We are paying TikTok to show ten short videos of our game for one month. We only count the people who play, win a run and come back. Then we compare the price with what Reddit charged us.
Written 27 September 2026, the day the videos went live, before a single result. We publish these whether they work or not: see Round 9 on Reddit, the round this one is measured against, and Round 6, where two of our five guesses were wrong.
The whole story in 20 seconds
🎬
What we test
10 videos
on TikTok, in 3 styles. Which kind brings real players?
💶
How much money
€35.62 a day
the same as Reddit got. About €1,140 in total.
📅
How long
32 days
27 Sep → 28 Oct. Results here by 7 Nov.
🏆
How we'll know
Price per player
who comes back another day. TikTok vs Reddit.
🎯 Why: a second place to find players🎥 and learn which kind of video is worth making
1What we already know
Last month we changed what we ask Reddit for. Not "anyone who clicks", but people who actually win a run. The price of each winner fell by more than half.
What one winning player cost us
Reddit, R8paid for clicks
$2.58
Reddit, R9paid for winners
$1.07
TikTok, TT1paid for winners
?
Shorter bar = cheaper. R9's first three weeks, from 3 Free Months. The full-month R9 number comes in early October.
🧪 We tried TikTok once before
In late June a tiny test sent us 54 visitors. One played. Nobody came back.
Back then TikTok had no idea who played. Now it hears about every win, so it can look for more people like them.
🧺 Why not just stay on Reddit?
One shop is one point of failure. If TikTok finds winners at Reddit's price, we are no longer stuck with one seller.
And we have made many kinds of video but never tested one against another.
2The trip we pay for
Every player walks the same five steps. TikTok is paid to bring us people who reach step 4. We judge it on step 5.
👀
Sees a videoscrolling TikTok
▼
👆
Taps itlands on our game, in the browser
▼
🃏
Playsstarts a run
▼
🏆
Wins a runTikTok is told
▼
🔁
Comes backon a later day
A picture, not to scale: each step keeps fewer people than the one before.
Step 4 is TikTok's target. A win is the signal it gets, just like Reddit did.
Step 5 is our scorecard. Someone who returns within a week is someone who liked the game.
We do the counting. Every player is traced to the exact video that brought them, by our own game, not by TikTok.
3The ten videos
A person talking · a face speaks to you, then the game2
Gameplay edit · the real game screens, cut fast2
Cinematic · the game's paintings brought to life6
Why 6, 2 and 2? These are simply the ten best of many drafts. That makes the style comparison weaker, and we say how in section 7.
Tap any video to play a 6-second clip; tap again to pause. Sound is off by default: tap the 🔇 button on a video to hear it (one video at a time).
A person talking2 videos
"Wait, this runs in your browser?!" A presenter reacts over a real run.h1-browser-run · 18 s
"One more run. Six hours ago." Two roommates, five real cards, then real play.AIthe couple is AI-made; nobody in it is a real player · c2 · 11 s
Gameplay edit2 videos
"Strike or defend?" A real fight, played at real speed.AIcard art animated at start and end · c01-b-r6 · 14 s
"A charm changes the opening." One lucky charm and the cards it powers up.AIanimated art inserts; the game's own music · n09-c-r10 · 15 s
Cinematic6 videos · no words, no buttons, no logo. They sell the mood, not the rules.
A sleeping door wakes up and swings open.c07-p2 · 12 s
The real win screen, keys swirling round the monument.c10-t1 · 8 s
Zael shoulders his hammer and turns to you.c6-t3 · 8 s
Into the boss arena, where Hushbarrow rises.f03-hushbarrow-t2 · 12 s
Creatures pass a torch, then The Undertow appears.p10-t3 · 12 s
Five real cards light up, one by one, on a stone altar.r06-t2 · 12 s
AIEvery cinematic film starts from a painting that is in the game, brought to life with an AI video tool.
4The money
One coin per day, 32 days
€35.62 × 32 days ≈€1,140at most (about $1,300)
€35.62 is exactly $40.62, Reddit's daily budget in Round 9. Same money, so the price comparison is fair on that point.
🎰 TikTok decides which video gets the money
We give TikTok the whole pot and all ten videos. It picks its favourites. On Reddit, one picture ate most of every day's budget:
1 favourite ad
82–90%
the other 9
the rest
Share of each day's spend in Reddit Round 9. We expect the same pile-up here (guess #2).
