Meta incremental attribution or 7-day click? What to optimise to and what to report

KEY TAKEAWAYS

The optimisation setting and the reported number are separate choices. Decide them separately.
Haus experiments from July 2025 to June 2026 put incremental attribution at 1.26x the incremental return of standard settings, 1.38x for DTC-only brands and 1.02x for omnichannel brands.
Incremental attribution is a model's prediction, not a holdout. Check it against a real experiment before trusting the column.
Any calibration factor measured before March 2026 is stale, because the definition of a click changed.

Meta's attribution setting does two jobs at once. It tells the delivery system which conversions to chase, and it decides which conversions appear in your report. Most advice treats that as one choice. It is two, and in late 2026 the evidence points to different answers for each. This article covers both, with a decision table and a reporting method.

What the attribution setting controls in October 2026

Every conversion ad set has an attribution model. There are two.

Standard counts conversions that fall inside time windows after an interaction. As Jon Loomer documented in March 2026, the default is 7-day click-through, 1-day engage-through and 1-day view-through. You can shorten the click window to 1 day or switch the other two off.

Incremental attribution drops the windows. Meta's help documentation, as quoted by Social Media Today in September 2025, says it optimises delivery using "models that predict whether a conversion is caused by an ad". Loomer's April 2025 walkthrough noted that once you select it, the click, engage and view windows can no longer be edited.

Whichever model you pick, delivery learns from the conversions that model counts. That is the first decision. The second is what you show the business, and Ads Manager lets you view other settings side by side without changing delivery.

The yardstick moved three times in 2026

Before comparing anything, note that Meta changed its own measuring stick this year.

January 12, 2026. The 7-day and 28-day view-through windows stopped returning data in the Ads Insights API, according to PPC Land's report on Meta's announcement.
March 2026. Click-through attribution began to require an actual link click. Likes, comments, shares and saves moved into the renamed engage-through category, which has a 1-day window. The video threshold for an engaged view dropped from 10 seconds to 5. Search Engine Land reported the announcement on March 3.
June 22, 2026. Meta replaced Nielsen's Designated Market Areas with Comscore Markets as its geographic standard, according to LiftLab. That matters if you run geo holdout tests.

The March change was the big one. Common Thread Collective compared 50 high-spend accounts, with $29.5 million in combined spend, across the 14 days before and after March 18. Reported 7-day click ROAS fell 13%. Sales did not change. The definition did.

The practical consequence: a year-on-year ROAS comparison that crosses March 2026 compares two different metrics. So does any multiplier you calculated before then.

Does incremental attribution drive more real sales?

The sources disagree, and the disagreement is mostly about dates.

Meta says yes. In a January 29, 2026 post, it said its fourth-quarter 2025 model rollout produced a 24% increase in incremental conversions against its standard model. That is Meta grading its own product.

Early independent results said no. Haus, which runs geo holdout experiments for advertisers, reported in July 2025 that incremental attribution beat standard settings in only 43% of its head-to-head tests, on what it called a limited sample. In December 2025 the agency Genie Goals published a single A/B test in which Meta-reported sales fell 73% and cost per acquisition rose 253%. It stopped using the setting.

The newest independent data says yes, with conditions. On July 16, 2026, Haus published an update that divides incremental return under each setting. For tests from July 2024 to June 2025 the ratio was 0.80x, meaning standard won. For July 2025 to June 2026 it was 1.26x, with a stated range of 1.11x to 1.43x. Haus did not publish the number of tests, and it cautions that once spend is controlled for, the improvement cannot be cleanly separated from noise at this sample size.

The split inside that result matters more than the headline. Brands that sell only through their own site saw 1.38x. Omnichannel brands saw 1.02x, which is parity. That fits an earlier Haus finding from 640 experiments: for omnichannel brands, 32% of Meta's effect landed in channels such as retail and marketplaces. Meta's model cannot learn from sales it never sees.

