Most Google Ads dashboards are excellent at telling you what happened clicks, impressions, attributed conversions, ROAS. What they are far worse at telling you is why it happened, and specifically, whether your ad spend actually caused a customer to buy, sign up, or fill out a form or whether that customer was going to convert anyway. That gap between correlation and causation is exactly what a Conversion Lift study is built to close, and in 2026 it has moved from a “nice-to-have enterprise feature” to a mainstream measurement requirement as third-party cookies fade and attribution models grow noisier.
This guide breaks down what a Conversion Lift study actually is, why performance marketers increasingly need one, which campaign types support it, the technical and operational criteria you have to meet before you can even launch one, and how holdouts and Study Power determine whether your results will be trustworthy.
What Is a Conversion Lift Study?
A Conversion Lift study is a randomized, controlled experiment run natively inside Google Ads. Instead of comparing “before vs. after” performance, which is easily skewed by seasonality, competitor activity, or organic demand shifts, Google splits your eligible audience into two statistically comparable groups automatically:
- Treatment group: users who are eligible to see your ads
- Control group: users who are deliberately withheld from seeing your ads for the duration of the test
Because the split happens randomly at scale, the two groups behave like statistical twins. Whatever conversion difference shows up between them at the end of the test period can be attributed to the ad exposure itself, not to outside noise. Google reports this as Incremental Conversions, calculated simply as treatment conversions minus control conversions, along with a Relative Lift percentage that expresses that gap as a proportion of the control group’s baseline.
This is a fundamentally different question than the one your standard “Conversions” column answers. Attributed conversions tell you which touchpoint gets credit under your chosen attribution model. Incremental conversions tell you whether that touchpoint changed the outcome at all. A campaign can look phenomenal in last-click reporting while contributing almost nothing incremental, because it was simply intercepting demand that already existed branded search traffic is the classic example.
Conversion Lift sits inside a broader family of measurement tools Google now groups together as “Lift Studies,” alongside Brand Lift (perception, recall, favorability) and Search Lift (impact on organic search behavior). In 2026, Google restructured its Ads interface so that all three lift study types now live inside the Experiments section rather than being scattered across separate reporting tabs a signal that Google now treats incrementality testing as a core discipline of campaign experimentation, not a side project reserved for big brand budgets.
Why a Conversion Lift Study Is Necessary
There are three structural problems in modern digital measurement that make lift studies necessary rather than optional.
1. Attribution models can’t see the counterfactual. Even the most sophisticated data-driven attribution model only distributes credit across touchpoints that were observed. It has no way of knowing what would have happened if the ad had never been shown. A holdout-based experiment is the only method that actually observes that counterfactual, because it creates a real population of users who genuinely never saw the ad.
2. Privacy changes have degraded signal quality. Cookie restrictions, consent-mode defaults, and in-app tracking limitations mean fewer conversions are observably tied to a specific click. Enhanced Conversions and offline conversion imports help recover some of that signal, but they still operate inside an attribution framework, not a causal one. Teams that have implemented Enhanced Conversions have reported measurable incremental ROAS gains on Search, but that number itself was only knowable because it came from a lift-style analysis run across many accounts underscoring that even measurement-quality fixes need incrementality testing to prove their value.
3. Budget conversations require causal proof, not correlation. When a CFO or CMO asks whether a channel is “driving net new revenue” or simply intercepting demand that would have converted through organic search or direct traffic anyway, pointing at the conversions column doesn’t answer the question. A lift study does, because it isolates the advertising effect from every other variable moving in the market at the same time seasonality, PR coverage, competitor promotions, or a spike in category demand.
This is precisely why lift studies are described as more rigorous than simple trend-reading. If search volume for your brand climbs the week after a campaign launches, that could be the campaign or it could be an unrelated news cycle, a competitor’s stockout, or a seasonal pattern that happens every year. Only a controlled experiment separates the ad effect from everything else.
