Digital marketers operating in the modern era face a persistent challenge: distinguishing between what is happening and what is actually being driven by their specific advertising efforts. This is particularly true for Google’s Demand Gen campaigns, which operate across high-visibility surfaces like YouTube, Discover, and Gmail. While these campaigns are visually stunning and reach users in their most engaged moments, they often suffer from what is known as the attribution illusion.
The attribution illusion occurs when platform reporting shows a high number of conversions, but you are left wondering if those users would have converted anyway through an organic search or a direct visit. To solve this transparency gap, Google launched asset uplift experiments in November 2025. These tests allow advertisers to move beyond guesswork and measure the true incremental impact of their creative assets through rigorous A/B testing. By isolating variables, you can finally determine which videos or images are truly moving the needle and which are simply riding the wave of existing brand awareness.
Why attribution doesn’t equal incrementality
In a standard reporting environment, if a user watches a Demand Gen video on YouTube, doesn’t click, but then later searches for your brand and completes a purchase, Google might assign partial or even full credit to that initial video view. On the surface, the campaign looks like a massive success. However, this is a correlation, not necessarily a causation. The critical question remains: Would that user have made that purchase even if they had never seen the YouTube ad?
Standard attribution models struggle to answer this because they lack a baseline for comparison. This is where incrementality testing—and specifically asset uplift tests—becomes essential. These tests utilize the scientific method by splitting your audience into two segments: a treatment group that sees your specific creative assets and a control group that does not. By establishing what the “natural” conversion rate is for users who aren’t exposed to the ad, you can measure the true “lift” or the additional conversions that were created solely because of the creative impact.
Relying solely on creative instinct or default platform reporting can lead to significant waste. Without incrementality data, you might be funneling your highest creative budgets into assets that look good on paper but offer zero actual lift to your bottom line. Asset uplift tests provide the empirical evidence needed to justify creative spend and optimize for genuine growth.
What you need before testing creative uplift
Launching an experiment without the proper foundation is a recipe for inconclusive results. Before you dive into the Google Ads experiment interface, you must ensure your account and your specific campaign meet several critical prerequisites. Failing to meet these standards often leads to “noise” in the data, making it impossible to reach statistical significance.
Conversion volume requirements
For a test to be statistically valid, the algorithm needs a significant amount of data to compare. Google recommends a minimum of 50 conversions across both the treatment and control arms of the experiment during the testing period. If your primary conversion—such as a completed sale or a high-value lead—doesn’t hit this volume, the test results will likely be labeled as “inconclusive.”
If you find yourself in a low-volume situation, a smart strategy is to optimize the test around high-intent micro-conversions. Instead of tracking “Purchases,” you might track “Add to Cart” or “Check-Out Initiated.” These actions occur more frequently and can still provide a strong signal regarding which creative assets are driving deeper user engagement.
Budget minimums and stability
Consistency is key in any scientific experiment. Your Demand Gen campaign must have an adequate budget to run continuously without being capped. If your campaign hits its daily budget limit and shuts off at 2:00 PM every day, the data for the control group becomes skewed. To get an accurate reading, the campaign should have enough funding to run for at least four weeks without interruption. This ensures that the algorithm can test the assets across different days of the week and times of day, providing a comprehensive view of performance.
Creative isolation
The most common mistake in A/B testing is changing too many things at once. If you change the audience targeting, the bidding strategy, and the video asset all at the same time, you won’t know which change caused the shift in performance. To measure creative uplift accurately, you must isolate the variable. Keep your audiences, bidding models, and standard image assets identical across both arms of the test, changing only the specific creative element you wish to evaluate.
How to run an asset uplift test in Google Ads
Setting up an experiment has become significantly more streamlined within the Google Ads ecosystem. However, the technical ease of setup should not overshadow the need for a disciplined approach. Follow these steps to ensure your test is built for success.
1. Define a clear hypothesis
Before touching any settings, write down exactly what you are trying to prove. A vague goal like “let’s see if this video is better” isn’t a hypothesis. A strong hypothesis looks like this: “By replacing our current corporate brand video with User-Generated Content (UGC) in our Demand Gen asset group, we will see a 12% incremental lift in ‘Sign-Ups’ over a 30-day period.” Having a specific target allows you to evaluate the success of the test with total clarity.