What the evidence supports: the model improved between 2025 and 2026, so older verdicts are out of date. It now looks like a reasonable optimisation setting for a brand whose sales all pass through its own pixel and Conversions API. It is not yet a proven gain for anyone else.

One caution on reading your own test. Reported conversions fall under incremental attribution by design. A drop in the Ads Manager total tells you nothing. Judge the test on revenue measured outside Meta.

Can you report the incremental number as fact?

No. Many guides describe incremental attribution as a holdout test running inside your campaign. Meta's own wording describes a prediction model. Those are different things.

A holdout withholds ads from a random group and compares outcomes. A model estimates what would have happened. Brett Gordon, Robert Moakler and Florian Zettelmeyer examined 663 large Facebook experiments and found that even sophisticated models with more than 5,000 user-level features missed badly. For lower-funnel outcomes the median experimental lift was 5%. The better of two modelling methods estimated 24%.

The same authors later showed that a model trained on experiments does far better. Across 2,226 Meta experiments, their approach explained 88% of the variation in incremental conversions per dollar, against 19% for 7-day last-click attribution. Meta has not said its product uses this method, so treat the link as an inference. The lesson holds either way: a prediction calibrated on many advertisers' experiments is an average. Your account may sit far from it.

Seer Interactive's April 2025 test across six accounts and $1.05 million in spend makes the same point from the other side. Meta's setting judged 87% of conversions incremental. A comparison in GA4 suggested 67%. Neither figure is an experiment.

There is a second disagreement here. Haus found that 7-day click attribution under-reported true incremental revenue by 15% on average for DTC-only brands. The academic work and Seer's test point to over-reporting. Both can be true. Click-only numbers leave out view-driven and off-site sales, so they run low for large prospecting accounts. Default settings with view-through, and accounts heavy on remarketing, run high. And since March 2026 the click number is narrower still, which is why Common Thread Collective raised its own multiplier for 7-day click from 1.2x to 1.38x.

Which setting to optimise to: a decision table

Your situation	Optimise to	Why
Sales only on your own site, steady volume, mostly prospecting	Test incremental attribution in an equal-budget cell against standard	Latest Haus data shows 1.38x for DTC-only brands
Meaningful retail, marketplace or offline sales	Standard, 7-day click with default engage and view	Parity at 1.02x, and the model cannot see off-site sales
Leads, sign-ups or other free conversions	Standard, consider 1-day click and view-through off	Loomer's guidance for low-commitment events, where views inflate most
Remarketing-heavy account	Standard with view-through off	Loomer expects the March 2026 changes to cut remarketing credit hardest
Too little volume to split a test	Defaults, unchanged	No way to verify a switch, and Loomer's rule is to keep defaults unless there is a specific problem

If you do test, keep the two cells identical except for the attribution model, and decide in advance which outside revenue figure settles it.

What to report: a five-step method

Mark the regime changes. Annotate January 12, the March click change (March 18 in Common Thread Collective's data) and June 22, 2026 on every dashboard. Do not compare periods across them without a note.
Break results apart. Use the breakdown by attribution setting. Ads Manager splits results into 1-day click, 2 to 7 day click, 1-day engage and 1-day view. Report click-through as the headline platform number. Show the rest as context.
Add the incremental column. Use Compare attribution settings to place Meta's incremental figure next to standard. Label it as modelled.
Run one real experiment. A Meta conversion lift test or a geo holdout. Haus notes that lift tests need at least seven days and at least 10% of the audience in each cell. Those are minimums. Haus's own geo tests ran 18.6 days on average, followed by a post-treatment window.
Turn the experiment into a factor. Divide experimental incremental revenue by the platform number for the same period. Apply that factor until the next test or the next definition change.

Here is the arithmetic, using invented numbers for a hypothetical brand.

