App Tracking Transparency (ATT): The iOS Privacy Change That Reshaped Paid Media
App Tracking Transparency (ATT)
TL;DR: ATT is Apple’s privacy framework (shipped in iOS 14.5 on April 26, 2021) that forces every app to ask permission before tracking a user across other companies’ apps and sites. Most users declined — of those shown the prompt, only about a quarter opted in worldwide (lower in the US). That single change collapsed the per-user signal advertisers relied on, broke deterministic attribution, pushed platforms to modeled conversions, and is the root cause behind the rise of glossary/marketing-mix-modeling, glossary/incrementality-testing, AI broad targeting (glossary/advantage-plus), and the principle that glossary/creative-is-new-targeting. It is the privacy change that actually landed on schedule — unlike Google’s repeatedly-delayed cookie deprecation.
What it means
App Tracking Transparency requires any iOS app to show a standardized, Apple-controlled prompt — “Allow” vs “Ask App Not to Track” — before it can track the user across other companies’ properties. If the user declines, the app loses access to the IDFA (Identifier for Advertisers), the per-device ID that ad networks used to match an ad impression to a downstream conversion.
Two points matter for marketers:
- It’s app-level, on Apple’s mobile OS. ATT governs the IDFA and cross-app tracking on iOS. It is parallel to, but separate from, third-party-cookie deprecation on the web (different company, different mechanism, different timeline — see below). The two are constantly conflated.
- The prompt can’t be gamed. Apps cannot restyle the dialog or pre-empt the “no.” This is why opt-in rates stayed low.
Key dated facts
| Date | Event |
|---|---|
| June 2020 (WWDC) | Apple announces ATT |
| April 26, 2021 | ATT ships in iOS 14.5; enforcement begins — apps must use the ATT framework to access IDFA |
| February 2, 2022 | Meta tells investors ATT will cost it ~$10B in 2022 (see below) |
How much signal was lost
Opt-in rates — mind the denominator. There is no single ATT opt-in number; reported figures span 6–46% because vendors measured different things. The defensible framing:
- Of users actually shown the prompt: roughly ~25% worldwide, ~17% in the US (Flurry’s ~2B-device tracker — vendor).
- Across all iOS users (including those never prompted): low-to-mid teens (~13–15%).
The safe statement: most iOS users declined; of those who saw the prompt, only about a quarter allowed tracking. The wide spread is itself a fact — early single-tracker figures of 6% (US, low-end) and 46% (an outlier) should not be cited as the headline number.
The $10B Meta figure. On Meta’s Q4 2021 earnings call (Feb 2, 2022), CFO David Wehner said the iOS headwind in 2022 was “on the order of $10 billion.” Cite it as a company estimate/forecast (“Meta estimated ~$10B”), not an audited result — Wehner called it a best guess. The same report triggered a record one-day market-cap drop. COO Sheryl Sandberg named the two damage vectors: targeting accuracy fell (cost-per-outcome rose) and measurement got harder (conversions harder to attribute).
How the industry replaced the lost signal
Deterministic, user-level attribution gave way to modeled, aggregate attribution:
- SKAdNetwork (SKAN) — Apple’s privacy-preserving attribution API. Returns campaign-level install/conversion data via delayed postbacks; you learn a campaign drove N conversions in value-bucket X, never which user.
- Meta Aggregated Event Measurement (AEM) — Meta’s response: up to 8 prioritized conversion events, with statistical modeling filling gaps for opted-out users. Conversions surface in Ads Manager as modeled, anonymized data points, not 1:1 matches.
The causal chain (why this page exists)
ATT (+ web cookie pressure) → most users decline → IDFA / cross-app signal collapses → deterministic attribution breaks → platforms shift to modeled conversions (SKAN, AEM) → advertisers can no longer trust last-click, so they lean on MMM, incrementality testing, and first-party data → manual interest/lookalike targeting loses its data foundation, so AI broad targeting wins (Meta Advantage+, Google PMax) → the one input the advertiser still fully controls is creative → “creative is the new targeting.”
Every downstream lever this wiki documents — the glossary/marketing-mix-modeling revival, the demotion of glossary/multi-touch-attribution, the rise of glossary/advantage-plus / glossary/performance-max, the glossary/creative-is-new-targeting shift, and the glossary/meta-andromeda engine — bottoms out here.
ATT vs Google’s third-party cookies — do not conflate
| ATT | Google 3rd-party cookies | |
|---|---|---|
| Company / surface | Apple / iOS apps | Google / Chrome web |
| Mechanism | IDFA opt-in prompt | Removing 3rd-party cookies (Privacy Sandbox) |
| Status | Shipped & enforced Apr 26, 2021 — real, in effect | Repeatedly delayed (2022→25), then walked back: July 22, 2024 Google said it will not deprecate as planned |
The part of the “cookie apocalypse” that genuinely landed on schedule and bit hard was largely ATT, not Chrome.
Key Takeaways
- ATT is the root cause of the post-privacy paid-media playbook — not a footnote, the starting point.
- Most users declined (≈25% opt-in among those prompted, worldwide), gutting per-user ad signal.
- The $10B Meta number is a Feb-2022 company estimate, not an audited result — attribute it precisely.
- Attribution went from deterministic to modeled (SKAN, AEM); last-click stopped being trustworthy.
- ATT actually happened; Chrome’s cookie deprecation didn’t — don’t treat them as the same event.
Related
- glossary/creative-is-new-targeting — the downstream conclusion: with targeting commoditized, creative is the remaining lever
- glossary/multi-touch-attribution — the measurement method ATT broke at the user level
- glossary/marketing-mix-modeling — the privacy-durable measurement that came back because of ATT
- glossary/incrementality-testing — the causal ground-truth layer that replaced trust in attribution
- glossary/advantage-plus — Meta’s AI broad targeting that wins once manual targeting loses its data
- glossary/performance-max — Google’s equivalent automation suite
- glossary/meta-andromeda — the 2026 retrieval engine that makes creative-as-lever engine-enforced
- marketing/marketing-analytics-in-2026 — the post-attribution measurement stack ATT forced into being
Sources
- Apple Developer — User Privacy and Data Use — primary; the ATT framework and prompt rules
- MacRumors — App Tracking Transparency guide — press; mechanics and timeline
- CNBC — iOS 14.5 release date confirmed (Apr 20, 2021) — press; ship date
- Flurry — ATT opt-in rate monthly tracker — vendor; opt-in rates (note denominator)
- CNBC — Meta says Apple privacy change will cost $10B (Feb 2, 2022) — press reporting Meta’s own estimate
- RevenueCat — iOS attribution: SKAN / AEM / probabilistic — vendor/technical; modeled-attribution mechanics
- DigitalCommerce360 — Google ends cookie deprecation plan (Jul 24, 2024) — press; the ATT-vs-cookies distinction
Do-not-cite: the 46% worldwide opt-in figure (outlier vs ~25% prompted); the bare “ATT cost Meta $10B” framed as fact (it’s a company forecast).