Incrementality is the measure of outcomes that only happened because of your marketing - the results you would have lost if the campaign had never run. It is established through a controlled experiment: expose one group, hold the marketing back from a comparable group, and the difference is the incremental lift. Incrementality answers a question attribution cannot: not which touchpoints got credit, but whether those conversions would have happened anyway. For AI marketers it is the rigorous way to prove that AEO drives net-new pipeline rather than capturing demand you were already going to win.
What is incrementality?
Incrementality is a randomized, controlled experiment that measures the additional outcomes caused by marketing, compared with a world where that marketing did not run. The design is simple: split people into a treatment group that is exposed to the campaign and a control, or holdout, group that is not, keep the groups comparable, and measure the gap in outcomes. That gap - the "lift" - is the incremental impact. Everything above the control's baseline was caused by the marketing; everything at or below it would have happened regardless.
The reason incrementality has become a priority is that the alternative measures flatter marketing. Attribution and return-on-ad-spend dashboards count conversions a channel touched, but touching a conversion is not the same as causing it. The classic example is retargeting: ads served to people already heading to checkout earn enormous attribution credit while adding little real revenue. Incrementality is the discipline that strips out that illusion. As the measurement vendors who specialize in it put it, attribution tells you where conversions came from; incrementality tells you whether they would have happened anyway.
How incrementality testing works
Every incrementality test is a variation on the same idea: deny the marketing to a comparable group and watch what changes.
Test and control (holdout) groups
The treatment group sees the campaign; the control group is held out. To trust the result the two groups must be alike in every way that matters except exposure, which is why randomization matters. The difference in conversion rate between them, scaled to the population, is the incremental lift. (Note: some platform documentation labels these groups inconsistently, but the convention is fixed - the exposed group is the test or treatment group, and the unexposed group is the control or holdout.)
Geo experiments
The most common method for measuring lift is the geo experiment, also called a geo holdout. You select comparable geographic regions - cities, states, designated market areas - run the marketing in the treatment regions, withhold spend in the control regions, and compare. Geo tests are popular precisely because they compare whole markets rather than individuals, so they need no cookies or user-level identifiers and survive privacy restrictions intact.
Platform lift tools
Where the platform controls exposure, it can run the split for you. Meta Conversion Lift and Google's Conversion Lift and Brand Lift products randomize exposure inside the platform; independent vendors such as Measured, Haus, and Triple Whale run geo and holdout experiments across channels for a more neutral read.
Incrementality vs attribution vs MMM
Incrementality is one of three measurement disciplines marketers conflate. Each answers a distinct question:
| Incrementality | Attribution | Media mix modeling | |
|---|---|---|---|
| Method | Controlled experiment | User-level credit assignment | Aggregate regression |
| Question | Did the marketing cause net-new outcomes? | Which touchpoints get credit? | How should I split budget across channels? |
| Proves | Causation | Correlation along a path | Contribution, top-down |
| Cadence | Point-in-time test | Continuous | Always-on model |
Incrementality and media mix modeling are natural partners. MMM runs continuously and allocates budget across channels using correlation; incrementality runs as a discrete experiment and proves causation. In practice, incrementality experiments are used to calibrate MMM, anchoring the model's correlational estimates to a causal ground truth.
Why incrementality matters for AI search
AEO and AI-visibility programs face a specific credibility problem: the easy metrics cannot prove they work. A last-click dashboard will show AI driving a tiny fraction of conversions because AI mostly influences the discovery phase, weeks before the trackable click - the heart of the attribution gap. Worse, when branded conversions do rise after an AEO push, a skeptic can fairly ask whether those buyers would have converted anyway.
Incrementality is the answer to that skeptic. By holding AEO investment back from some markets or periods and comparing outcomes, you isolate the net-new demand the work actually created. It moves the AEO conversation from "AI gets the credit my model assigns it" - which a CFO can dismiss - to "pipeline rose X percent in markets where we invested in AI visibility and did not rise where we held back." That is causal evidence, and it is the strongest case an AEO budget can make.
How to run an incrementality test for AEO
- Geo holdout. Concentrate AEO and AI-visibility effort on a set of treatment markets and deliberately hold back in comparable control markets, then compare branded search, pipeline, and conversions between them.
- Time-based holdout. When geo splitting is impractical, alternate on and off periods against a stable baseline and a control market, watching for lift that tracks the on periods with the expected lag.
- Conversion lift for paid AI. For ChatGPT Ads and similar placements, use the platform's audience-split lift test rather than reading ROAS at face value.
- Pick the right outcome. Measure something the test can move: branded search volume, pipeline, or citation rate. Our AI Visibility Checker gives you a citation-share reading to compare across test and control.
Our AI Analytics module tracks citation share and AI-referred outcomes over time, which gives an incrementality test a clean before-and-after signal to compare across your treatment and control groups.
Common misconceptions
Incrementality and attribution are the same thing
They are opposites in spirit. Attribution distributes credit among touchpoints that participated in conversions; incrementality questions whether those conversions needed the marketing at all. A channel can dominate your attribution report and be nearly non-incremental.
A positive ROAS proves the spend was worth it
Not on its own. ROAS counts credited revenue, not caused revenue. Spend that mostly reaches already-converting buyers shows a healthy ROAS while contributing little incremental lift. The holdout is what separates the two.
Incrementality is only for paid media
Paid media makes splits easy, but organic work is testable too. Geo and time-based holdouts let you measure the incremental effect of AEO, content, and other organic investments that have no click to track.
You need massive volume to test
Geo experiments do need enough markets and conversions to reach significance, which can be a barrier for small programs. But time-based holdouts against a control market, and lift tests on higher-volume paid channels, bring incrementality within reach of teams that cannot run a full geo design.
Frequently asked questions
#What is incrementality in simple terms?
Incrementality is the share of results that only happened because of your marketing - the outcomes you would have missed if the campaign had never run. You measure it with a controlled experiment: show the marketing to one group, hold it back from a comparable group, and the difference between them is the incremental lift. It separates results your marketing caused from results that would have happened anyway.
#What is the difference between incrementality and attribution?
Attribution assigns credit for conversions that already happened to the touchpoints a buyer interacted with. Incrementality asks a harder question: would those conversions have happened without the marketing at all? A channel can win all the attribution credit and still be barely incremental - for example, retargeting ads served to people who were already going to buy. Attribution tells you where conversions came from; incrementality tells you which marketing actually caused them.
#How do you run an incrementality test?
You split your audience into a test group that is exposed to the marketing and a control (holdout) group that is not, keep the two groups comparable, and measure the difference in outcomes. The most common method is a geo experiment: pick similar regions, run the marketing in some and withhold it in others, then compare. Because geo holdouts compare whole markets rather than tracking individuals, they work without cookies or user-level data.
#Can you measure incrementality for AEO or AI search?
Yes, with holdout design. For paid AI ads, platform lift tools split audiences directly. For organic AEO, the workable approach is a geo or time-based holdout: invest in AI visibility for some markets or periods and not others, then compare branded search, pipeline, or citation share between them. It is the only rigorous way to prove that AEO work drives net-new demand rather than capturing demand that would have converted anyway.
#Why isn't a good ROAS enough to prove incrementality?
Return on ad spend counts the conversions a channel was credited with, but credit is not causation. If an ad mostly reaches people who already intended to buy, it can post a strong ROAS while adding almost no incremental revenue - the business would have earned most of those sales without it. Only a holdout test, where some buyers never see the marketing, reveals how much of that ROAS is genuinely incremental.
