When A/B testing is worth the effort
A thumbnail experiment is useful when the island already has a clear audience and the current image is not the only bottleneck. First confirm that the title, description, code and playable experience are ready; otherwise a thumbnail change can only diagnose the browse step, not fix the full funnel.
Use the Creator Portal testing workflow when it is available to your island. Keep the island, title and publishing window stable so the comparison answers one question: which visual promise makes the right player want to inspect this experience?
- Run a test after the island hook is stable, not while the title and description are changing every day.
- Keep both variants within the same 16:9, 1920×1080 and file-size baseline before upload.
- Record the start and end time, island version and exact thumbnail prompt so the result is reproducible.
Change one meaningful creative hypothesis
The strongest variants communicate the same island through different visual priorities. Variant A might lead with the vault escape; variant B might lead with the squad race. Both should remain accurate, but each should make a distinct action win at browse size.
Avoid meaningless tests such as a two-pixel crop or a barely different hue. Those changes are difficult to interpret and do not teach you what players understood. Keep the title treatment, focal subject and contrast intentional in both concepts.
- Hook test: objective or moment first, such as escape, race, defend or solve.
- Composition test: close character silhouette versus a wider environment that explains the mode.
- Message test: a clean editable title treatment versus an image-led card with no baked-in copy.
Read impressions, clicks and CTR together
Impressions tell you whether the island was shown, clicks tell you whether the card created curiosity, and CTR relates those two counts. A higher CTR with very few impressions is not automatically a durable winner, so note the exposure context instead of declaring a result from a single small slice.
Keep a simple experiment log: variant description, dates, impressions, clicks, CTR and any notable island or audience change. When one version wins, reuse the winning visual principle in the next thumbnail or launch asset rather than copying the exact image across every channel.
- Compare the same metric definitions and time window for both variants.
- Look for a consistent direction across impressions and clicks, not CTR in isolation.
- Treat low-exposure results as a hypothesis to revisit, not as a universal benchmark.
Keep every variant policy-safe and truthful
A test is not permission to use misleading artwork. Both variants should represent the live island, use original or licensed assets and avoid XP or AFK claims, platform-specific controls, copied third-party IP and any promise a player cannot experience.
Before uploading, inspect both images as small Discover cards and at 100%. Remove tiny text that disappears, check the file limit, and confirm that the focal action still matches the current island build. A compliant loser is still useful evidence; a misleading winner is expensive churn.
- 16:9, at least 1920×1080 and no larger than the current 5 MB limit.
- Same island promise in both variants; no invented mechanics, rewards or locations.
- Original, general-audience artwork with no XP/AFK language, coin imagery or platform button prompts.
- Save the winning hypothesis and carry its palette and silhouette into the next launch asset.