Fitnit Lab · built in public · measured 27 Aug 2026

What it takes
to compete

Seventeen apps that own the Health & Fitness category, pulled live from the App Store — every screenshot, subtitle, price, rating count and release date — beside Fitnit's real install numbers from PostHog. Then the part that actually matters: which of their advantages are buyable, which are structural, and what the honest sequence looks like from 17 installs a week.

01Where we actually are

Every plan below has to survive this table first. The gap is not a percentage.

Installs have been flat-to-slightly-up all summer — 14, 16, 21, 16, 9, 13, 14, 14, 14, 20, 18, 22, 11, 17, 16, 28, 21, 17. Shipping 90 versions in five months moved product quality a long way and moved installs almost not at all, which is the single most useful fact on this page: we do not have a product problem in the funnel we can see, we have a distribution problem upstream of it. Signups track installs almost 1:1 (17 installs → 8–23 signups per week, helped along by the web app), so the store page is converting the handful of people who reach it. Almost nobody reaches it now — see the correction below, because that has not always been true.

02The 5,017 we already have

Every number in this section is a direct count from our own Postgres, not an estimate and not a proxy. It is the most important section on the page and it did not exist in the first draft.

What the shape says

• Something worked, for four months, and then stopped. May, June, July and August 2025 brought 1,369, 996, 1,044 and 633 accounts — 4,042 people, 81% of everything we have ever had, at roughly 1,000 a month. Today's rate is ~70 a month. Whatever that channel was, it produced 13× our current acquisition and it has been off for a year.
• They were real people, not a bot wave. 88% arrived through Google sign-in and 30–40% of each of those cohorts logged real data — a workout, a meal, a weigh-in, a step day. A scripted signup wave does not eat lunch and record it.
• The product kept almost none of them. 1,659 people used Fitnit at least once and 78 have opened it in the last month. That is a ~1.6% surviving fraction of everyone who ever signed up.
• Activation is climbing again, on tiny volume. The 2026 cohorts run 7–14%, then 30.3% in July and 46.3% in August — the best months since the 2025 wave, and they are the Apple-sign-in iOS cohorts. The current product converts arrivals far better than the 2025 product did. There are just eighty of them.
• The social layer never formed. Across 5,017 accounts: 3 friendships, 8 posts, 1 referral code. We built a feed, DMs, reactions, challenges and a referral programme for a network that does not exist yet. Nothing is wrong with the code; there was never enough simultaneous density for a second person to be there.

What the list is actually worth

4,895 mailable addresses is not one asset, it is three, and they want completely different emails:

And one live channel, already running

Search Console, the only attributed traffic we have, since it started recording in June 2026:

03The field, measured

Sorted by rating count — the closest public proxy for how many people are actually in each app. Note the second column: nearly everyone at the top has been on the store for six to seventeen years. The two exceptions are the interesting ones.

And the subtitle — 30 characters, one job each

Read straight off the live listings. Every one of them names a single job, in the buyer's own words, with no conjunction:

04Momentum, measured

Nobody publishes a download curve. Sensor Tower and Appfigures hold those behind a paywall, and every "growth chart" in a blog post is one of their estimates re-drawn. Two things are measurable for free, and between them they answer the question better than one estimate would.

What each app is doing right now

Apple publishes the most recent written reviews for any app as an RSS feed. Pull the 250 newest for each app, measure the span of time they cover, and you get a real number: how fast that app is accumulating written reviews today. Hevy's 250 most recent reviews span eight days. Alpha Progression's span five and a half years.

And what each one built, per year, to get here

The second measure is the lifetime average: total ratings divided by years on the store. It is a chord, not a curve — it cannot show a bad quarter or a viral month, and it flatters a young app while punishing an old one that has plateaued. Read as a record, not as a trend, it is still the clearest picture of who built fast.

Ours, on the same axis

This one is a real curve, because it is our own data and we can see all of it.

05What they actually make

Only Peloton has to publish real numbers. Everything below is either a company statement or a third-party estimate, dated per row — order of magnitude, not accounting.

