The Broken Debate
"Google is just remarketing."
We've heard it from brand founders, Twitter pundits, and Meta-agency loyalists. It's become the default take in every marketing Slack channel and conference panel. Google Ads isn't incremental. You're paying for customers who would've bought anyway.
After auditing 100+ Google Ads accounts, we have a different view. Not because we're Google defenders - we've cut Google spend for clients when the data warranted it. But because the debate needs to distinguish targeting categories from causal evidence.
Here's what most people get wrong: they build their entire Google Ads opinion around brand search. Someone Googles your brand name, clicks your ad, buys. Would they have bought without the ad? Probably. Fair criticism.
Brand share varies by account, window and denominator; a universal four-percent figure is not established here.
The remaining non-brand activity - non-brand Shopping, non-brand Search, Performance Max with brand exclusions - operates on completely different mechanics. And the incrementality debate rarely addresses it with any precision.
This isn't a pro-Google or anti-Google argument. It's a measurement argument. Both sides are using the wrong numbers to make their case, and brands are making six- and seven-figure budget decisions based on a debate that's fundamentally misdirected.
A study can inform measurement design only within its stated population, spend, exposure and methodology. This article does not retain the primary study records needed to transfer the earlier PMax incremental-ROAS figures to another account.
These figures are attributed to a third-party report in the original article; this public summary does not independently validate the dataset or transfer its result to another account. And they tell a very different story than the "Google is just remarketing" crowd would have you believe.
A survey about perceived impact does not establish the incremental return of a particular campaign.
The question is whether the activity being evaluated creates incremental contribution in this account, at what cost, with what uncertainty.
Let's break this apart.
Targeting Mechanisms and Prior Exposure
Search queries, audience signals, product data and prior interactions can all matter across Google's campaign types. Meta also uses multiple signals. Treating one platform as purely keyword-based and the other as purely audience-based is too simple for a measurement design.
A query can reveal the current question without revealing the person's full prior exposure. Someone may see a social ad, later search a category or brand, and finally purchase through another route. Each platform's credited conversion describes its reporting contract, not necessarily its incremental contribution.
Separate brand capture, non-brand discovery and remarketing in the analysis where the account supports those distinctions. Then test the actual causal claim. An observed decline in branded searches after another channel stops is a directional signal; seasonality, offers and concurrent changes can also explain it.
Campaign structure helps make the investigation interpretable. It cannot settle the incrementality question by itself.
The Denominator Trap
A share needs a defined denominator: clicks, impressions, spend, attributed revenue or another measure in a specific account and window. Brand share can differ sharply across those measures.
No universal four-percent brand share is established here. Even a measured four-percent share would not prove the remaining ninety-six percent incremental.
A brand-search holdout can address a particular question if the design handles eligibility, time, seasonality, competitors and spillover appropriately. Simply switching it off for two weeks and observing organic revenue is not a universal causal test.
The question is which outcome changed compared with a defensible counterfactual. Keep the experiment's result separate from attributed credit, revenue redistribution and contribution after acquisition costs.
Frame a Non-Brand Hypothesis
Non-brand Search and Shopping can introduce an unfamiliar product while someone compares options. That makes incremental demand a plausible hypothesis, not a result established by the targeting label.
A person searching “white running shoes” may already know a brand from another channel. They may have purchased through another route without this particular ad. Conversely, an ad can introduce a product they would not otherwise have considered. A query alone cannot identify the counterfactual.
Define the product group, eligible audience, exposure, primary business outcome and window. Choose a holdout or another defensible comparison when the scale and constraints support it. If only observational data is available, state the confounders instead of converting non-brand share into incremental share.
Feed and page improvements remain useful work. Their effect on demand, attributed conversion and incremental contribution are separate questions.
The Measurement Problem Nobody Wants to Admit
Both sides of the incrementality debate share a common weakness: they're reading the wrong numbers.
Platform attribution and incremental measurement answer different questions. A difference reported for one brand does not establish a market-wide overstatement percentage.
A campaign showing 3.0x ROAS in Google's dashboard might be delivering 0.5x incrementally. That's not a rounding error. It's a 6x discrepancy.
A platform-reported conversion may be genuine while its assigned credit differs from the effect measured by an experiment. Compare definitions and evidence rather than assuming either view is universally correct.
Neither side is measuring incrementality directly. They're arguing about proxy metrics.
Here's what direct measurement looks like:
Confirm current lift-test eligibility, available study designs and expected power in the actual account. A universal minimum budget is not established here.
Detectable effects depend on the design, variance, exposure, contamination and outcome rate. Estimate power for the proposed test; a conversion count alone does not guarantee a particular precision.
Use a retained experiment record before labeling revenue truly incremental. Keep assigned groups, dates, treatment, costs, outcomes and uncertainty together; an attribution-method change alone cannot create causal evidence.
The real risk isn't that Google Ads doesn't work. It's that you're making decisions based on numbers that don't reflect reality - in either direction.
Most brands are either over-investing based on inflated ROAS or under-investing based on the "it's just remarketing" narrative. Both positions cost money. The answer is measurement, not philosophy.
The Incremental Stack
After auditing 100+ accounts and studying the measurement data, we've developed a framework for how to think about Google Ads incrementality. We call it The Incremental Stack - 4 layers, each building on the one below.
