Meta’s March 2026 updates have fundamentally changed how link clicks are counted and introduced "Engage-Through" logic that can skew your ROAS. If you are seeing a discrepancy between Meta and your back-end, it’s likely hidden in these five settings.
When you set up a website conversion campaign in Meta today, the default attribution is:
- Click-through: 7 days
- Engage-through: 1 day
- View-through: 1-day
- Attribution model: Standard
- Conversion count: All Conversions
That's five distinct layers, and each one affects both what gets reported and how Meta delivers your ads. Miss any of them, and you're misreading your results.
Layer 1: Click-Through Attribution
In March 2026, Meta restricted "click-through" to include only actual link clicks, excluding previous "social" clicks such as likes or comments. While a "link click" still includes interactions like click-to-call or Shops, it primarily signifies a site landing.
Your options now are 7-day or 1-day click-through windows — the old 28-day window is no longer available as an optimization setting.
For purchases, try sticking with 7 days, as it reflects the reality that people often research before buying.
For free offers that don't require much deliberation, 1-day is more accurate. Just know that switching from 7-day to 1-day will drop your reported volume, so factor that in before you make the change. Also, you can still view 28-day data using the Compare Attribution Settings feature for historical context.
Layer 2: Engage-Through Attribution
This replaced "engaged-view" in March 2026, and it's a meaningful expansion.
The old engaged-view only counted video views of 5+ seconds followed by a conversion within 1 day. The new engage-through captures all of that, plus any non-link click on your ad — likes, shares, saves, reactions, comments — if that interaction was followed by a conversion within 1 day.
It's on by default with a 1-day window, and you can turn it off by setting it to "none."
When to Keep it On vs. Turn it Off:
Keep this active for purchase campaigns, as social interactions signal genuine intent. However, consider disabling it for lead gen or remarketing, where a simple "like" followed by a conversion is often a stretch of the ad’s actual influence.
Layer 3: View-Through Attribution
This credits a conversion when a user is served an ad and converts within one day without clicking or engaging. This is the primary source of inflated results, particularly in remarketing, where Meta often takes credit for conversions actually driven by email or SMS.
Like engage-through attribution, view-through is turned on by default but can be set to “none.” Turn off 1-day view-through for all non-purchase events and always disable it for remarketing to avoid misleading, coincidental data.
Layer 4: Attribution Model
This is a newer setting that most people haven't touched yet.
Standard attribution optimizes delivery based on your attribution windows and engagement types. Incremental attribution uses machine learning to predict whether a conversion was actually caused by your ad — and only counts those.
Incremental attribution rolled out in 2025. When you select it, the model takes over that logic entirely, and you lose the ability to manually edit your attribution settings.
In theory, incremental is more accurate. It should filter out conversions that would have happened anyway and focus your optimization signal on the ones your ads genuinely drove.
In practice, many advertisers haven't seen a dramatic difference in results — likely because careful manual attribution management already handles most of the noise.
Practically speaking, you should only toggle "Incremental" if you have high-budget campaigns with strong conversion volume (ideally 50+ conversions per week).
For smaller accounts, however, Incremental can lead to a ‘Death Spiral’ — the AI stops reporting "non-proven" sales, which lowers the signal, which causes the algorithm to stop spending on potentially good traffic. If you're already struggling to exit the learning phase, don't add another constraint that limits your signal.
Layer 5: Conversion Count — All Conversions vs. First Conversion
This one trips people up constantly and is a big reason why Meta numbers don't match your back-end or third-party tools.
By default, Meta counts all conversions — meaning if the same person buys from you twice within your attribution window, both purchases get reported. That can make your results look better than they are when you're comparing to systems that only log one transaction per customer.
First conversion changes this so Meta only counts the initial action per person per attribution window. It was reporting-only when it launched in 2024, but Meta made it an actual optimization setting in late 2025 — meaning the algorithm will now optimize toward first conversions only if you enable it.
The catch with using it as an optimization lever: when Meta only counts first conversions, it treats a customer who buys once the same as a customer who buys twice. You're actively telling Meta not to learn from repeat purchase behavior — and that's a signal you want it to have for most ecom campaigns.
First conversion is more valuable as a reporting lens than as an optimization input. The limited use cases where it makes sense as an optimization setting are lead gen (to avoid rewarding duplicate form fills) or engagement events where repeat actions would inflate results.
Key Takeaways for 5-Layer Attribution
Your Meta reporting now combines five settings, each affecting both what you see and how the algorithm delivers.
If you and a client or teammate are looking at different reports with different settings active, you're not even looking at the same data. Get aligned on a standard configuration, document it, and keep your benchmarks consistent going forward.
Finally, don't customize attribution settings just because you can. The defaults exist because they work in most situations. Make changes when you have a specific problem to solve — inflated results on remarketing, lead gen campaigns where view-through conversions are clearly noise, etc. Otherwise, leave them alone.
Related Posts:
