Meta Measurement After iOS Privacy: Pixel, CAPI and Attribution

Apple's privacy change, known as app tracking transparency, permanently reshaped Meta advertising. Since users started being asked for explicit permission for cross-app tracking, browser and app-side measurement has no longer been complete. That does not mean the ads stopped working; it means how you set up measurement matters far more than it used to. This article covers what still holds today, what no longer does, and the correct order of setup.

What actually changed

Previously, when a user saw an ad and later came to your site and purchased, that journey could largely be tracked through a device identifier. Once the privacy prompt appeared, a significant share of users declined. As a result Meta can no longer observe some conversions directly and estimates them through modelling. The conversion count in your report is therefore not the sale itself, but the measurable and modelled portion of it. Understanding that distinction is the first condition for reading the number correctly.

Three steps that still hold

First, domain verification. Without verifying your domain in Business Manager, what you can do on the web conversions side stays limited. It is a one-off step, usually finished in minutes with a verification tag or a DNS record, but every later step runs incomplete when it is skipped.

Second, the Conversions API. The pixel runs in the browser and is affected by ad blockers and privacy settings. The Conversions API sends the event straight from your server, where those obstacles do not apply. The two are installed together and neither replaces the other. To avoid counting the same event twice, each event carries a shared identifier that Meta uses to deduplicate. We cover pixel installation step by step in a separate article.Meta pixel kurulumu

Third, a real conversion test before going live. Using the test tool inside Events Manager, visit the site and complete the form or the purchase flow, and watch the event arrive from both the browser and the server. Decide based on seeing the event actually flow, not on the setup looking correct.

No longer valid: the eight-event prioritisation

This is the outdated instruction we encounter most often. In the early period after the iOS change, Meta allowed only eight web events to be prioritised per domain, and you had to rank them by hand. That restriction was removed and the tab was taken out of the Events Manager interface; Meta now collects eligible events itself and there is no event count limit on the web side. Despite that, most guides online still explain how to rank eight events.

The check that replaced it is different and more important: are your events being sent completely, and are the event names consistent? If the purchase event carries one name on the web and another on the server, Meta cannot match them as the same event. Use identical naming across web, server and app. The privacy protocol itself is still very much alive; what disappeared was the manual eight-event configuration.

How to read modelled conversions

Part of the conversions you see in the report were observed directly and part were modelled. There is no button separating the two, and you should not expect one. The practical consequence is this: on small accounts and low-volume campaigns the modelled share is relatively high, so judging by daily fluctuation misleads you. Rather than shutting a campaign after a three-day dip, look at weekly totals. Modelling behaves far more stably over a longer window.

For the same reason it is worth keeping an independent count on your own side. Counting form submissions in your own database or spreadsheet is the cheapest way to compare what the platform says with what actually happened. Treating the ad platform as the single source of truth is a risky habit in a period when measurement itself has weakened.

Whose job is the server-side setup

Because the Conversions API sends events from your server, this step usually touches the development side. On a hosted platform a plugin or integration is often enough. On a custom-built site a developer has to place the point from which the event is sent. This is the most frequently postponed step of measurement setup, and while it is postponed your data stays incomplete. Finish it before launching campaigns; when it is added later, the earlier period's data does not come back.

If you hold a customer list, uploading it is also part of measurement. It becomes both a lookalike source and an exclusion list, and both of those protect your budget directly.

Attribution windows changed in 2026

This is the change that caused alarm on many accounts last year and is frequently misread. In January 2026 the seven-day and twenty-eight-day view-through windows were removed. Some accounts lost a significant share of their reported conversions overnight while their actual sales had not moved at all. A second change followed in March 2026: a click now means a genuine link click, interactions such as likes and comments moved to a separate category, and the view threshold was lowered.

The practical rule that follows is clear: compare periods only under the same window. Comparing this June with last June and declaring performance down is an invalid comparison, because the window itself changed. When reporting on older periods, note the break; a chart presented without that note creates a false impression.

The app side is a separate topic and should not be mixed with the web side. For businesses with a mobile app the event flow is built through a different path and distinct constraints still apply there. Carrying an instruction you read for web conversions into an app campaign produces a setup that does not work. Plan the two sides separately.

How much Meta and GA4 divergence is normal

The two systems use different attribution logic, so the numbers never match exactly. Meta credits the person who saw the ad and later converted; GA4 weights the last interaction. Some gap is expected. Our threshold is this: if the gap exceeds thirty percent, we do not accept it as natural and we audit the setup. The cause is usually in the installation rather than in attribution: a missing event, double counting, or an incorrect source tag.

The order of setup

The order runs as follows: domain verification, a map of conversion points, pixel and Conversions API installation, event name consistency, a live test, and only then the campaign. Breaking that order and launching the campaign first wastes the opening days of budget, because the system has no signal to optimise against. We covered which level to set the budget at in a separate article.bütçe modeli yazımız

Our main guide collects the full setup, the account audit checklist and the campaign structure.Meta Ads rehberimiz

If you would rather work with a team that runs both the measurement setup and the campaign management that follows, we are here.Meta reklam yönetimi