apple-ads-benchmarks

name

apple-ads-benchmarks

description

Use when someone asks whether an Apple Ads or subscription metric is unusual, what CPA/CPT/TTR/CR was published for a country, App Store genre, or niche, how Apple Ads subscriber economics compare with other paid channels, or what uplift the 2026 report measured for Custom Product Pages and keyword-matched paywalls. Also use when someone says CPI: this report publishes CPA per download, so the denominator must be clarified before comparison. Reads local CSV files and needs no account, CLI, network access, or subscription.

Apple Ads benchmarks for subscription apps

Use Adapty's 2026 report data without turning marginal aggregates into false precision.

Read local CSV files only. Call nothing and require no account.

Answer in the language the user writes in. Keep these instructions and data keys in English.

Start with the metric definition

Use the PDF terminology encoded in data/2026/manifest.json:

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

The report does not provide CPI as spend / installs. If the user says CPI, ask whether their denominator is downloads or installs. Compare only when it is downloads, and call the benchmark CPA. Do not create or store a CPI alias.

The 28.9%-77.9% niche range is cr, from Time Planner to PDF Reader. It is not install-to-trial or trial-to-paid, and it is not a target range.

Select a row without inventing an intersection

1. Read data/2026/taxonomy.csv and resolve the requested country, App Store genre, or sub-niche.
2. Read the relevant numeric file:
   • acquisition.csv for TTR, CR, CPT, or CPA;
   • subscriber-economics.csv for funnel, year-1 LTV, or subscriber-efficiency rows;
   • effects.csv for CPP or tailored-paywall measurements.
3. Match every available dimension: metric, scope, category/sub-niche, geo, traffic source, variant, cohort day, statistic, and period.
4. Require exactly one matching row. Never choose the first of several partial matches.

Country, genre, and sub-niche acquisition rows are separate marginal aggregates. There is no category-by-country matrix. For a question such as "Utilities CPA in Brazil", either:

• show the global Utilities CPA and all-category Brazil CPA as two explicitly separate marginals;
• or say a combined benchmark is unavailable.

Never average, multiply, interpolate, or otherwise synthesize the missing intersection.

Comparability checks depend on the metric

Acquisition: TTR, CR, CPT, CPA

Match the metric definition, scope, geo when using a country row, Apple Ads traffic source, and 2025 period. Do not ask about trial length, price, paywall type, or cohort age: they are not the denominators of these acquisition metrics.

Subscriber funnel

Match the exact step (install_to_trial is not install_to_paid), traffic source, geo/category when present, and relevant paywall/trial context from the user. The report has global channel comparisons, twenty country install-to-paid comparisons, and ten US category medians. It does not have a general category-level trial-to-paid baseline.

LTV and ROAS

Match the time window. year_1_ltv cannot be compared with Day-30 LTV. Tailored-paywall ROAS rows are relative uplift at the listed cohort days, not absolute ROAS benchmarks.

CPP and paywall effects

Report these as measured relative effects from the study population. Never treat an uplift as a guaranteed result or a universal target.

No numeric normalization without coefficients

The report publishes no adjustment coefficients for price, trial length, paywall type, traffic mix, or cohort-age mismatches. If dimensions differ:

• identify the mismatch;
• say the likely direction only when the report supports it;
• do not calculate a normalized value, corrected gap, or "normal band".

If the mismatch can change the conclusion, do not issue a verdict.

Source and sample sentence

Every answer containing a benchmark must state:

• that it is the 2026 vintage covering January-December 2025 campaigns;
• the report-wide sample context: 8,000+ apps and 1,000,000+ ad groups;
• that per-row app counts were not published, when the answer could otherwise imply row-level statistical confidence.

Do not use a fabricated per-row threshold. The published niche eligibility requirements are in references/methodology.md.

Verdict format

When comparison is valid, use four parts:

1. the user's metric and denominator;
2. the exact published aggregate and scope;
3. the difference in points, percent, or dollars, clearly labeled;
4. the practical next check or action.

Say "above/below the published aggregate", not "statistically normal/abnormal". A single aggregate is not a distribution and cannot establish significance.

Refuse or narrow the answer when

• the user's denominator is unknown or differs from the report;
• a requested category-by-country intersection does not exist;
• trial length, price, paywall type, traffic source, or cohort window prevents a valid subscriber comparison and cannot be matched;
• the report contains only a relative lift but the user asks for an absolute baseline;
• the requested category/niche has no taxonomy row;
• the question asks for an individual competitor's data;
• the user asks for a bid derived from a benchmark.

Benchmarks do not set bids. Allowed CPA comes from the app's own unit economics.

References

• references/methodology.md - sample, aggregation, source hierarchy, and limitations.
• references/how-to-compare.md - metric-specific comparison and refusal examples.
• references/taxonomy.md - scopes, stable keys, and non-intersection rule.
• references/behavior-fixtures.md - required behavior for representative prompts.

To act on an account, use the Adapty CLI and the apple-ads skill: https://github.com/adaptyteam/apple-ads-cli

Full report: https://adapty.io/apple-ads-for-subscription-apps/