Frank Michael Smith: An 8-Hour Quiz Game Doing $40K/Month

By Wanli Lu · Updated October 2026 · 11 min read

The story in one paragraph

Frank Michael Smith is an independent sports creator with ~4M social followers who does not code. In June 2026, driving home from a wedding, he remembered a map-guessing game (MapTap) and mashed it with his own sports trivia. He opened Claude, described the idea in plain English, and spent about 8 hours that Sunday building GeoSports — a daily 5-question sports-geography quiz where you click a globe to guess a city. Three months later it had ~700K monthly active users and ~$40K/month in revenue. He spent $0 on advertising.

What the product actually is

GeoSports is deliberately tiny. Five rounds a day: a clue (e.g. "a tailgate superfan injured himself climbing over a table in this NFL city"), you click the globe to guess the location, proximity = score. No tutorial is needed — players just see a result and share it. That shareable result is the entire growth engine: a 53% share rate means more than half of players post their score. The game sits under an umbrella brand, GeoGames, with spin-offs GeoHistory and GeoFooty reusing the exact same mechanic.

The distribution play (the real story)

The headline is "8-hour build." The actual lesson is distribution. Frank's own 4M followers drove only ~2% of traffic; 98% came from word-of-mouth. How:

• One comment, not an ad. He replied to an NBA thread with "I made a game, try it" — that single reply brought 40K players day one, 150K day two.
• A built-in reason to compare. Sports fans love measuring themselves against each other; the shareable score card turned every player into a billboard.
• Borrow, don't rent, audiences. He invited sports bloggers and retired athletes as "question guests," tapping their follower pools to roll another round.
• Replicate the mechanic, not the topic. Same engine → GeoHistory, GeoFooty. One piece of content became a row of products.

Notice what AI did not do: it didn't decide "fans love comparing scores" or "this one sentence makes people forward it." That's 13+ years of sports-content instinct. AI only turned the idea into something clickable.

The money: two legs, ads do the heavy lifting

$32K + $9K ≈ $40K. The structure matters more than the total: ads are ~80% of revenue, which means the product had to earn attention before it could earn money. That's why he obsessed over share rate, not paywall conversion. A game nobody opens daily makes its ad slots worthless.

The method, as a copyable skeleton

The original write-up distilled five cases (Frank, Connor/Payout, Hasaam/Launch Fast, Ryan/Cat-on-Chair, Sarah/Stella) into a repeatable path. The version worth stealing:

The 5 steps

• 1. Pick a crowd, not a category. "AI + fitness" is too big; "pace reminders for marathoners in their last 2 weeks" is doable. Start where you already belong.
• 2. Find a repeated, compare-or-lookup action. Sourcing products, checking rebates, arguing which city a game was in — if it recurs, people will pay to save the trouble.
• 3. Ship a runnable v1 in 48 hours. Plain-English brief to an LLM; screenshot errors back. Ugly is fine; clickable and shareable is not.
• 4. Say it in one human sentence where the crowd already is. No ads — test whether anyone fights to use it. Frank's comment beat any campaign.
• 5. Charge before you polish. Start at "a milk tea": $4.99/mo overseas, ¥9.9–29.9/mo domestic. Someone paying proves demand is real.

The 7-question filter (do NOT code if you can't answer most)

• Can I get into this crowd?
• How many times a month does this problem hit them?
• Is anyone already paying to solve it?
• Can I build v1 in 48 hours?
• Can I find 50 testers today?
• Would they pay "a milk tea" to save the time?
• If nobody pays in 3 months, do I stop?

The cold water: the odds are ugly

Every person in that article was reported because they won. RevenueCat's State of Subscription Apps 2026 (115K+ apps, $16B+ revenue) shows the base rate:

• Median 1-year revenue for a new app: ~$72/month.
• Only 4.6% ever reach $10K/month — roughly 1 in 22.
• Pre-2020 apps still earn 69% of all revenue; post-2025 apps only 3%.
• New subscriptions shipped per month ballooned from ~2,000 (Jan 2022) to ~14,700 (Jan 2026).

AI compressed build time from 3 months to 8 hours. It did not compress "nobody shows up." The 8-hour build is the least replicable part — Frank's 13-year content head start and 6-year niche-game reps are what made those 8 hours pay off.

Replicability for us

The durable core — web-first product, English/USD surface, AI-built, distribution via shareable value, ads + subscription — is exactly what we already run. Mapping it:

So the honest answer: we already own the bones. What we lack is not the build (done) or the indexing (done) but (1) an approved ad network + USD payout, and (2) a shareable hook + borrowed-community distribution. Those are marketing problems, not coding problems — and the article's whole point is that the coding problem is already solved.

Key moves you can copy

• Ship web-first, skip the app store. Frank's near-zero friction (open a link, play) is why it went viral. calcuxa is already web-first — keep it that way.
• Engineer a reason to share. A brag card beats a bare answer. Give calcuxa results a "share your number" button.
• Borrow audiences, don't buy them. One sentence in the right thread > a budget. Find where people ask "what's my mortgage payment" and answer with a link.
• Charge a milk tea early. calcuxa's Pro (ad-free + extra tools) at $4.99 mirrors Frank's exactly — but only after traffic justifies it.
• Replicate the mechanic, not the topic. calcuxa's calculator engine could spawn sibling tools (the GeoGames move) once one proves demand.
• Run the 7-question filter first. If you can't name the crowd and 50 testers today, you're not ready to build — AI or not.

Lessons and anti-patterns

• "8 hours" is a headline, not a method. Frank had 19 years of accumulated judgment (13 yrs content + 6 yrs niche games) before the 8 hours. Copy the judgment, not the clock.
• Ads need eyeballs first. Frank's 80% ad revenue only works because 700K people show up daily. A quiet site monetizes poorly on ads — that's why calcuxa's near-term bet is affiliate + Ezoic, not a paywall.
• Survivor bias is the trap. 1-in-22 hits $10K/mo. Treat any "$X/month" screenshot as the top of a distribution, not the middle.
• USD payout is structural, not a prompt. From mainland China, AdSense/Ezoic payout and many networks need Payoneer/Wise or a Stripe-Atlas entity. Solve this before chasing ad revenue.
• Try the 48-hour ugly v1. "Perfect is the enemy of shipped." Frank's first version was rough and immediate — that's the lesson, not his particular game.