Case Study Apr 3, 2026

We A/B Tested Three Steam Capsule Styles - Wishlist and Click-Through Lessons From One Week

A one-week Steam store page experiment with three capsule art variants—how we measured wishlist adds vs page clicks, what we changed between waves, and what actually moved the needle without breaking Valve rules.

By GamineAI Team

We A/B Tested Three Steam Capsule Styles - Wishlist and Click-Through Lessons From One Week

Your Steam capsule is not “art for art’s sake.” It is the compressed argument for why someone should click through noisey discovery feeds. We spent seven days rotating three capsule treatments on a single project page, keeping everything else as stable as possible so we could see which visual story produced more qualified visits and wishlist adds, not just vanity clicks.

This write-up is a method tour first and a results snapshot second. Numbers are rounded and contextual—your game, genre, and traffic mix will differ. The goal is to copy the measurement habit, not chase our exact percentage points.

If you have not read our breakdown of Steam discovery and what shoppers actually click yet, skim it first; this experiment assumes you already respect tag discipline and capsule safe zones.

Squirrel - illustration for Steam capsule experiment article


What we held constant

Freeze list (do not touch mid-week):

  • Core trailer and short description opening line.
  • Tag set and genre choices—we were not fighting an algorithm re-index during the capsule wave.
  • Price visibility and release window text on the page.
  • Primary social posts cadence—the only change in outbound copy was which capsule image we attached when we had to show art.

What we changed: only the main capsule image (and matching small capsule export from the same master), synced through the Steam partner tools on a fixed schedule so analytics rows stayed comparable.


The three capsule hypotheses

We numbered treatments A / B / C internally:

Variant Hypothesis Visual idea
A – Readable UI hero Players reward clarity of genre + readable logo in small sizes. Large logo, weapon silhouette, cool background clutter pushed down.
B – Motion-still fantasy Emotional pose + environment beats UI chrome for wishlists. Character mid-action, warm rim light, logo smaller.
C – Social-style crop A tighter face crop might win feed CTR even if busier. Tighter framing, higher contrast, more filmic grain.

Each was exported to Steam’s published pixel specs with the same safe margins—we did not crop differently per variant in a way that would break partner validation.


How we measured (without fooling ourselves)

  1. Primary metric: wishlist adds per day with unique page visits as a sanity check (wishlists without visits is a tracking smell).
  2. Secondary metric: click-through from the places we controlled—newsletter and Discord pinned post—using simple UTM tags so aggregator traffic did not drown the signal.
  3. Guardrail: if a variant spiked bounce-like behavior (very high visits, flat wishlists), we flagged misleading crop rather than celebrating CTR.

Steam’s own dashboards are authoritative for wishlists; we still mirrored totals in a spreadsheet row per day so we could annotate “trailer retweet day” vs “quiet day.”

Pro tip: log external beats (press pick-ups, demo drops) in the same sheet. Capsules are never isolated variables in real life.


Week timeline we actually ran

  • Day 0: Baseline screenshot of partner analytics + export all three masters.
  • Days 1–2: Variant A live.
  • Days 3–4: Variant B live.
  • Days 5–6: Variant C live.
  • Day 7: Roll back to the winner for a confirmation beat (optional, but saves regret).

We avoided swapping at midnight in our timezone—afternoon swaps meant the first traffic wave saw the new art in a coherent daily bucket.


What we saw (directionally)

Variant A (readable UI hero) produced the most stable wishlist adds per visit in our run—not the highest raw clicks from social teasers, but the best conversion once someone hit the page. Readers understood the offer faster; fewer people bounced after clicking a “mystery mood.”

Variant B won outbound tease CTR when we posted the capsule alone without trailer context—useful if your growth loop is image-first platforms where people will not autoplay video.

Variant C spiked feed attention early in the week, then cooled once the same audience saw repeats—tight crops fatigue fast unless your brand mark is unforgettable.

We are not publishing exact percentage lifts because sample size in one indie week is thin; treat our ranking as A best for wishlist quality, B best for curiosity clicks, C situational for bursts.


Operational mistakes we almost made

  • Almost shipped different small capsules that did not geometrically match the main art—bad for brand recall when someone tabs between library and store.
  • Almost changed short description on day three because a new idea felt cute—would have poisoned the capsule experiment.
  • Almost credited a wishlist spike to art when it was actually a curator list mention—annotate external events or you will mis-learn.

Applying this without a full art team

  1. One master PSD (or Affinity file) with layer comps for A/B/C, not three unrelated paintings.
  2. Batch export to Steam’s sizes in one script or action—fewer “oops we uploaded the 616 wide to the 460 slot” incidents.
  3. Weekly review even after you pick a winner—feeds and competitors move; capsules stale faster than developers expect.

For a wider solo-dev narrative with more funnel context, see From prototype to Steam page - a solo dev case study.


FAQ

Does Steam offer native multivariate capsule testing?
Partners should rely on documented Steamworks tools and policy. This article describes a manual rotation with clean controls—verify current Valve guidance before you schedule changes.

How large a sample do I need?
Higher than a single indie week unless traffic is huge. Use our method as a template; extend to two-week blocks per variant when possible.

What if my game is not visually flashy?
Lean into logo legibility + one nouns-driven hook (“co-op extraction roguelite”) in the capsule-safe zone, not painterly clutter.

Trailer vs capsule—which matters more?
For wishlists on page, trailer often dominates—we held ours still so capsules had a fair test. Do not optimize PNGs while your trailer hook is muddy.

Can AI tools generate capsule concepts?
Yes for variants and mood boards, but ship human-approved typography and readable silhouettes—generative noise often fails at 230 px wide.


Conclusion

Capsule tests are short, honest experiments: freeze what should not move, rotate what should, and log the world outside your art file. In our week, readability beat mystery for wishlist quality, while mood-first crops won narrow outbound CTR skirmishes.

If this saved you a pointless redraw cycle, bookmark it for your next Next Fest dress rehearsal, and rerun the spreadsheet when you change genres—the winner moves when your tags and audience move.


Further reading