Figure for the test month	Revenue	Return on $60,000 spend
Default setting, all windows	$210,000	3.50
7-day click only	$150,000	2.50
Incremental attribution column	$140,000	2.33
Geo holdout estimate	$165,000	2.75

The factor for 7-day click is 165,000 divided by 150,000, or 1.10. For the default setting it is 0.79. For the incremental column it is 1.18. Next month, with no test running, the report takes that month's 7-day click revenue and multiplies by 1.10. Each row tells a different story about the same month. Only the last one was measured.

If you are too small to test

A holdout needs enough conversions to detect a difference. Many accounts do not have them. In that case, do three things. Keep the default optimisation setting. Report click-through conversions as the platform figure, stated as a floor or a ceiling according to how much remarketing the account runs. And track total revenue against total ad spend each week, because that number does not depend on anyone's attribution model.

Where this stops

Attribution tells you what paid demand produced this month. It does not tell you whether buyers would have found you anyway. That depends on digital authority: how much publications, search engines and AI engines trust the brand. Ads do not build it directly, because AI engines draw on sources, and ads are not sources. The two work together. Strong earned presence can raise the baseline that your holdout group converts at, and that makes honest incrementality reporting matter more. The free Digital Authority report shows where that baseline stands.

Questions

What is Meta incremental attribution?

It is an attribution model you can select in an ad set in place of the standard click, engage and view windows. Meta's delivery system then optimises for conversions that its models predict were caused by the ad, and Ads Manager reports those instead of every conversion inside a time window.

Is incremental attribution the same as a lift test?

No. A lift test withholds ads from a random group and compares outcomes. Incremental attribution is a prediction made by a model. It can be a good prediction, but it is still Meta estimating its own effect, so it needs checking against an experiment.

Why did my Meta ROAS fall in March 2026?

Meta narrowed click-through attribution to link clicks. Likes, comments, shares and saves moved to engage-through attribution with a 1-day window. Common Thread Collective measured a 13% fall in 7-day click ROAS across 50 accounts. Reporting changed. The underlying sales did not.

Can I see incremental conversions without changing my campaigns?

Yes. In Ads Manager, open Columns, choose Compare attribution settings and tick Incremental attribution. It adds columns next to your standard results without touching delivery.

Sources

Haus, Is Meta's Incremental Attribution Outperforming Standard Attribution? What the Data Shows haus.io
Haus, The Meta Report, Lessons from 640 Haus Incrementality Experiments haus.io
Haus, Understanding Meta incrementality testing haus.io
Meta, 2026 AI Drives Performance about.fb.com
Social Media Today, Meta Shares More Info on Incremental Attribution Tracking socialmediatoday.com
Common Thread Collective, Why Your Meta ROAS Dropped Overnight commonthreadco.com
Jon Loomer Digital, How Meta Ads Attribution Works in 2026 jonloomer.com
Jon Loomer Digital, Meta Ads Attribution Reporting, What Your Results Really Mean jonloomer.com
Jon Loomer Digital, Incremental Attribution Added to Compare Attribution Settings jonloomer.com
Jon Loomer Digital, New Incremental Attribution Details jonloomer.com
Search Engine Land, Meta introduces click and engage-through attribution updates searchengineland.com
Gordon, Moakler and Zettelmeyer, Close Enough? A Large-Scale Exploration of Non-Experimental Approaches to Advertising Measurement arxiv.org
Gordon, Moakler and Zettelmeyer, Predicted Incrementality by Experimentation (PIE) for Ad Measurement ideas.repec.org
Seer Interactive, We Tested Meta's New Incremental Attribution Setting on $1M in Ad Spend seerinteractive.com
Genie Goals, What We Learned Testing Meta's Incremental Attribution geniegoals.co.uk
PPC Land, Meta restricts attribution windows and data retention in Ads Insights API ppc.land
LiftLab, Meta Attribution Changes, Impact on Geo Lift and MMM liftlab.com

Esh

Founder, Ethos Engine. Builds the Digital Authority Score and runs the placement marketplace. esh@ethosengine.io

NEXT STEP

See where your brand stands.

Your Digital Authority Score, who AI recommends for your target questions, and the placements that close each gap. Your first report is free.

Get your free report →