A revised economics paper tracking 580,000 apps for four years after Apple's tracking prompt finds the shift to paid business models never arrived. Data collection went up instead.

Reinhold Kesler, assistant professor of economics at Heinrich Heine University Düsseldorf and a researcher at the Duesseldorf Institute for Competition Economics, published a substantially expanded version of his study on Apple's App Tracking Transparency framework on July 31, 2026. The 46-page paper, first circulated in April 2022, now carries evidence running four years past the policy's enactment. Its central claim cuts against one of the most widely repeated assumptions in mobile advertising: that restricting cross-app tracking would push app developers off advertising and onto payments.

It did not. According to the paper, the framework raised the likelihood that an app on the Apple App Store charged an upfront price by 0.071 percentage points relative to a control group on Google Play, and the likelihood of offering in-app purchases by 0.104 percentage points. Both estimates are statistically significant. Both are, in relative terms, increases of 0.82 percent and 0.65 percent against baseline shares. Kesler describes them as economically negligible, and the wider set of specifications in the paper does not rescue them.

What developers did change, according to the study, was the data they collect.

A quasi-experiment built on two app stores

Apple enacted App Tracking Transparency with the iOS 14.5 release on April 26, 2021, introducing a one-time prompt asking users to permit tracking across other companies' apps and websites. Refusal blocks access to the IDFA, the device-level identifier that underpinned deterministic attribution and cross-app profiling for the previous decade.

Kesler's identification strategy treats Google Play as the counterfactual. Monthly web crawls covering February to December 2021 produced a balanced panel of 583,834 apps on Apple and 901,182 on Google, with each app store page scraped for pricing, in-app purchase presence, size, description length, website presence, and privacy policy status. Missing observations, caused by crawl failures, account for 1.52 percent of the sample and were imputed.

Baseline monetization differed sharply between the platforms before the policy. On Apple, 11 percent of apps carried an upfront price and 22 percent offered in-app purchases. On Google, the corresponding figures were 6 percent and 13 percent. Fifty-eight percent of the Apple sample were single-homing apps, available on Apple only, identified by cross-referencing app identifiers against listings held by the analytics firm apptopia.

The preferred specifications exclude multi-homing apps. That choice matters: developers publishing on both stores can offset iOS losses with Android revenue or implement pricing changes simultaneously across platforms, either of which would attenuate the estimate. When multi-homing apps are included, the coefficients shrink further while retaining their sign, which the paper reads as direct evidence of spillovers.

Statistically real, economically trivial

The before-and-after specification, run on Apple apps alone, produces increases of 0.088 percentage points for paid apps and 0.108 percentage points for in-app purchases. The difference-in-differences estimates that add Google as a control return 0.071 and 0.104 percentage points. Time dummies tell two different stories underneath those headline numbers. For upfront pricing, the trend across both platforms is negative, with the Apple-specific divergence appearing after the iOS 14.5 update reached near-full adoption. For in-app purchases, the time trend is positive and statistically significant on both platforms, indicating a movement that predates and outlasts the policy.

Kesler flags a methodological problem rather than smoothing it over. An event study interacting monthly indicators with the Apple platform dummy finds pre-treatment coefficients that are insignificant for upfront pricing, satisfying the parallel trends requirement, but negative and statistically significant for in-app purchases. Apple and Google apps were already on divergent paths for in-app payments before the prompt appeared, which weakens the causal reading of that particular outcome.

Price levels moved barely at all. Averaged across every app in the Apple sample, the mean price rose from roughly 0.55 to 0.56 US dollars over the observation window. Conditional on an app carrying a price in at least one month, the average moved from 4.87 to 4.96 US dollars.

Raw counts of adoption and abandonment show the platform asymmetry more plainly than the regression coefficients do. Between the first and last observation, 1,535 Apple apps adopted an upfront price while 1,340 dropped one. On Google, 42 adopted and 452 abandoned. For in-app purchases, Apple recorded 2,574 adoptions against 626 abandonments; Google recorded 3,286 against 1,343. Among Apple apps that adopted upfront pricing, 85 percent were single-homing. Education and Games accounted for most of the adopters.

