When ATT rolled out, the predictions were everywhere. Apple was ending mobile advertising. Apple was ending Facebook. Apple was overstating its privacy benefits. ATT would never get widespread adoption because users would all opt in. ATT would have minimal impact because everyone was already using probabilistic methods. None of the predictions were exactly right and most were significantly wrong.
This piece looks at what actually happened across three and a half years and what we can now say with reasonable confidence about the impact.
Opt-in rates settled lower than the optimists predicted
Before ATT launched, some industry voices predicted opt-in rates would be high — sixty, seventy percent in some categories — because users would understand the value exchange and choose to share data with apps they liked.
The actual opt-in rates landed much lower. Industry-wide rates have stabilized in the twenty to thirty percent range, with significant variation by app category. Gaming apps tend to see lower rates. Social apps see slightly higher. Utility apps vary widely.
The variation by category is itself informative. Apps where users perceive direct value from data sharing — finding friends, personalized recommendations they value — see higher rates. Apps where the value exchange is less visible see lower rates. The variation is roughly what you would predict from a user-rationality model but the levels are uniformly lower than the optimists predicted.
The Facebook prediction was directionally right but the magnitude varied
Facebook's public guidance after ATT launched suggested ten billion dollars of revenue impact. The actual impact has been hard to isolate because so much else has happened — economic conditions, competition from TikTok, AI investment cycles, regulatory headwinds. But the directional impact was real.
What changed for Facebook specifically was less the absolute volume of advertising and more the loss of high-precision targeting and conversion measurement on iOS users. The platform adapted by leaning more heavily on its own modeled conversions and on broader audience targeting rather than the high-precision custom audiences that had been the core product.
The adaptation has been partial. Facebook performance for mobile app advertisers on iOS is meaningfully worse than it was pre-ATT. Performance for the same advertisers on Android is generally fine. The split between the two operating systems' performance is one of the most legible effects of ATT and has not gone away.
Smaller advertisers got hit harder than the big ones predicted
The pre-ATT discussion sometimes treated the change as primarily a problem for Facebook and Google. The actual distribution of impact was more complicated.
The largest advertisers had the resources to invest in alternative measurement approaches, build internal data infrastructure, run incrementality testing, hire dedicated post-IDFA measurement specialists. They got hit but they had buffers.
Smaller advertisers had less buffer. The teams running mid-budget UA campaigns without dedicated measurement infrastructure took the worst of it. Their attribution surface narrowed. Their ROAS reporting got noisier. Their ability to optimize creative and audience tightened.
The cumulative effect was to widen the operational gap between large and small mobile advertisers in ways that were not the explicit policy intent of ATT but are real.
Probabilistic methods got more constrained than people initially thought
For a while after ATT, the assumption was that fingerprinting and probabilistic attribution would compensate for the loss of IDFA. Apple has continued to tighten the constraints on those methods, both through enforcement and through technical changes that make them less reliable.
The current state is that probabilistic attribution at the user level is largely gone and is unlikely to come back. What remains is probabilistic attribution at the campaign and channel level, which is useful but is a different methodology with different operational characteristics.
The teams that built business processes around the assumption that fingerprinting would persist as a usable signal had to reconfigure their measurement when it became clear that signal was unreliable. The teams that planned for the post-fingerprint reality had a smoother transition.
SKAdNetwork actually became usable, eventually
SKAdNetwork was widely mocked when it launched as inadequate for serious campaign measurement. The mockery was deserved at the time. SKAN 2 and 3 had real problems with conversion bucket constraints, postback timing, and privacy threshold suppression that made the data difficult to use.
SKAN 4 is meaningfully better. The expanded postback windows, the more flexible conversion value configuration, and the better privacy threshold handling all add up to a system that can actually carry real measurement load if configured competently.
It is still not as good as the IDFA-based measurement it replaced. It is good enough that teams running serious UA can use it as one of several signals in their measurement stack. The "SKAN is broken" framing that was common in 2021 and 2022 is mostly out of date.
The privacy benefit was real but harder to measure than expected
The fundamental policy claim of ATT was that user privacy would benefit from forcing explicit consent for cross-app tracking. Whether this benefit has materialized is harder to measure than the industry impact.
The mechanical change is real. Cross-app data flows that previously happened by default now happen only with explicit user consent. The percentage of users whose cross-app data is being collected has dropped from approximately one hundred percent to approximately twenty-five percent. That is a substantial change.
Whether that change has produced commensurate benefits in user experience, reduced ad creep, or other privacy-positive outcomes is harder to assess. The benefits are diffuse and individually small even when the aggregate is meaningful. Most users probably do not notice ATT's downstream effects on their daily app experience even when those effects exist.
What I think the next phase looks like
Apple has continued to tighten the privacy framework on iOS in incremental ways. The next phase will probably include further restrictions on cross-app fingerprinting, additional consent gates for various data flows, and continued evolution of SKAdNetwork.
Google's Privacy Sandbox for Android represents a different approach to similar objectives — building privacy-preserving measurement APIs rather than restricting existing methods. Whether it lands in a usable state and whether mobile advertisers actually adopt it remain open questions.
The cross-platform divergence between iOS privacy posture and Android privacy posture has been one of the lasting effects of ATT. Cross-platform mobile advertisers now operate under meaningfully different measurement frameworks on the two platforms. This is operationally annoying but is now the steady state.
The mobile measurement landscape is in a more stable place than it was three and a half years ago. The teams that adapted competently are running viable practices. The teams that are still hoping for the pre-ATT regime to come back are operating on outdated expectations. ATT changed the floor of mobile measurement; that change is now permanent.
