The lift in brand mentions or visibility in AI-generated search results that is directly caused by a marketing or content action, after subtracting the baseline.
Marketing professionals analyzing brand performance and campaign impact within AI search environments.
01What it is and how it works
Incrementality isolates the causal effect of a single action on your brand’s appearance in AI search. AI search models (like those behind ChatGPT, Gemini, or Claude) generate answers from their training data and from real-time retrieval of indexed content. When you publish a new article, update a knowledge panel, or run a paid placement, the model may incorporate that new information. Incrementality compares a test group (where the action was taken) against a control group (where it was not) to measure the net new appearances. The mechanism relies on randomized experiments or time-series analysis: you observe brand mentions in AI outputs before and after the action, while controlling for other factors like seasonality or algorithm updates. The difference is the incremental lift.
It tells you how much more your brand shows up in AI answers because of something you did, compared to if you had done nothing.
02What to do about it
Start by defining a clear baseline. Use a holdout group — for example, a set of queries or a geographic region where you do not deploy the new content or campaign. Measure brand mentions in AI search responses for both groups over a consistent period (at least two weeks to account for model update cycles). Then run your action: publish the content, launch the campaign, or update the entity. After the action, measure again. The difference in the change between test and control is your incrementality. Repeat this process for every major content or campaign change. Document the results to build a library of what drives real lift versus what only looks like lift because of external noise.
03How it is measured or noticed
You measure incrementality by comparing brand appearance rates in AI search outputs across a test group and a control group. The key metric is the incremental appearance rate: (appearances in test after action minus appearances in test before action) minus (appearances in control after minus appearances in control before). You can also measure incremental share of voice — the percentage of relevant AI answers that include your brand, adjusted for baseline. Tools that log AI search responses over time can automate this. Look for statistically significant differences; a common threshold is a p-value below 0.05. Without a control, you risk mistaking organic growth for campaign impact.
04Common mistakes
- Comparing before-and-after without a control group — this confuses seasonal trends or algorithm updates with your action's effect.
- Measuring too short a window — AI models may take days or weeks to reflect new content, so a one-day measurement is noise.
- Ignoring model version changes — if the AI model updates during your test, the baseline shifts and your incrementality estimate is invalid.
- Using the same queries for test and control — overlap contaminates the control; use disjoint sets of queries or regions.
- Attributing all lift to one action when multiple actions ran simultaneously — isolate one variable at a time.
05Limits
Incrementality does not apply when you cannot create a valid control group — for example, if the action is global and affects all queries equally. It also struggles with very small effects: if the lift is less than the natural noise in AI outputs, you need a large sample size to detect it. Incrementality is often confused with attribution, which assigns credit across multiple touchpoints. Attribution assumes a path; incrementality assumes a counterfactual. They answer different questions: attribution says 'which channel helped?', incrementality says 'did this action help at all?' Finally, incrementality measures short-term causal impact, not long-term brand building. A single action may have delayed effects that the test window misses.
06Worked example
A brand publishes a new help center article about a common customer problem. They randomly split 1,000 search queries into two groups: 500 queries where the article is indexed and 500 where it is blocked via robots.txt. Before publication, both groups show the brand in 10% of AI answers. After one week, the test group shows the brand in 22% of answers; the control group shows 11%. The incrementality is (22% - 10%) - (11% - 10%) = 12% - 1% = 11 percentage points. The article caused an 11-point lift in brand visibility in AI search.
Frequently asked questions
How is incrementality different from simply tracking total brand mentions in AI search?
Incrementality isolates the causal effect of a specific action, whereas total mentions include baseline appearances that would have happened anyway. Without incrementality, you might mistake correlation for causation and invest in actions that don't actually drive lift.
Should we measure incrementality for every content or marketing action?
It depends on the scale of the action and your ability to create a valid control group. For large, isolated campaigns with a clear test group, it is highly recommended. For small or global actions that affect all queries equally, incrementality may not be applicable.
How do you set up a test and control group for AI search incrementality?
Define a set of queries that are exposed to your action (test group) and a similar set that are not (control group). Then compare the rate of brand mentions in AI search outputs between the two groups over the same time period.
Does incrementality still work if the AI model updates frequently?
Yes, but you must run the test and control groups simultaneously to control for model changes. If the model updates during the test, both groups are affected equally, preserving the validity of the comparison.
What happens if we ignore incrementality and just look at total brand mentions?
You risk overinvesting in actions that do not actually cause the lift, because you cannot distinguish between organic growth and campaign-driven effects. This can lead to wasted budget and misinformed strategy.
How long does it take to see incrementality from a new content piece?
It typically takes days to weeks, depending on how quickly the AI model indexes and reflects new content. You should measure the baseline before the action and then track mentions at regular intervals after launch.
Asked out loud
spoken, not typedThe same term in the words somebody uses speaking to an assistant rather than typing into a box — written from the situation, which is why each one carries the situation it came from.
You need to run a controlled test comparing queries that saw your content against a similar set that didn't. That's called incrementality measurement, and it will give you the real lift caused by your action.
Compare your brand's lift against a control group of similar brands or queries that weren't affected by your actions. That's incrementality, and it separates your impact from external noise.
You should use incrementality, which measures the lift directly caused by your actions after subtracting the baseline. It will show you which actions truly drive visibility and which ones don't.