5The calendar
Reddit R9
TikTok TT1
1 Sep1 Oct1 Nov
28 AugR9 starts
8 Octfirst look
7 Novresults
The two rounds overlap for about one day (27–28 Sep). TikTok stops on 28 Oct at 23:59; the striped part is the week the last visitors get to come back.
6Reddit vs TikTok: what's different
This is not a clean fight, and we say so now. The goal and the money match. Almost everything else changes at once.
Reddit · R9
TikTok · TT1
🎯 What we pay for
a player who wins a run
a player who wins a run
SAME
💶 Money per day
$40.62
€35.62 = $40.62
SAME
🧾 How a win counts
only after a tap on the ad, within 7 days
only after a tap on the ad, within 7 days
SAME
🗂️ Number of ads
10
10
SAME
🖼️ Format
still pictures
tall videos, 8–18 s
DIFFERENT
📱 Where it shows
Reddit, phones only
TikTok + partner apps
DIFFERENT
🌍 Who sees it
worldwide, card-game fans
adults 18+; no India, no US
DIFFERENT
📒 History with us
two rounds already
one tiny test
DIFFERENT
📅 Month
28 Aug – 28 Sep
27 Sep – 28 Oct
DIFFERENT
🇮🇳 The biggest difference: India
About half of Reddit's visitors came from India. There, far fewer first games ended in a win. TikTok is not available in India at all.
India
9 in 100
everywhere else
24 in 100
First games that ended in a win, Reddit R9. So TikTok could look better for a reason that has nothing to do with video.
7How we will decide
One number decides it: what it cost to find one player who came back another day. We set the line at 30%, the same gap Reddit's new approach had to beat.
🥇TikTok winsmore than 30% cheaper
🤝A matchwithin 30%, either way
👎TikTok losesmore than 30% dearer
← cheaper than Redditdearer than Reddit →
We do not decide here what happens next. That also depends on Reddit's final number, which is not in yet.
Four reasons the answer may be fuzzy
🔀
Many changes at once
New app, videos instead of pictures, other countries. A result is about "TikTok with these videos, this month", not "video beats pictures".
🌏
No India
Half of Reddit's visitors were from India, where wins were rarer. TikTok has none, which flatters it.
🪙
Maybe too little money
€1,140 over ten videos may not be enough to tell the three styles apart. Our Reddit rounds never could.
🎰
TikTok splits the pot
A style that gets €30 can't be fairly judged against one that gets €900. We don't control the split.
So a style only gets a verdict if at least 150 people tapped its videos and at least 20 of them won. Below that we print "not enough data", not a ranking.
8Five guesses, written down first
We write our guesses down before we see results, so we can't fool ourselves later. A test we cannot lose is not a test. On 7 November we tick each one.
🏷️
GUESS 1
TikTok will cost more per winner than Reddit
A brand-new account, tracking switched on the same day, and ads in partner apps all push the price up. Missing India pushes it down. We think the first group wins.
How sure: medium
7 Nov check:✔ right✘ wrong
💰
GUESS 2
The money piles onto one or two videos
TikTok will pick a favourite early and feed it, just as Reddit fed one picture most of every day's budget.
How sure: high
7 Nov check:✔ right✘ wrong
🎬
GUESS 3
Pretty films get cheap taps but few winners
The cinematic films show no play: no hand of cards, no fight, no choice. They bring people who liked a picture. Videos that show the game should win more runs per visitor.
The one we most want to know
7 Nov check:✔ right✘ wrong
🔁
GUESS 4
We won't see which style brings people back
Too few returning players, split three ways. It has been like that in every Reddit round so far.
A note on our own method
7 Nov check:✔ right✘ wrong
📲
GUESS 5
Some money leaks into partner apps
A real share of the budget will be spent in other apps TikTok works with, not on TikTok itself. Those visitors will start games less often. We check this at the first look, 8 Oct.
How sure: low · we've never run there
7 Nov check:✔ right✘ wrong
9The scoreboard we'll fill in
Reddit · R9 · pictures
Price per winner
early Oct
Price per player who came back
early Oct
Money spent
early Oct
TikTok · TT1 · videos
Price per winner
by 7 Nov
Price per player who came back
by 7 Nov
Money spent
by 7 Nov
For the data nerds: the exact rules
Everything above in its technical form. These rules were fixed on 27 September, before launch (pre-registered), and are what the 7 November readout is graded against.
