Meta changed attribution twice in 2026, and the two changes compound. If your reported conversions fell this year and you changed nothing, the cause is more likely a definition than a decline.
Engage-through attribution is the Meta attribution category, introduced on March 3, 2026, that credits conversions following an ad interaction that is not a link click: likes, shares, saves, comments, profile visits and video views of 5 seconds or more. It is on by default with a 1-day window, and it holds conversions that were counted as click-through before March.
The two changes, in order:
- 1.January 12, 2026. The 7-day view and 28-day view windows were removed from the Ads Insights API. Reporting that read those windows lost the conversions inside them.
- 2.March 3, 2026. Click-through attribution was narrowed to link clicks only. Every other interaction moved into engage-through, and the video threshold for an engaged view dropped from 10 seconds to 5.
The default setting for campaigns optimizing toward website conversions is now 7-day click, 1-day engage-through, 1-day view.
What each attribution type counts now
| Type | What triggers it | Windows available |
|---|---|---|
| Click-through | A link click that sends the person to a website, app, lead form, Messenger or other destination | 1 day or 7 days |
| Engage-through | A like, share, save, comment, profile visit, or a video view of 5 seconds or more | 1 day only |
| View-through | An impression with no interaction | 1 day, or off |
Two details decide whether your reporting is still telling you what you think it is.
Engage-through is not the old social-click bucket renamed. It absorbed the previous engaged-view video metric as well, and lowered its threshold from 10 seconds to 5. A video-heavy account can see engage-through grow without any change in behaviour, simply because more views now clear the bar.
Engage-through is on by default. Many advertisers who believe they are reading a clean click-based number are reading click plus engage plus view. Equally, many custom column sets built before March show only click-through, and their owners are reading a narrower number than the platform is optimizing on.
Three effects that get confused with each other
| Effect | Date | What happens to the numbers | Accounts that feel it most |
|---|---|---|---|
| Long view windows removed | Jan 12, 2026 | Reported conversions fall | Accounts that reported on 7-day or 28-day view |
| Click narrowed to link clicks | Mar 3, 2026 | Click-through falls, engage-through rises, total roughly flat | High-engagement creative, awareness-heavy spend |
| Video threshold moved to 5 seconds | Mar 3, 2026 | Engage-through rises | Video-heavy accounts |
The first is a real loss of reported credit. Third-party analyses put it as high as 30 to 40 percent of conversions for accounts that leaned on the 8-to-28-day view window, and near zero for accounts that never used long view windows. The second is a reclassification, not a loss: conversions moved from one column to another. The third inflates a column rather than deflating one.
That distinction is the practical point of this article. A reclassification calls for a reporting fix. A real decline calls for a campaign fix. Responding to the first with the second — pausing ads, cutting budget, rebuilding creative — is how accounts get damaged by their own dashboards.
A worked reconciliation
Worked example. These are model inputs chosen to demonstrate the arithmetic, not AdRiseLab customer data and not a benchmark.
An account reports 1,000 attributed conversions in February on a 7-day click, 1-day view setting, displayed as one column. In April, on the same spend and the same creative, a custom column set built in January reads 780.
| Column | February | April |
|---|---|---|
| Click-through | 1,000 (click and view combined) | 780 |
| Engage-through | did not exist | 190 |
| View-through | inside the 1,000 | 35 |
| Total attributed | 1,000 | 1,005 |
Read only the click-through column and performance fell 22 percent. Read all three and the total rose by 5 conversions, half a percent. Nothing happened to the business. 190 conversions were refiled as engage-through, and 35 view-through conversions that the old single column had folded in are now shown separately.
Check in this order, because each step can make the next one unnecessary:
- 1.Confirm the attribution setting on the report, not on the campaign. They can differ, and the report is what you are reading.
- 2.Add the engage-through column explicitly. It is frequently missing from custom column sets built before March 2026.
- 3.Compare totals across the change date, not the click column. Only the total is comparable across March 3.
- 4.Only then look for a real performance change — in cost per result, conversion rate on site, and your own order count.
Rebuilding reporting you can compare
Annotate both dates. Any chart that crosses January 12 or March 3, 2026 contains two definitions. Mark both in whatever tool you report from. An unannotated year-over-year comparison across these dates is not a comparison.
Choose one decision metric and write it down. For most direct-response accounts the defensible choice is 7-day click-through, because it is the narrowest and the most stable from here forward. Engage-through is useful context and a poor thing to optimize a budget against.
Stop treating Meta's attributed number as revenue. It is a platform-modelled estimate of credit under a platform-defined window. Your order ledger is a count of orders. They answer different questions, and they are more useful side by side than merged into one figure. This is the design choice behind AdRiseLab Revenue: it keeps the website order ledger and Meta's reported conversions as separate measurements, marks Shopify and server-signed orders as verified and browser-reported orders as unverified, and shows orders that never passed through the storefront as revenue credited to no campaign.
Improve the signal before re-tuning the campaign. If click-through is now the number that matters, the quality of click identification matters more than it did. That is a Conversions API and event-matching question, covered in Meta Pixel vs Conversions API.
What not to conclude
Engage-through conversions are not fake. A person who saved an ad on Tuesday and bought on Wednesday is a real customer. The change is about which bucket the platform files them in, not about whether they exist.
A narrower definition does not make Meta's numbers causal. It makes them narrower and more consistent. Every attribution setting is a rule for assigning credit; none establishes that the ad caused the sale. That requires a holdout or a lift test, which is a different exercise with a different cost.
Historical data cannot be restated to the new definitions. The interactions were not bucketed that way when they happened. Annotate instead of recomputing.
A drop that survives all four checks is real. Reclassification explains a falling click column with a flat total. It does not explain a falling total, and it does not explain a fall in your own order count. If those moved, go to the campaign — why Meta ROAS drops splits a decline into CPM, click-through rate, conversion rate and order value so you can see which one moved.
Sources
Jon Loomer: How Meta ads attribution works in 2026 for the settings and windows. Dataslayer: what engage-through attribution is for the March 3 change and the default setting, and Dataslayer: the January 2026 view-window removal for the API change and the reported impact range. Dates were checked against these sources on September 27, 2026.
Related Reading
What is a good ROAS for Facebook ads covers what the number you are now reporting should be compared against. The Meta ads audit framework puts attribution settings inside a wider account review, and the ROI calculator turns reported conversions into a margin question. For how hook and hold rate were affected by the 5-second threshold, see what is a good hook rate.