Layer 1: Architecture Understanding
Before measuring incrementality, define the actual treatment, exposed population and outcome. Query and audience information help describe that treatment; neither targeting label determines incremental value.
Non-brand and brand describe query relationships. Either can create incremental value or capture purchases that would otherwise occur; a suitable comparison is needed to distinguish them.
Most accounts never get past this layer because they treat all Google Ads spend as one bucket.
Layer 2: The Brand/Non-Brand Split
Separate brand from non-brand. Measure them independently. This is the baseline - if you haven't done this, nothing else matters.
A brand holdout needs an appropriate assignment, duration, business-risk boundary and contamination check. A short before/after pause can be confounded by seasonality, demand or other campaigns.
Layer 3: The Coverage Engine
Separating brand and non-brand can make allocation easier to inspect. Coverage still describes eligibility and delivery, rather than the fraction of orders caused by advertising.
Spend concentration describes the account's allocation. It does not identify incremental orders on those products or prove the rest of the catalog has no demand.
Layer 4: The Measurement System
Build KPIs you trust. Platform ROAS is not a KPI - it's a vanity metric. Set up:
- Shorter attribution windows in your MTA tool (24-48 hour click-to-conversion)
- Regular incrementality lift tests (quarterly at minimum)
- Brand vs non-brand ROAS tracked separately
- New customer rate by campaign type
The Stack is sequential. Layer 3 doesn't matter if Layer 1 is wrong. Layer 4 is meaningless without Layer 3 in place. Most accounts jump straight to Layer 4 (measurement) without building Layers 2 and 3 first. They measure a broken system and conclude that Google doesn't work.
It's not that Google doesn't work. It's that their Google setup doesn't work.
The Feed Quality Multiplier
We've saved this section for a reason. Feed quality is the most underrated variable in the incrementality equation, and it's the one we have the most data on.
Your product feed isn't a data file. It's your auction entry ticket.
Google uses supported product data for matching and eligibility. Missing or vague facts can limit relevance, but a templated title alone does not establish that Google cannot understand the item or match useful queries.
Repeated wording can conceal useful product differences, but it does not establish a Google duplicate-content or Quality Score penalty. Review valid product identity, approved attributes and actual item-level eligibility before testing content changes.
The earlier feed example used an internal text-difference heuristic. It is not a Google quality measure; retain valid product identity, factual distinctions and actual test results instead.
The author reports new query visibility after title changes. Without the actual comparison and concurrent-change record, the story does not establish that demand stayed constant or isolate the title change as the cause.
Broader eligible coverage can create additional opportunities to test. A claimed share of eligible auctions cannot be converted into an equal share of incremental revenue without outcome and counterfactual evidence.
Most brands debate Google incrementality at Layer 1 (is it remarketing or not?) while their real constraint sits at Layer 3 (coverage limited by feed quality).
You're not demand-capped. You're structure-capped.
How to Actually Answer the Question for Your Account
Philosophizing about incrementality is free. Testing it costs money and attention. Here's a practical framework for answering the question for YOUR account.
Step 1: Assess your product uniqueness.
A differentiated product can still face substitutes, existing awareness and different purchase routes. Product uniqueness alone does not establish whether brand or non-brand activity is incremental.
If you sell in a competitive category (most e-commerce), both brand and non-brand matter - but for different reasons. Brand protects against competitor conquest. Non-brand drives new acquisition.
Step 2: Evaluate your non-brand replaceability.
Ask what the customer would have done without this exposure. The answer is not observed from the generic query or an operator’s belief; design a suitable comparison where feasible.
Step 3: Set conservative attribution windows.
Changing an attribution window changes credited outcomes, not the counterfactual. Choose the window for the buying cycle and reporting question; a shorter window is not inherently closer to true incrementality.
Step 4: Run incrementality tests quarterly.
Google's lowered minimums make this accessible for most brands. Start with your highest-spend non-brand campaigns. Run geo-based or conversion lift tests. Get real incrementality data instead of arguing about it.
Check detection limits. Estimate power from the chosen design, baseline rate, variance and available exposure. An inconclusive study does not establish either no effect or a fixed minimum detectable percentage.
Step 5: Build the KPIs you trust, then optimize against them.
The goal isn't perfect measurement. Perfect measurement doesn't exist in marketing. The goal is a measurement system you trust enough to make decisions against. Set your benchmarks, test regularly, and scale what's incremental at an acceptable ROI.
The Real Takeaway
The Google Ads incrementality debate is valid. But it's been stuck at the wrong level for years.
Critics focus on brand search - the smallest, least interesting slice of the platform. Defenders cite platform-reported ROAS - numbers that overstate impact by a third. Neither side is measuring what matters.
Targeting architecture describes how an ad may reach someone. Queries, product data and audience signals can all matter. None establishes incremental demand by nature; test the actual account decision and contribution.
The Incremental Stack is how we think about it:
- Understand the architecture (keyword vs audience)
- Split and test brand independently
- Build the coverage engine (product, angle, feed quality)
- Measure with KPIs you trust
Most brands are arguing at Layer 1 while their real constraints live at Layer 3. Fix the engine. Build the measurement system. Then debate the philosophy.
Stop asking "is Google incremental?" Start asking "is MY Google setup structured to capture incremental value?"
The answer to the first question is nuanced. The answer to the second question is something you can actually fix.
Put the framework to work
Design the incrementality question