Campaign setup
Objective: Sales → Websiteoptimising for a won run; the TikTok pixel reports a win as Purchase, as on Reddit
€35.62 / day= $40.62 at the 25 Sep ECB rate (1.1403). Cap about €1,140 ($1,300)
1 ad group, 10 adsone per video; the optimizer allocates spend across them
Broad, 18+every country TikTok offers our account; the US is not on that list, India has no TikTok
PlacementsTikTok + Pangle (Ad Network) + Global App Bundle; TikTok-only was not allowed
Creative as madeSmart+ on; smart creative, auto music and auto enhancements off
Attribution: 7-day clickno view-through; a win counts only after a tap on the ad
Consentthe TikTok pixel waits for the same cookie-bar yes as Reddit's (v7.3.1); the terms name TikTok
The metric
Primary: $ per returning player (CPA on D1–D7 returners): a visitor who plays again on a later calendar day within 7 days of first arrival. R9's own verdict metric.
Secondary: $ per winner (the optimisation event). Shows first, same day.
Counted in our telemetry, not TikTok's. Players are attributed by each ad's utm_content tag, consenting or not. Spend is the only platform number, converted at the fixed 1.1403 rate so FX drift cannot fake a difference.
Age-matched: day N of TT1 against day N of R9, never calendar date against calendar date.
Verdict band: within ±30% of R9's full-round $/returner = match; >30% cheaper = beats; >30% dearer = loses. 30% is the margin R9 had to beat R8 by.
The per-style read
Spend, visitors, runs started, winners and returners summed per style (2 + 2 + 6 ads); compared on $ per winner.
A style gets a verdict only with ≥ 150 visitors and ≥ 20 winners; below that it reads "insufficient data".
Two styles differ only if wins per visitor differ at p < 0.05. Returners per style are reported, never ranked.
Style is confounded with the specific videos: "cinematic" is six films, "a person talking" is two.
Prior: in R9, r9-path took 82–90% of daily spend. In Round 3 per-ad return rates were statistically identical (p = 0.80); Round 4's three finalists tied.
R9 vs TT1, side by side
R9 · Reddit
TT1 · TikTok
Optimisation event
won a run (Purchase)
won a run (Purchase)
same
Daily budget
$40.62
€35.62 = $40.62
same
Attribution
7-day click, no view-through
7-day click, no view-through
same
Structure
1 ad group, 10 ads
1 ad group, 10 ads
same
Format
flat 4:5 images
9:16 videos, 8 to 18 s
different
Inventory
Reddit feed and threads, mobile only
TikTok plus partner apps
different
Targeting
global, anchored on deckbuilder communities and keywords
broad, 18+; no India, no US
different
Account history
had run R7 and R8; pixel live 2 days before launch
one tiny test in June; pixel live the same day
different
Window
28 Aug to 28 Sep
27 Sep to 28 Oct
different
India: 48.7% of R9 visitors through 6 September; first-run win rate 9.0% there vs 24.4% elsewhere.
Scoreboard, per style
TT1 style
Ads
Spend
Visitors
Started
Won
$ / winner
Came back
A person talking
2
pending
—
—
—
—
—
Gameplay edit
2
pending
—
—
—
—
—
Cinematic
6
pending
—
—
—
—
—
Milestones
27 Sep ✓ v7.3.1: TikTok pixel ships behind the cookie bar; wins reach TikTok as Purchase, checked end to end. TT1 published, in TikTok review, runs from 17:15 Warsaw time.
28 Sep R9 closes on Reddit.
8 Oct First look, read-only: spend by video, by placement, by country.
28 Oct 23:59 Round closes. The last visitors get their 7-day return window; readout published here by 7 Nov.
Sources. R8 and R9 figures: our own telemetry, published in 3 Free Months on 20 September (Round 8, 4 to 27 August; Round 9 from 28 August, cohorts to 19 September) and the R9 plan (India share and win rates, read 8 September). Every TT1 figure will come from our telemetry except spend, which comes from TikTok's report. How the videos were made: the gameplay edits and cinematic films were animated from the game's own art with an image-to-video model; c2's couple was generated by a video model; h1 was made outside our usual pipeline.
