people built the original under constraints I can't see.

01 · Context

the internet noticed before I did.

I wanted my step count. I got a detection log, a percentage that admitted it was made up, and the same eight chips after every walk. I wasn't alone:

"Why is the biggest button on the screen for AI questions? You're a fitness app; the biggest button should be to add an activity."

User review · via Kotaku

"Several screens have become Walls of Words, bombarding the senses with too much stuff. Impenetrable. Off-putting."

the5krunner · design critique

"Google's single-minded focus on AI has ruined what was a much better experience… a huge step back in usability."

Android Authority · one-week review

To be fair: the visual craft is genuinely good: reviewers called it "gorgeous," and the sleep visualization earned a 9/10 from the app's own harshest critic. That's exactly what makes it interesting: this isn't a paint problem. It's an information-architecture problem, which means it can be argued with principles, not taste.

02 · The audit

four screenshots, four broken laws.

My own screenshots, July 10, 2026.

Google Health Today feed 1 2 3 4
Detection chip cloud 1 2 3
The same chip cloud again, one walk later 1
Sleep detail, five stacked stage tracks 1 2
Exhibit A: a log, not a dashboard.
  1. 1184% of a target that doesn't exist yet.
  2. 2Status hidden behind carousel dots.
  3. 3Three actions, none primary.
  4. 4"Readiness"? "Cardio load"? Never defined anywhere. My biggest pet peeve.
Exhibit B: the machine's homework, outsourced.
  1. 1Eight choices for a yes/no question.
  2. 2Delete dressed as a suggestion. One mis-tap kills the workout.
  3. 3Chips + thumbs + menu: three feedback systems.
Exhibit C: two scrolls later, the same question.
  1. 1Asks after every activity, never learns. Banner blindness, earned.
Exhibit D: five charts for one night.
  1. 1Green means three different things.
  2. 2Two screens of scrolling to see one shape.
03 · Principles

five rules the redesign never breaks.

P1

Show status before history

The home screen answers "how am I doing?" in one glance. History demotes to a timeline.

P2

Give each fact one clear home

Sleep lives in exactly one place per screen. Duplicates are deleted, not restyled.

P3

Ask one question at a time

Detection asks a binary question, learns, and promises to ask less. The taxonomy hides behind "edit".

P4

Design honest loading and calibration states

Loading and calibrating are designed states with visible progress and an arrival date, never a fake number.

P5

Work within Material 3 instead of inventing another system

Strict Material 3 Expressive: its tokens, type scale, 48dp targets, springs, and shapes. No invented system.

04 · The redesign

screen by screen, before and after.

Big numerals, generous space, one accent that always means "act here." Light by default, dark on tap. Flip any panel back to the original.

the home screen

Before: Google Health Today feed (status region)
← drag to compare →
Every fact visible, timeline demoted, the biggest button logs an activity.

the detection flow

Want smarter detection?

Confirm your activity or specify a new one to get deeper insights and smarter detection

WalkPower walkHike RuckingStroller walkNordic walk listMoredeleteDelete
thumb_upthumb_down···
Detected automatically. Did we get it right? checkYes Edit

Eight choices become two, asked once. Rename hides behind "edit"; delete moves to the detail view, styled destructive.

sleep, one picture

Before: the sleep-tracked card (score + five stacked tracks)
← drag to compare →
Four flat lanes, one per stage. Every waking, every REM cycle, readable at a glance. One color family, and lightness alone means depth.

the 184% problem

184%
Weekly Cardio 184% · Calibrating · 3 days left · +42
Cardio load 42 this week. Your target arrives in 3 days.

The earned load stays; the fake percentage waits for a real target. Calibration becomes a loop worth closing. And the jargon (readiness, cardio load, zone minutes) now explains itself in plain words, one tap away.

three decisions I sweated

D1 · dark, or Google-default light?

First pass: M3's light baseline. Correct, and lifeless. Dark became the identity: every M3E rule kept, one volt accent meaning "act here". Light ships too: flat neutral canvas, white cards, the accent swapped for black, because a health app gets read in morning sun.