Which Campaign Types Support Conversion Lift
Eligibility is one of the most common points of confusion for teams attempting their first study, because access is not universal across the platform.
User-based Conversion Lift, where individual users are randomly assigned to treatment or control, is most commonly available and discussed for:
- Video campaigns
- Discovery campaigns
- Demand Gen campaigns
- App campaigns (in select configurations)
Search and Performance Max campaigns can be more restrictive. Availability often depends on account size, historical spend, and whether your account has a dedicated Google account team, since these campaign types typically require geo-based experimental designs rather than pure user-based randomization to avoid contaminating the control group (a Performance Max campaign, by nature, can serve across Search, Display, YouTube, Gmail, and Maps simultaneously, which makes a clean user-level holdout harder to enforce).
Geo-based Conversion Lift is the fallback design when user-level randomization isn’t feasible. Instead of splitting individual users, Google splits matched geographic regions: some regions get ads, matched “twin” regions don’t, and lift is measured at the aggregate regional level. This approach is common for retail, QSR, and other business models where offline or in-store conversions matter as much as digital ones.
The practical takeaway: before you plan a study, confirm eligibility for your specific campaign type and account tier. Some formats are self-serve inside the Experiments section; others require reaching out to your account team to unlock access.
The Core Requirements Before You Launch a Study
A Conversion Lift study is not something you should bolt onto a campaign that’s already live and politically difficult to touch. The requirements are best addressed at the planning stage, before budgets are locked and stakeholders are expecting results on a fixed timeline.
1. A Meaningful, Well-Defined Conversion Action
The study needs a conversion event that genuinely reflects business value not a proxy so shallow it doesn’t matter, and not so deep in the funnel that volume collapses. For lead-gen advertisers, a raw form-fill count is often insufficient on its own; connecting the lift readout to downstream pipeline quality (via offline conversion imports or CRM-stage tracking) produces a far more defensible result than form-fills alone.
2. Adequate Conversion Volume
Google’s Study Power tool the feasibility calculator built into the Experiments section, estimates the probability that your test will produce a conclusive result given your current conversion volume, budget, and expected lift size. Low-volume conversion actions frequently need one of three fixes: a longer test duration, a broader conversion definition that captures more volume, or a different experimental design altogether (geo-based instead of user-based, for instance).
3. A Documented, Intentional Holdout
Every lift study requires a control group that is deliberately denied ad exposure for the test’s duration. This is a real business tradeoff: some portion of your addressable audience is intentionally not being marketed to, and it should be documented and agreed upon with finance or leadership stakeholders before the test begins, not discovered after the fact when someone asks why conversion volume dipped.Also Read: What Are the 5 Benefits of Advertising(Google Ads)?
4. Sufficient Duration to Capture Conversion Lag
Digital experiments run for too short a window are one of the most common causes of inconclusive results. The study needs to run long enough to capture the average delay between ad exposure and conversion, what’s often called conversion lag or post-click delay. High-consideration purchases and longer B2B sales cycles need proportionally longer test windows than impulse-purchase e-commerce categories.
5. Clean, Verified Conversion Tracking
This is the requirement most frequently overlooked. Before launch, teams should audit GA4 key event configuration, UTM consistency, offline conversion import mapping, CRM stage definitions, and consent-mode assumptions. A lift study built on top of broken or double-counted conversion tracking will produce a confidently wrong answer rather than an honest inconclusive one arguably a worse outcome, because it looks authoritative.
Understanding Holdouts and Study Power in Detail
Since holdouts and Study Power are the two concepts advertisers misunderstand most often, they deserve a closer look.
Study Power is expressed as an estimated certainty-of-lift percentage, typically shown on a scale from 50% to 95% in five-point increments. It answers a specific question: given your current setup, how likely is it that this study will produce a conclusive, statistically reliable result rather than an ambiguous one? Google’s tooling generates a recommended configuration automatically, but you can adjust it manually. If study power comes in below roughly 90%, the platform will typically surface budget guidance describing what it would take to reach that higher confidence threshold.