2. Navigate to the Experiments interface
To begin, log in to your Google Ads account and look at the left-hand navigation menu. Go to Campaigns and then select Experiments. From here, click the plus (+) icon to start a new project. You will want to select Asset tests provided by you and specifically designate it as a Demand Gen campaign experiment. This tells Google that you are testing specific creative variations rather than bid strategies or landing pages.
3. Configure a 50/50 split
Google will ask how you want to split your traffic. For the most accurate and statistically sound results, always choose a 50/50 cookie-based split. This method ensures that a single user is assigned to either the control group or the treatment group and stays there for the duration of the test. This prevents “pollution,” where a user might see the treatment ad and then later fall into the control group, which would ruin the baseline data. In this setup, your existing campaign serves as the control, and a duplicated version containing your new assets serves as the treatment.
4. Lock your variables
Once you hit “Start,” you must resist the urge to tinker. It can be tempting to adjust bids or narrow down an audience if you see a spike in costs, but doing so during a test will invalidate your results. Any change made to one arm must be perfectly mirrored in the other, but ideally, you should avoid changes altogether. Treat the testing period as a “lock-down” phase where the algorithm is allowed to work without human interference.
5. Set the duration
The standard recommendation is a minimum of four weeks. This isn’t an arbitrary number; it’s based on how Google’s machine learning operates.
- Week 1: This is the learning phase. The algorithm is figuring out how to distribute the budget within the split and learning which users within your audience are most likely to respond to the new creative.
- Weeks 2-4: This is where the actionable data is collected. Performance during this window reflects the stabilized reality of the campaign.
For industries with longer consideration cycles—such as B2B software or high-end luxury goods—you should consider extending the test to six or eight weeks to capture the full path to conversion.
What your experiment results actually mean
After the test concludes, Google will provide a report in the Experiments dashboard. This report will show the performance of both arms and provide a confidence interval. Interpreting these results correctly is the difference between making a smart business move and making a costly mistake.
Outcome 1: Positive lift (Statistically Significant)
If the results show a positive lift with at least 95% confidence, you have a winner. This means that the new creative asset didn’t just “get conversions”—it drove conversions that otherwise wouldn’t have happened. At this point, you should calculate your Incremental Cost Per Acquisition (iCPA). You do this by taking the extra spend from the treatment group and dividing it by the extra conversions generated over the control group. This iCPA is your new “truth” metric and should be used to justify further investment in that creative style.
Outcome 2: Negative lift
It can be bruising to the ego of a creative team, but sometimes a new asset actually performs worse than the original. This might happen if a video is too polarizing, too long, or has a high skip rate that signals the algorithm to stop showing it to high-value users. If you see a statistically significant negative lift, pause the treatment asset immediately. The experiment has done its job by preventing you from scaling a failing asset.
Outcome 3: Inconclusive result
This is perhaps the most common outcome. If the difference between the two groups is minor and the system cannot find a clear winner, the result is inconclusive. This usually happens for two reasons: either the conversion volume was too low, or the creative assets were too similar. If you were testing a blue “Sign Up” button versus a green one, you likely won’t see a massive lift in Demand Gen. To get a clear signal, you need to test “big” changes—such as a lifestyle video versus an animated explainer. If your result is inconclusive, try running the test for two more weeks, or go back to the drawing board for a more distinct creative concept.
Prove creative impact with incrementality testing
In a world where AI handles much of the bidding and audience targeting, creative assets have become the most powerful lever available to the modern marketer. It is no longer enough to simply produce high-quality video or User-Generated Content; you must be able to prove that those assets are driving business results.
Demand Gen is a formidable tool for visual storytelling, but its value is often questioned by stakeholders who only look at last-click attribution. Asset uplift experiments provide the scientific evidence needed to defend your creative strategy and your marketing budget. By establishing a baseline and measuring the true lift of your visuals, you transition from “guessing” what works to “knowing” what grows. Start with a single holdout test, let the data speak, and use those insights to build a high-performance creative roadmap that delivers genuine incremental value.