Meta separated click-through and engage-through attribution in March 2026. ROAS didn't actually fall — the metric definition changed. Here's what the numbers now mean, why video ROAS is likely inflated by 15–25%, and how to rebuild your benchmarks.
In March 2026, Meta changed how it counts clicks. The result was that ROAS appeared to collapse overnight for a lot of accounts, CPA appeared to spike, and conversion volume appeared to fall — none of which reflected a real change in how ads were actually performing.
What changed was the metric definition, not the performance. But that distinction took weeks to understand, and many accounts made budget cuts or creative pivots in response to a measurement artifact rather than a real signal.
Here's the full picture, including a complicating factor that's running in the opposite direction — video ROAS that's probably inflated right now.
Before March 2026, Meta's default attribution model was 7-day click, 1-day view. Any conversion that occurred within 7 days of someone clicking your ad — regardless of what the click was — counted as a conversion attributed to that ad.
"Any click" included:
This was always a messy attribution model. A user who liked your ad and then converted three days later through a Google search was being counted as a Meta conversion. But it was the model everyone had calibrated their ROAS targets around.
Meta split conversions into two categories with separate attribution windows:
Click-through conversions: Only conversions from actual link clicks — clicks that navigate to your website, app, lead form, or shop. Attribution window: 7 days.
Engage-through conversions: Conversions from all other interactions — likes, comments, shares, saves, post clicks that don't navigate away. Attribution window: 1 day.
These now appear as separate line items in Ads Manager: purchases_click_through and purchases_engage_through. The total is the sum of both.
The reclassification itself doesn't change what actually happened. Users who converted still converted. The ad still drove the same behavior. What changed is how those conversions are categorized and — critically — the attribution window applied to each.
For remarketing campaigns, the impact was largest. Remarketing relies heavily on users who see an ad, engage with it in some way (a like, a save), and then convert a few days later. Under the old model, that conversion had a 7-day window to be attributed. Under the new model, those same conversions fall under engage-through attribution with a 1-day window. Conversions that happened on day 2 through day 7 simply stopped appearing in reports.
The fix is mechanical: don't compare click-through conversions to your pre-March baseline. Compare total conversions (click-through + engage-through) to your pre-March baseline.
Pull a report with these columns side by side:
One agency that tracked accounts through the entire March rollout — 340 accounts across e-commerce, lead gen, and retail — found that when click-through and engage-through are combined, most accounts land within 8–12% of where their total conversions sat before March. The rest of the apparent gap was the attribution window change for engage-through.
If your combined total is significantly below your pre-March baseline, that's a real performance signal worth investigating. If the gap closes when you add both categories, the drop was measurement, not performance.
Here's where it gets less clean. At the same time Meta clarified the click attribution model, it expanded its engaged-view attribution — conversions that occur after someone watches a video ad without clicking.
The definition of an "engaged view" was quietly changed: it now requires watching just 5 seconds of a video ad (down from 10 seconds) or 97% of a short video, whichever comes first. Any conversion within 1 day of that view is attributed.
Five seconds is a very low bar. Someone who scrolled past your video in their feed and watched 5 seconds before continuing counts as an engaged view. If they convert within 24 hours — through any channel — that conversion appears in your Ads Manager as an engaged-view attribution.
Accounts running video creative heavily (Reels, Stories, in-feed video) are seeing 15–25% inflation in reported conversions from this change, according to AdBeacon's analysis of account cohorts before and after the threshold change. The conversions are real — users who were exposed to the video did convert — but attributing them to the video ad specifically, when the user spent 5 seconds watching and then continued scrolling, is optimistic.
This creates an ironic situation in some accounts: click-through ROAS looks worse than before March (because the window is now stricter), while video ROAS looks better (because the engaged-view threshold dropped). Both movements are at least partly artifacts of measurement changes rather than real performance shifts.
Remarketing accounts were hit harder than prospecting by the attribution change, for a specific reason that's worth understanding.
In prospecting, users who click a link are the primary conversion driver. The 7-day click-through window still applies to those users, so prospecting performance metrics are relatively stable post-March.
In remarketing, you're targeting users who already know your brand. The conversion behavior is different: they often see the ad, engage with it (a like, a reminder-type save), and then convert later through a different channel — sometimes typing your URL directly, sometimes through Google. Under the old model, those conversions got attributed to Meta. Under the new model, they're in the 1-day engage-through bucket, so any conversion after day 1 disappears from Meta's reporting.
This doesn't mean remarketing stopped working. It means Meta is now reporting a more conservative view of what remarketing is directly responsible for. If you cut remarketing spend in response to the apparent ROAS drop in March, it's worth reviewing whether overall revenue from that audience segment actually declined or whether you just stopped seeing it credited to Meta.
Three concrete things to do:
1. Reset your ROAS floors. Your pre-March ROAS targets were calibrated to the old click definition. A direct comparison isn't valid. Run a 60-day analysis of post-March performance using click-through + engage-through combined, and set new floors from that baseline. Don't average pre-March and post-March data — use only the post-March period.
2. Separate your video and non-video campaign performance. Video campaigns are now running with engaged-view attribution inflating their numbers. Static and carousel ads are not. If you're averaging across creative types, your overall ROAS metric is a blend of a conservative model (click-through) and a generous one (engaged-view). Segment your reporting by creative type to see what's actually happening in each.
3. Cross-reference with MTA or revenue data. Meta's Ads Manager is increasingly one signal among several rather than a source of truth. Run an incremental hold-out test (Meta offers this through its Conversion Lift tool) to understand what percentage of your Meta-attributed conversions are incremental — conversions that wouldn't have happened without the ad exposure. The attribution model tells you who converted after seeing your ad; incremental testing tells you whether the ad caused the conversion.
One more complication that most coverage of the March update ignores: if you're running Advantage+ campaigns, Meta's AI delivery system now makes attribution even harder to audit because it's optimizing across placements in real time.
When Advantage+ places an ad in Reels, applies an engaged-view attribution window, and then attributes a conversion, you're looking at: automated placement selection + creative optimization + a 5-second view threshold + engaged-view attribution, all bundled into a single reported metric. The performance number that comes out the other end is real in the sense that conversions happened, but it's very difficult to decompose into what the ad itself caused versus what the attribution model included.
The brands handling this best are running Advantage+ for prospecting (where the automation advantage is real and measurable through hold-out tests) while maintaining manual placement campaigns for remarketing where they want clean attribution data.
Meta's March 2026 attribution changes had three effects that interacted in confusing ways:
If you're making budget decisions based on the apparent ROAS change in March, the most important thing to verify is whether you're looking at click-through in isolation or the combined total. Almost all of the accounts that cut spend in response to the apparent ROAS drop made that decision based on click-through alone.
The attribution model changed. The performance, for most accounts, didn't.
MarketerTools
Marketing Practitioners
Written by the MarketerTools team — practitioners who build and use tools for marketers every day.
Put this into practice
Use the free tools on MarketerTools to apply what you just read.
Browse all tools →