Where the effect concentrated

If the tracking restriction drove the response, the strongest reactions belong to the apps most exposed to it. The paper tests this two ways.

Interacting the post-period indicator with the share of a developer's portfolio available only on Apple yields coefficients of 0.090 for upfront pricing and 0.215 for in-app purchases. Developers exclusively multi-homing show no effect on paid apps and a slight decline in in-app purchases. Interacting the post-period with a tracking indicator drawn from Apple's privacy labels, which identifies apps that collect data to track users outside the app, produces 0.046 and 0.752.

Robustness tests substitute alternative exposure measures. Where advertising is the stated purpose of data collection, the interaction terms are 0.050 and 0.614. Where apps fall into the ten of twenty-six categories classified as advertising-intensive, upfront pricing rises 0.043 while in-app purchases show no statistically significant difference at all.

The gradient runs in the predicted direction. The absolute magnitudes remain small.

Four years of exit and entry

Kesler recrawled the balanced panel one, two, three and four years after the policy. By March 2025 the Apple panel had fallen to 321,797 apps, meaning roughly 44 percent had left the store. The steepest declines fall between 2021 and 2023, with the largest single drop in 2022, coinciding with Apple's April 2022 policy of removing inactive apps. That confound is acknowledged in the paper rather than argued away.

Two years after the framework took effect, the share of paid apps among survivors had risen from 10.92 to 11.67 percent, a relative increase of 6.9 percent. In-app payments moved from 21.81 to 24.32 percent, up 11.5 percent. By 2025 the paid share had fallen back to 10.38 percent while in-app payments continued climbing to 27.06 percent, a pattern consistent with a broader industry drift toward subscriptions rather than a privacy shock.

Entry data assembled from archived snapshots of the analytics firm 42matters show monthly new listings averaging 39,300 on Apple and 114,800 on Google during 2020, falling to 35,900 and 90,200 in 2021. The Apple decline coincides with full adoption of the iOS version carrying the prompt; the Google decline begins earlier.

A cohort analysis built from 2,186,201 apps, assembled from continuous 2021 crawls plus Apple sitemap retrievals in August 2024 and July 2025, examines monetization by birth month. The share of paid apps declines steadily across cohorts and settles between 2 and 3 percent for the most recent ones. In-app payments begin rising around GDPR enactment and reach nearly 20 percent three years after the tracking prompt. Neither series shows a visible discontinuity at April 2021.

The ad load theory does not hold

One competing explanation is that developers simply showed more advertisements to compensate for weaker targeting. Kesler tested it against user reviews, collecting up to three per app in December 2021 for a total near 600,000, then repeating the crawl in July 2025 for 854,237 reviews.

The share of reviews mentioning advertisements peaks around the implementation date and slips afterward. The share combining an advertisement mention with a one-star rating shows no increase during 2021 and, if anything, a modest decline. Writing on LinkedIn, Kesler summarized the result as offering little evidence that users complained more about advertising after the change.

Maintenance activity offers no stronger signal. Update rates on both platforms rise shortly before the policy and decline gradually afterward, tracking each other closely enough that the paper declines to attribute the drop to Apple specifically. Kesler positions this as more cautious than adjacent literature, noting that Cheyre et al. characterize the magnitude of their own estimated update decline as trivial.

Data collection is where the adjustment happened

Among apps in the balanced panel that survived to 2025, the share disclosing any data collection through Apple's privacy labels rose from 16.9 to 40.9 percent. The share collecting linked data, meaning data tied to an identified user, climbed from 10.4 to 24.1 percent. Increases concentrate in Identifiers, Contact Info, Usage Data and Diagnostics. Health and Fitness and Sensitive Info categories barely moved.

The balanced panel understates the shift by construction, since it consists of older cohorts that survived. In the July 2025 cross section, 25 percent of balanced-panel apps report collecting linked data against 37 percent of the remaining apps, most of which entered after the policy. Privacy label coverage had reached 85 percent by 2025, up from a minority position in 2021, which makes the later cross section the more complete picture.

