Google Search advertising produces inflated ROAS figures because it takes credit for clicks that organic search would have delivered for free. Measuring true incremental return requires pause testing, geo holdout experiments, and separating branded from non-branded activity.
Paid search is often the highest-reported ROAS channel in a marketing stack. That strong performance is partly real and partly a measurement artefact. When someone searches for your brand name and clicks your paid ad, they would likely have found you through the organic results directly below it. Google counts that click as a paid conversion. You paid for something that was probably free.
The organic cannibalisation problem
Organic cannibalisation occurs when your paid search ads take clicks from your own organic search results. For branded search terms where your organic result ranks first, this happens constantly. The user searches your brand name, sees the paid ad at the top, clicks it, and converts. Google Ads counts it as a paid sale. Your organic listing, which would have received that click if no paid ad existed, gets nothing.
Cannibalisation is highest on branded terms because your organic ranking for your own brand name is usually strong. On competitive non-branded terms where you rank lower organically, paid ads genuinely extend your reach to users who would not have found you otherwise. The incremental value of paid search varies significantly between these two categories.
Pause testing: the simplest incrementality test
A pause test turns off paid search for a defined period and measures what happens to total traffic and conversions. The logic is straightforward. If paid search is truly incremental, organic traffic should not compensate when you pause it. If organic traffic rises when you pause paid, the paid ads were cannibalising organic clicks.
- Choose a period that is not seasonally unusual for your business. Pausing during peak sale periods distorts the results.
- Run the pause for at least two to three weeks to see past any initial novelty effects.
- Monitor branded and non-branded terms separately. The results will often be very different.
- Calculate the net revenue difference: total revenue during the pause versus the equivalent prior period, adjusted for any trend.
- Apply the same adjustment when you resume spending to confirm the lift reappears consistently.
Geo holdout testing for search
Geo holdout tests for paid search divide the country into matched regions. In the test regions, your paid search campaigns run normally. In the control regions, you pause or reduce paid search. Because the regions are matched in terms of user demographics and baseline conversion rates, the difference in outcomes is attributable to the paid search activity.
Google's own Conversion Lift tool offers a version of this for Search, though it requires coordination with your Google account team. Third-party measurement providers can run independent geo holdouts using your CRM data and regional sales figures, which many advertisers prefer because the methodology is not controlled by Google.
Branded search terms typically show incrementality of 30 to 60 percent, meaning 40 to 70 percent of those clicks would have come through organically anyway. Non-branded terms where you rank low organically can show incrementality above 80 percent.
Marketing mix modelling and paid search
MMM estimates paid search contribution using spend variation over time rather than click attribution. One limitation is that search spend and organic search traffic often move together, because both tend to rise during periods of high consumer intent. Separating the two requires careful model specification and, ideally, periods in your data where paid search spend varied independently of organic performance.
Despite this limitation, MMM is valuable because it estimates search contribution alongside all your other channels. This lets you compare paid search ROI directly with TV, social, and email on a consistent basis, which click-based attribution cannot do.
Splitting branded and non-branded budgets
Treating branded and non-branded paid search as two separate channels in your measurement framework produces much clearer results. Branded campaigns often look excellent in platform reporting but have low true incrementality. Non-branded campaigns often look weaker in reporting but drive genuinely new customer acquisition. Separate reporting reveals this split and informs smarter budget decisions.
For a deeper treatment of this split and how to measure the two types of search independently, see our dedicated article on branded versus non-branded search measurement.
Will pausing branded search hurt my brand if competitors are bidding on my brand terms?
This is a legitimate concern and a reason to pause for short windows rather than permanently. During a two-week pause test, monitor whether competitors' ads appear on your brand terms in their place. If they do, the cost of pausing includes whatever traffic those competitors capture. Factor this into your incrementality calculation. In many cases, even accounting for competitive interception, branded search ROI is lower than platform data suggests, but the cost of pausing may still exceed the saving.
How does smart bidding affect incrementality measurement?
Smart bidding strategies like Target ROAS and Target CPA use machine learning to allocate bids in real time. This makes spend patterns less predictable week over week, which can complicate MMM. For geo holdout tests, smart bidding does not fundamentally change the methodology, but you should confirm that the bidding strategy does not automatically shift budget away from your control regions, which would contaminate the test. Separate campaigns by geo and apply bidding strategies within each region independently.
Is Performance Max (PMax) harder to measure incrementally?
Yes. PMax combines search, display, YouTube, and Gmail inventory in a single campaign with limited segmentation available in reporting. This makes it difficult to isolate the search component for incrementality testing. The best approach for now is to run PMax alongside a separate standard Search campaign for your most important terms and use geo holdouts across the whole Google activity rather than trying to measure PMax in isolation. This is an evolving area as Google releases more reporting transparency over time.