Rejected twice: monetizing a browser game in 2026

Building DrawLa was the fun part. Then we tried to make it sustainable – every round costs real server money, because a neural network runs eight inferences per second per player – and collected two rejections in quick succession. This post is the honest account, written on the very website one of those rejections was about.

The economics of a free AI game

Most free browser games cost almost nothing per player. Ours doesn’t: the live recognition that makes DrawLa fun means an always-warm cloud instance doing constant model inference. It’s not dramatic money, but it scales with every player – so “just leave it free and unfunded” quietly becomes “lose money on every game played”. Ads are the standard answer for this genre, which led us to door number one.

Rejection #1: AdSense – “low value content”

Google AdSense reviews the website an ad unit will live on. Ours failed review with the verdict Minderwertige Inhalte – low value content. Our first reaction was the usual indignation; the second, more useful reaction was actually auditing our own site the way a reviewer would. The finding was uncomfortable: besides the game itself, the site’s “content” was a dozen near-identical mini-pages of about 200 words each – the kind of thin, templated SEO filler every guideline explicitly warns about. We had written pages for a crawler, and the crawler correctly noticed.

The fix wasn’t clever: we deleted the filler and replaced it with things we would honestly want to read – a German-language drawing school with real step-by-step tutorials and original diagrams, and this English devlog with real engineering detail. Whether the next review passes is not ours to promise; the site being worth reading now is.

Rejection #2: CrazyGames – “too close to existing titles”

Door number two was distribution: game portals like CrazyGames bring their own audience and their own ad monetization, no AdSense approval needed. We integrated toward their requirements and submitted. The review came back friendly and firm:

Here’s the frustrating part: we are differentiated – no skribbl-style clone has an AI referee judging strokes live, and none has ants eating your drawing while you defend it. But a reviewer plays a few minutes, and in those minutes our game looked like “draw a word, something guesses it”. The differentiation existed; the first-minute experience didn’t show it. That’s not the reviewer’s failure. A portal reviewer is a proxy for a bored player scrolling a grid of thumbnails – if the hook isn’t visible in minute one, it may as well not exist.

The pattern behind both rejections

Two very different gatekeepers told us the same thing in different words: what you actually built is not what you presented. The site had a real game and fake content; the submission had a real hook and a generic first impression. Both rejections were, annoyingly, accurate reviews of our presentation rather than our product. That reframing changed our roadmap more than any feature request ever has:

• Content that exists because it’s useful, not because a crawler might count it (you’re reading the result).
• Front-loading the chaos: new players should meet the signature effects in their first round, not after politely climbing a difficulty curve.
• Presenting the game as what it uniquely is – an AI-refereed sabotage battle – in every thumbnail, tagline and store description, instead of the safe-but-generic “fun drawing game”.

Notes for fellow small teams

• Read your rejection twice. The first read is for feelings; the second usually contains an accurate description of your weakest surface.
• Don’t write content for reviewers. Thin SEO pages cost us more reputation than having no pages would have.
• Ad networks are not interchangeable. We briefly ran a lesser-known banner network; one of its tags force-redirected players mid-game. We shipped a same-day removal and a policy: no ad partner whose failure mode is hijacking the game.
• Rejections are per-version, not per-team. Every portal we know accepts resubmissions after substantial updates. The plan is not a new game; it’s making this one impossible to mistake for a clone in sixty seconds.

This is the last post of the launch series – architecture, latency, chaos design, and now the money chapter. The story continues as it develops; the honest version, as always.