Sorting that cross section by birth month produces the study's most pointed finding. Apps born before the tracking prompt more often disclose non-linked data collection. Apps entering afterward disclose linked data collection at higher rates, narrowing the gap between the two categories. Decomposing linked disclosures by type shows the increase driven by Contact Info, a category covering names and email addresses.

That combination of data types is well suited to fingerprinting techniques, which can construct persistent user profiles without the IDFA. Kesler's LinkedIn summary put the conclusion directly: "The strongest evidence of adaptation comes from developers' data collection practices."

Why the finding matters for advertisers and publishers

The economic stakes attached to the policy were never small. The paper cites Meta's expectation of roughly 10 billion US dollars in lost advertising revenue in the year following the change, and CMA findings that app developers reported revenue losses around 25 to 30 percent or more. Kraft et al. estimate that the share of trackable users in the United States fell from 72.6 percent to 17.9 percent, with analytics firms reporting 60 to 80 percent of users declining permission. PPC Land has reported the same research line elsewhere, with a weighted average reduction of roughly 45 percentage points in trackable traffic across seven EU markets.

Against a shock of that size, a monetization response of 0.071 percentage points is the point of the paper. Payments, on this evidence, are not a close substitute for advertising. The Apple App Store ecosystem generated close to equal billings in 2025 from in-app advertising at 151 billion US dollars and from digital goods and services at 149 billion, and neither pool absorbed the other.

For media buyers, the practical implication is that iOS inventory did not become materially less ad-funded. It became harder to measure while remaining ad-funded, which is consistent with the attribution workarounds that have accumulated since: the migration to AdAttributionKit, the rebranding of SKAdNetwork and its 77 percent coverage of referral-based App Store conversions, and the consortium investment by Google, Meta, Moloco and Unity into AppsFlyer to keep measurement neutral.

For publishers and regulators, the second-order effect is the sharper one. A restriction aimed at one identifier appears to have redirected effort toward first-party data collection and alternative identification methods that no equivalent consent prompt governs. The paper frames this as a design question rather than a compliance failure, arguing that governance must account for "the broader incentives firms face to collect and exploit user data" and not only for the specific technologies being restricted.

That argument arrives while European authorities are still litigating the framework itself. France's competition authority fined Apple 150 million euros in March 2025 over the prompt's asymmetric design, Germany's Bundeskartellamt reached preliminary findings in February 2025 that the rules may treat third-party developers unfairly, and Italy's competition authority imposed a fine of 98,635,416.67 euros in December 2025. According to Kesler's LinkedIn post, the Italian authority cited the paper in that case.

The consistent regulatory complaint has been that the framework burdens third-party developers disproportionately. Kesler's evidence adds a different measurement to that record: the burden did not change how those developers charge users. It changed what they know about them.

Timeline

Summary

Who: Reinhold Kesler, assistant professor of economics at Heinrich Heine University Düsseldorf and researcher at the Duesseldorf Institute for Competition Economics and ZEW Mannheim, whose earlier version of the work was cited by Italy's competition authority in its case against Apple.

What: A revised 46-page study using difference-in-differences on a panel of more than 580,000 apps, finding that App Tracking Transparency raised paid-app prevalence by 0.071 percentage points and in-app purchase prevalence by 0.104 percentage points, while apps entering after the policy disclose markedly higher rates of linked data collection, particularly contact information.

When: Posted July 31, 2026, extending an analysis first circulated on April 14, 2022, and covering the period from February 2021 through July 2025.

Where: The Apple App Store and Google Play Store, with monthly web-scraped panel data covering both platforms and cohort data drawn from Apple sitemaps.

Why: The result contradicts the expectation that restricting cross-app tracking would push developers toward payment-based business models, and points instead to substitution into alternative data collection, a mechanism that the consent prompt does not govern and that carries direct consequences for measurement, targeting and regulatory design in mobile advertising.