Holdout size is the single biggest lever inside your control. A larger holdout up to 50% of the eligible audience gathers a bigger control sample faster, which shortens the time needed to detect a real difference, but it comes with a proportionally higher opportunity cost, since a larger share of your addressable audience is going untargeted during the test. A smaller holdout preserves more of your working ad exposure but extends the time required to accumulate enough conversions in each group to detect a statistically meaningful gap.
There’s also a funnel-stage nuance worth planning around: upper-funnel and mid-funnel actions pageviews, lead form starts, product-page visits tend to show clearer, higher-magnitude lift than bottom-funnel purchase events, simply because they occur more frequently and closer in time to ad exposure. Google’s own guidance recommends measuring both a primary bottom-funnel KPI and a secondary upper- or mid-funnel action, using the latter to sanity-check that people are responding positively to the ads even in categories where the ultimate purchase conversion is rare or delayed.
Reading and Interpreting Your Results
Once a study concludes, resist the temptation to read the headline lift number in isolation. A few interpretive habits separate a rigorous readout from a misleading one:
Distinguish platform-level lift from business-level lift. A Google lift study measures incremental impact strictly within Google’s observable environment. It cannot see whether that incremental conversion cannibalized a sale that would otherwise have happened through a different channel, or whether it represents genuinely new revenue for the business. Cross-referencing the lift result against CRM data, revenue trends, or downstream funnel metrics helps confirm whether platform-level lift is translating into real business-level growth.
Use incremental cost per conversion (iCPC), not just raw lift. Comparing iCPC across channels the cost of only the conversions the ads actually caused, not all attributed conversions imposes financial discipline that a simple lift percentage doesn’t. A channel with modest relative lift but a very low iCPC can still be a better budget allocation than a channel with dramatic lift but an expensive incremental conversion.
Account for effects the study structurally can’t see. Because Conversion Lift and geo-based experiments quantify incremental value strictly within Google’s environment, they typically underweight view-through influence, assisted conversions on other platforms, and brand-reinforcement effects that pay off later through a different channel. A paid search campaign, for example, may appear incremental within Google’s test while still depending on brand-building spend happening elsewhere an interaction the test simply cannot observe.
Wait for the full study period before drawing conclusions. Cumulative or in-flight results may be visible during the test, but Google explicitly recommends waiting until the official end date for the most reliable read, especially for Demand Gen studies, which include modeled “delayed incremental conversions” conversions expected to land after the study’s official close, based on the conversion delays observed during the test window itself.
Common Pitfalls That Undermine a Study
A handful of mistakes account for most inconclusive or misleading lift studies:
- Launching without eligibility confirmed. Discovering mid-setup that your campaign type or account tier isn’t supported wastes planning time. Confirm access first.
- Choosing a holdout size without modeling the opportunity cost. A 50% holdout might satisfy statistical needs quickly but can meaningfully dent revenue during the test window if not sized deliberately.
- Running the study for too short a duration. Especially for high-consideration purchases, ending the test before the average conversion lag has elapsed produces an artificially low, unreliable lift figure.
- Measuring only a single, deep-funnel conversion event. This limits volume and can leave the study underpowered when a broader or secondary KPI would have produced a clearer signal.
- Skipping the tracking audit. Inconsistent UTMs, misconfigured GA4 key events, or unmapped offline conversions introduce noise that a statistical test can’t distinguish from genuine treatment effects.
- Treating in-flight numbers as final. Reading cumulative results too early, particularly on Demand Gen campaigns with delayed conversion modeling, can lead to premature and incorrect conclusions.
Setting Up a Study: A Practical Walkthrough
Beyond the eligibility checklist, it helps to see how the pieces fit together operationally, from kickoff to readout.
Step 1 Confirm eligibility and pull historical data. Before opening the Experiments section, check whether your campaign type, account tier, and region support the design you want (user-based or geo-based). Pull at least 60–90 days of historical conversion volume, average order value, and existing conversion rate by campaign. This baseline data is what the Study Power calculator will use to model feasibility.
