Rank Sonar
All docs

Concepts

Concepts: ASO Copilot

How the Claude-powered Copilot generates keyword opportunities, metadata rewrites, and screenshot critiques.

The ASO Copilot is the differentiator that turns Rank Sonar from a tracker into an active recommendation engine. Available on the Studio tier and above.

What the Copilot generates

Four recommendation types, all powered by Claude Sonnet 4.6 via the Vercel AI Gateway:

1. Keyword opportunities

Identifies keywords where you have a meaningful chance to gain rank — high search volume, manageable difficulty, and competitor presence (validating the keyword is real). Each recommendation includes:

  • The keyword
  • Your current rank (or "not in top 250")
  • Your competitors' average rank for the term
  • Estimated rank delta if you add the keyword to subtitle or description
  • Plain-English rationale ("Three of your top competitors added this term to their subtitle in Q1 2026...")
  • Priority score (0-100)

2. Subtitle rewrites

Reads your current subtitle, your top keywords, and your competitors' subtitles. Proposes rewrites that integrate keywords naturally. Output:

  • The current subtitle
  • The suggested rewrite
  • Diff highlight of what changed
  • Which keywords were added
  • Predicted rank impact and conversion impact

3. Description rewrites

Same approach as subtitle rewrites but for the longer description field. More room for natural language; less direct keyword impact (Apple weights description less than subtitle, Google weights it more). The Copilot adapts its recommendations per platform.

4. Screenshot critiques

Analyzes your current screenshots (using Claude Sonnet's vision capability) against your top competitors' screenshots. Output:

  • For each screenshot position: suggested improvement
  • Pattern observations across competitors
  • Specific caption text recommendations

Generating recommendations

Visit Apps → [your app] → Copilot and click "Generate recommendations." The workflow runs server-side:

  1. Loads your current metadata, 30-day rankings, and competitor metadata
  2. Calls Claude Sonnet for each recommendation type
  3. Persists results to your workspace
  4. Streams updates to the UI as each step completes (~30-90 seconds total)

Recommendations are cached for 24 hours; click "Refresh" to regenerate sooner.

Acting on recommendations

Each recommendation card has three actions:

  • Apply — opens your draft metadata editor with the suggestion pre-filled. You still submit the change to App Store Connect / Play Console manually (we don't push directly).
  • Dismiss — removes the recommendation. Future recommendations learn from your dismissals.
  • Save for later — moves to a later-review list.

How the AI uses your data

The Copilot reads from your Rank Sonar workspace:

  • Your apps' metadata (title, subtitle, description, screenshots, current rank for each tracked keyword)
  • Your competitors' public metadata (same fields, scraped daily)
  • Search volume and difficulty data for keywords you track

It does not read:

  • Your App Store Connect / Play Console raw data (downloads, revenue)
  • Other workspaces' data
  • Anything you haven't explicitly added

The model: Claude Sonnet 4.6, called via the Vercel AI Gateway with zero-data-retention enabled. Inputs and outputs are not used to train Anthropic's models.

Cost and limits

Copilot generation costs us roughly $0.10-0.50 per app per refresh, depending on how many keywords and competitors are analyzed. Limits per tier:

  • Studio: 100 generations per workspace per month
  • Agency: unlimited

Excess generation requests queue and run when capacity is available.

See also