Apple Ads Benchmarks

Published Apple Ads baselines for subscription apps, with the comparison rules that keep unlike populations apart.

This agent skill packages Adapty's 2026 report as local, source-traceable CSV data. It covers the 2025 campaign period across 8,000+ apps, 1,000,000+ ad groups, 90 countries, 22 primary App Store genres, and 59 report niches. It needs no account, CLI, network call, or subscription.

Full report: Apple Ads for subscription apps 2026

What is published

• Acquisition: TTR, CR, CPT, and CPA for 90 countries, 22 App Store genres, and 59 niches.
• Subscriber economics: published install-to-trial/install-to-paid comparisons, the top country and US-category tables, year-1 LTV, cost per paying subscriber, and paying subscribers per $1,000.
• Measured effects: Custom Product Page baselines/lifts and relative tailored-paywall uplift, including the published Day 0-Day 92 ROAS uplift series.
• Machine-readable taxonomy and provenance: stable keys, PDF locators, internal workbook locators, source hashes, sample context, and the report's inclusion thresholds.

Only derived CSV/JSON is distributed. The source PDF/XLSX and raw impressions, taps, downloads, and spend are deliberately excluded from the repository and installable plugin.

Metric definitions

The names follow the PDF exactly:

Metric	Denominator
ttr	taps / impressions
cr	downloads / taps, measured on the App Store product page
cpt	spend / taps
cpa	spend / downloads

The report does not publish CPI as spend per install. CPA/download is not silently aliased to CPI: when a user asks about CPI, the skill first checks whether their denominator is downloads or installs.

The published 28.9%-77.9% niche range is cr (tap-to-download): Time Planner is the low end and PDF Reader the high end. It is not install-to-trial or trial-to-paid.

Install

claude plugin marketplace add adaptyteam/apple-ads-cli
claude plugin install apple-ads-benchmarks@adapty

Other agents

npx skills add adaptyteam/apple-ads-benchmarks --all

Pairs with apple-ads-cli: this repository describes what was published; the CLI operates an Apple Ads account.

Updates and releases

Claude Code installs marketplace plugins into a local cache; it does not read this repository live. Because .claude-plugin/plugin.json declares an explicit version, every published change must bump that version. Pushing new commits without a version bump will not produce an update for installed users.

Release checklist for maintainers:

1. Update the plugin and run the validation commands below.
2. Bump the version using semantic versioning: PATCH for fixes, MINOR for new compatible features, and MAJOR for breaking changes.
3. Commit and push the plugin repository. Update its entry in the adapty marketplace if the description, source, or other discovery metadata changed.
4. For Cowork distribution, build and publish a new .plugin archive. The archive is a static delivery artifact and does not update itself.

Claude Code users can fetch a release manually:

claude plugin marketplace update adapty
claude plugin update apple-ads-benchmarks@adapty

Run /reload-plugins in an existing Claude Code session, or start a new session, to load the new version. Auto-update can be enabled for the adapty marketplace in /plugin → Marketplaces; third-party marketplaces do not enable it by default.

Skills bundled inside this plugin update together with the plugin. A skill installed separately with npx skills add follows that tool's separate update/reinstall flow. A Cowork .plugin must be rebuilt, redistributed, and imported again.

See the Claude Code documentation for plugin updates and version resolution and marketplace auto-updates.

Example questions

What is the published CPA for Remote Control apps?
How does Brazil's Apple Ads CR compare with the country median?
What was the year-1 LTV reported for the US?
How much did keyword-matched paywalls lift ROAS at Day 92?
Can I compare my Utilities CPA in Brazil with this dataset?

The last acquisition question has no combined category-by-country row. The skill can show the global Utilities and all-category Brazil marginals separately, but it will not synthesize an intersection.

Data layout

data/2026/
  manifest.json                 schema, coverage, sample context, rounding, source hashes
  acquisition.csv               684 matrix rows plus 6 published distribution summaries
  subscriber-economics.csv      published funnel, LTV, and subscriber-efficiency tables
  effects.csv                   CPP and tailored-paywall measurements
  taxonomy.csv                  stable genre, report-category, and sub-niche keys
skills/apple-ads-benchmarks/
  SKILL.md                       comparison and refusal protocol
  references/                   methodology, comparison, taxonomy, behavior fixtures

The numeric CSVs use one schema:

metric,scope,category,sub_niche,geo,traffic_source,variant,cohort_day,stat,value,unit,
period,source_id,source_locator,value_source_id,value_source_locator,n,sample_note,note

scope is one of overall, country, app_store_genre, or sub_niche. Country, genre, and niche rows are separate marginal aggregates, not a cross-product.

Source and sample policy

The PDF is authoritative for definitions, publication scope, methodology, printed values, and display precision. The non-distributed workbook fills matrix values that are not printed on a slide. If a workbook-derived rounded value conflicts with a printed value, the PDF wins and the row records that override.

The report does not publish per-row app counts, so n is blank for this vintage. Answers must state the vintage and report-wide sample context; they must not invent a row-level sample size. The PDF publishes these niche eligibility thresholds:

• at least 2 contributing apps;
• at least 200,000 impressions;
• at least 1,000 downloads;
• at least $10,000 annual spend per app.

What this data cannot support

• It is not per-app competitor data, a forecast, a target, or a guarantee.
• Its acquisition matrix has no category-by-country intersections. The separate subscriber file includes the PDF's US-only category install-to-paid table.
• It has no category-level trial-to-paid baseline or LTV curve.
• It has no numeric adjustment factors for trial length, price, paywall type, or cohort-age mismatches.
• A single aggregate point does not define a statistical normal band.
• Tailored-paywall ROAS rows are relative uplift, not absolute ROAS benchmarks.

Contributing

Stable taxonomy keys and CSV columns are the public interface. The utilities, health-fitness, photo-video, and education keys are already referenced by companion vertical guides.

Before committing:

node scripts/validate-data.mjs
node scripts/validate-skill.mjs
bash scripts/pack-plugin.sh /tmp/apple-ads-benchmarks.plugin
node scripts/validate-package.mjs /tmp/apple-ads-benchmarks.plugin

The validators reject schema drift, missing segments, source leaks, impossible dimensions, unsupported metrics, broken CPA/CPT/CR relationships, and changes to the published golden anchors.