The number worth stealing: what one install is worth

Where both figures exist, dividing them gives the only cross-app benchmark on this page that is directly usable by us — and the spread is twelve-fold:

• Fitbod — $10.00 per download. $79.99/yr, a narrow high-intent audience, and an app that is useless without the subscription. Fewest downloads of the four, most revenue.
• Cal AI — $2.86 per download. $29.99/yr against enormous volume. The creator machine buys reach, and the low price converts it.
• BetterMe — $1.50 per download (2024). Quiz-priced, paid-social-fed.
• StepsApp — $0.80 per download. The thinnest monetisation here by far, and it still clears ~$400k a month, purely because half a million people a month download a step counter.

06What their first screenshot does

Apple's own data says the decision happens in about seven seconds and the first two screenshots carry most of the conversion. Here is slot 1 from eight of them, at the size a person actually sees it in search results.

Cal AI is the one to study, because it is the one we could actually copy

Four screenshots. Not ten. They are not a feature tour — they are one job, told as four consecutive beats, in one colour, in one typeface:

A 2.4-year-old app with two founders sits above Fitbod, Ladder and Strava on rating count with four screenshots and one sentence of positioning. Whatever else is true, the store listing is not where their advantage lives — and it is not where ours is lost either. But it is the cheapest thing on this page to fix.

StepsApp: the narrowest job in the category, and #20 on the free chart

Worth its own look, because it is the counter-example to everything Cal AI teaches. No AI, no coach, no plan, no creator budget — it counts steps, a thing iOS already does for free, and it has 292,950 ratings at 4.81★, sits at #20 on the US free Health & Fitness chart, ships in 31 languages, and claims 100 million people in the first line of its description.

• It is beating its own niche 30:1. Pedometer++ has 180,782 ratings and 0.87 reviews/day; StepsApp has 26.92. Being in a big category is not the advantage — out-executing inside it is.
• It monetises worse than anything else on the page and earns more than most. $0.80 per download × ~500k downloads a month. Volume is doing every bit of the work.
• Its recent additions are our existing features. Streaks, challenges with friends, leaderboards, groups, Apple Health, widgets. A pedometer bolting on social — while we have the social and bury the steps.
• The uncomfortable read: "count my steps" is a job millions of people search for and instantly understand; "count my reps with the camera" is a job almost nobody searches for yet. StepsApp wins a commodity with volume. We cannot — entering that category means starting at #200 behind a ten-year incumbent at #20. The rep counter is the opposite trade: tiny demand, no incumbent, and a demo that plays as video.

07Ours, honestly

Same treatment, same scale, no defending it.

• Five slots of ten used. The unused five cost nothing and are worth more than any feature we could ship this month.
• No proof of any kind. No award, no review count, no user count, no press. Partly because we have none yet — which is what makes the ratings number in §01 a marketing problem, not a vanity one.
• No owned colour. Red, green-on-white, white, dark map, white. Nothing about the row is recognisable at thumbnail size.
• Every caption ends in an exclamation mark and names a feature. "Automatically track reps!", "Every Workout!", "Track walks and runs!". Nobody in §03 does this.

08The positioning problem, in one measurement

The app is called Fitnit: AI Rep Counter. Here is where it sits for the terms that name it (iTunes Search API, US, 27 Aug 2026 — directional, not App Store search rank):

Our keyword field is workout, calorie, food tracker, gym, weight loss, personal trainer, macros, barcode, squat, pushup, form, diet — twelve head terms, each defended by an app with six figures of ratings and a decade of downloads. Not one of them is winnable, and the long-tail terms we could win are being spent on instead.

09How the big numbers were actually bought

Four playbooks, sorted by whether we can run them.

What paid would cost, before we consider it

10The plan

Sequenced so each phase makes the next one cheaper. Each has a test that can fail.

11The scoreboard

Six numbers. If these do not move, nothing above worked, regardless of how much shipped.

Sources — iTunes lookup & search API (metadata, positions, 27 Aug 2026) · US App Store product pages (subtitles, IAP pricing) · PostHog (installs, signups, actives; 18 weeks to 23 Aug 2026) · our own Supabase Postgres (accounts, activation, segments, Search Console rollups; counted 27 Aug 2026) · AppTweak and Adapty Apple Ads benchmarks 2026 · Business of Apps H&F benchmarks · Sub Club on Ladder · Starter Story on Cal AI.
Rating counts are a proxy for scale, not a download report. Keyword positions are the iTunes Search API, which ranks differently from App Store search — read them as direction, not as rank.
Regenerate: python3 scripts/lab/externalize.py compete.html --out src/public/lab --name compete