Step 2 Run the Study Power estimate before committing to a launch date. Enter your expected budget, expected lift, and desired holdout into the feasibility tool and review the certainty-of-lift percentage it returns. If the estimate lands below your organization’s comfort threshold, adjust one variable at a time extend duration, widen the conversion definition, or increase the holdout rather than changing everything simultaneously, so you can see which lever actually moves the needle.
Step 3 Get sign-off on the holdout tradeoff in writing. Because a holdout means deliberately withholding ads from a meaningful slice of your addressable audience, this is the point where finance, brand, or leadership stakeholders should explicitly acknowledge the tradeoff. A quick internal memo stating the holdout size, expected duration, and estimated opportunity cost avoids awkward questions three weeks into the test when someone notices conversion volume looks lower than usual.
Step 4 Freeze major campaign changes for the test window. Avoid changing bids, budgets, creative, or targeting mid-study wherever possible. Every structural change you make during an active test introduces a new variable that muddies the comparison between treatment and control, and can force you to restart the clock on Study Power.
Step 5 Let the study run to its full, planned duration. Resist the urge to check results daily and react to early signals, particularly for categories with long conversion lag. Cumulative in-flight numbers can shift meaningfully as the sample matures, and Demand Gen studies in particular model delayed incremental conversions that only appear in the final report.
Step 6 Read the results against a pre-agreed decision framework. Before the study starts, define in advance what a “successful” lift result would look like, and what action follows from a null or negative result. Without this framework, teams often rationalize inconclusive results after the fact rather than making the harder call to reallocate budget.
Who Should Prioritize Running One First
Not every advertiser needs to run a Conversion Lift study on every campaign simultaneously. A few account profiles get outsized value from prioritizing it early:
- Accounts with strong branded search performance. If a large share of “conversions” is coming from users searching your brand name directly, a lift study is the fastest way to find out how much of that volume your paid campaigns are actually causing versus simply intercepting.
- Accounts shifting significant budget into Performance Max or Demand Gen. Because these automated formats reduce placement-level visibility, a lift study is one of the few remaining ways to validate that the machine-learned optimization is producing real incremental results rather than reallocating credit from other channels.
- Lead-gen and B2B accounts with long sales cycles. These accounts benefit most from connecting lift results to downstream CRM stages rather than stopping at the form-fill, since a shallow conversion definition can produce a misleadingly small or large lift number.
- Any account preparing for a budget conversation with finance or leadership. A documented incrementality result is simply a stronger argument for (or against) continued investment than an attributed-conversions screenshot.
Conversion Lift in the 2026 Measurement Landscape
Two shifts make Conversion Lift more relevant in 2026 than in prior years. First, Google’s interface consolidation moving all lift study types into the Experiments section reflects a platform-level push toward treating incrementality testing as a routine part of campaign management rather than a specialized enterprise service. Second, the rapid rise of automated, feed-driven, and AI-assisted campaign types (Performance Max, AI Max for Search, Demand Gen) has made attribution murkier by design: these formats intentionally reduce keyword-level and placement-level visibility in exchange for broader reach and machine-learned optimization. When you can no longer see exactly which query or placement drove a click, a controlled experiment that measures the aggregate causal effect becomes one of the few remaining ways to validate that automated spend is actually working.
Final Takeaway
A Conversion Lift study is the difference between assuming your ads work and proving it. It requires more upfront planning than reading a standard performance report eligible campaign access, a well-defined conversion action, enough volume to power the test, a deliberately sized and documented holdout, sufficient duration to capture conversion lag, and clean tracking underneath all of it. But the payoff is a number your finance team can actually trust: not “conversions attributed to this campaign,” but conversions this campaign genuinely caused. In an advertising environment where automation is steadily reducing platform transparency, that causal proof is becoming less of a luxury and more of a baseline expectation.