Trade-off: no shadows on dark. Elevation comes from M3's surface ladder.

D2 · where does delete live?

Overflow menu: buried. Long-press: undiscoverable. Winner: the detail view, styled destructive. Deliberate, un-mis-tappable.

Trade-off: deleting takes two taps, the price of never losing a workout.

D3 · what owns the hero?

Readiness is the smartest number, and blank during calibration. Steps won: always true, always actionable. Readiness reclaims the hero when it has something to say.

Trade-off: the most differentiated metric sits below the most reliable one. On purpose.

05 · The system

strictly google's rules, receipts attached.

"Follow the design system" is easy to say. Here's what it means, mechanically.

one accent, tokenized

Neutral dark surfaces carry zero meaning; color is reserved for roles. Volt is the only action color in the entire app. If it glows, you can act on it.

volt · act #D6F14E teal · sleep #43E5C6 cyan · cardio #54C7F4 mint · goal #8BE6A8 surface-low #1C1C21

the M3 type scale, verbatim

Roboto Flex on Material's exact ladder. Hierarchy comes from the scale, not from inventing sizes.

display · 449,389
title · 22/28Evening walk
body · 14/20Detected automatically. Did we get it right?
label · 12/16CARDIO LOAD

expressive motion, two springs

M3 Expressive replaces easing curves with spring physics: spatial springs overshoot (hero moments), effect springs settle clean. Every transition in the prototype uses one of these two.

expressive · overshoots
standard · settles

48dp or it doesn't ship

Material's minimum touch target is 48×48dp. The original's confirm chips missed it; so did my first pass. I held myself to the same rule.

checkYes

36px visual pill, invisible 48dp hit area (dashed). Anything smaller is a mis-tap waiting to happen, especially mid-walk.

06 · In your hands

built, not mocked. tap it.

Not a picture of an app. An app-shaped thing that runs. Switch tabs, open a walk, confirm a detection, log from the volt button.

Hand-written HTML/CSS/JS, no framework, same as this site. Open it full-screen →

07 · What I'd test

hypotheses, not victory laps.

An unsolicited redesign can't claim outcomes. It can commit to measurable predictions:

H1 · Glanceability

Users answer "how am I doing today?" from the redesigned home in under 3 seconds, without scrolling.

Test: 5-second test + first-click, before vs after · metric: time-to-answer, scroll events before first fixation on status

H2 · Detection trust

One binary question raises confirmation rates and stops the prompt from being ignored as noise.

Test: A/B chip-cloud vs confirm-strip · metric: confirm rate, dismiss/ignore rate, model-correction quality over 4 weeks

H3 · Calibration retention

Framing calibration as visible progress ("4 of 7 nights") increases wear-through-calibration versus a dead "calibrating" label.

Test: cohort comparison, new-device users · metric: D7 wear rate during calibration window

H4 · Sentiment recovery

Fixing the top three roadmap complaints (glanceability, prompts, customization) moves store sentiment measurably.

Test: release-over-release review mining · metric: rating trend, complaint-theme frequency

What I couldn't know from outside
  • Why the feed is a log: there may be an engagement or AI-Coach strategy I'm arguing against blind.
  • The Premium funnel: the AI Coach is paid, and my "quiet" detection strip removes one of its loudest surfaces.
  • Sensor truth: maybe cardio load genuinely can't show a partial target sooner. (Though then it shouldn't show 184%.)
  • The roadmap: some of this may already be in flight inside Google. Their 39-item list overlaps my audit heavily, which I take as validation, not competition.
08 · Reflection

what I learned.

The fix was subtraction, not addition.

The redesign adds no features. It deletes duplicates and hides the taxonomy behind one honest question. Less app, more answer.

Constraints are the argument.

Leaning on Google's own system and named laws made the case impossible to wave off as taste. The rules were already theirs.

same pixels, different priorities.

Google Health isn't badly drawn. It's badly prioritized. The redesign reorders what already exists around the one question people open the app to ask.

▶ the working prototype let's talk →