App Store Optimization:
An AI-powered
internal tool

The starting point

Growing a game means constantly
testing its store listing

The app icon, name, and description are what decide whether someone installs it or not. Testing them worked, but it didn't scale:

• Creative teams or developers made variants by hand.
• They uploaded each one to Google Play themselves and waited on results.
• The business teams responsible for growth couldn't move on their own.
• Tracking each test's results meant jumping out to Google Play and back.

My approach

Designed with AI, end to end

I led the design end-to-end among a squad with a PM and a developer, and used AI throughout the process itself. Before committing to the real design, we built interactive AI-generated prototypes to test the structure and flow — and showed those mockups to the teams to gather feedback before moving on.

Later, when I was ready to design the UI with our design system, we connected Claude to our git repo and I built the design fully with Claude Code. I even ran the design QA myself in the local environment — the first time we took a feature from idea to production this way.

Grounding the AI

Suggestions built on real data, not guesses

The tool starts by reading the game's own store listing, keywords, and reviews, and forming an editable understanding of the game. Every suggestion is grounded in that analysis — so the AI works from how the game actually performs and what players actually say.

The core idea

Guiding the AI, not just triggering it

For the AI to be trusted, people needed to feel in charge of it. Every generation gives the user room to guide the outcome before a single variant is made:

• A creativity level — medium by default, adjustable to high or low.
• A reference for icon tests — the current icon, a competitor, or an upload.
• Custom instructions, global or per variant.

Building trust

Suggestions you can weigh

Each generated variant comes back with an AI strategy explaining why it might work, and a confidence score — the AI's own estimate of how likely it is to convert, based on the analysis, reviews, and keywords. Instead of a wall of equal options, users get a ranked, reasoned set they can judge.

Closing the loop

From idea to live test, in one place

Because ASO Buddy integrates with Google Play, teams generate variants and run the A/B test directly in our admin platform, then track every result in the logs page. Generating, launching, and measuring all live in the same feature, so a test never has to leave the platform.