Google Display Ads have only a few seconds to attract attention, communicate a message, and guide users toward an action. Small design decisions – from headline size and spacing to CTA placement and background graphics – can significantly change what viewers notice first.
But optimizing a display ad is not simply a matter of making every element more noticeable.
In this case study, we used Attention Insight’s predictive eye-tracking technology to optimize a Google Display Ad for a live webinar, The Future of AI in Marketing. Across four design iterations, we tested how changes to visual hierarchy, element visibility, contrast, and complexity affected predicted attention.
The process increased the Visual Performance Score from 54 to 70, while producing a much clearer hierarchy between the main message, offer details, and primary CTA.
More importantly, the experiment revealed something useful about how to optimize Google Display Ads: improving one metric does not necessarily improve the design as a whole.
The Goal: Optimize a Google Display Ad for Registrations
The banner promoted a free live webinar titled The Future of AI in Marketing.
Its primary conversion goal was straightforward: encourage users to click REGISTER NOW.
We therefore defined three key Areas of Interest (AOIs):
- CTA Button – “Register Now”
- Main Message – “The Future of AI in Marketing” headline
- Offer Details – the webinar date and time
The objective was not to distribute attention equally between these three elements. A successful ad needed a clear hierarchy: users should first understand the main message, then register the supporting event information, and finally notice the CTA.
We used CTA Visibility as the analysis goal and tested each iteration using Attention Insight.
Why Predictive Attention Matters for Google Display Ad Optimization
Traditional design review can tell us whether something looks prominent, but that does not necessarily tell us where visual attention is likely to go.
Predictive eye-tracking provides another layer of evidence.
Attention Insight generates AI-powered attention heatmaps and predicts how viewers are likely to distribute their visual attention before a campaign goes live. Instead of relying only on subjective judgment, designers can compare iterations using predicted attention patterns and measurable Areas of Interest.
For this experiment, we looked at both individual AOI percentages and the broader Visual Performance Score, which evaluates the design across several categories:
- Element Visibility
- Visual Hierarchy
- Attention Focus
- Contrast
- Visual Complexity
- Text Load
This distinction became particularly important during the optimization process.
Setting Up the Google Display Ad Analysis
Before making design changes, we uploaded the original 300 × 600 Google Display Ad to Attention Insight and selected CTA Visibility as the analysis goal.
We then defined three Areas of Interest according to their role within the ad:
Primary CTA – Critical
Main message – Major
Offer Details – Minor
Defining these AOIs allowed us to measure how predicted attention was distributed between the key elements rather than judging their visibility only by eye.
Reading the Attention and Score Evaluation Results
Once the design is analyzed, Attention Insight provides several layers of information.
The heatmap shows where visual attention is predicted to concentrate, while Percentage of Attention measures how much predicted attention falls within each defined AOI.
The platform also provides Insights that compare individual AOIs with reference values. For example, an Insight may indicate that CTA visibility is lower or higher than the average for similar elements.
These messages provide useful context for the percentages: instead of seeing only that an element received a certain share of attention, we can understand whether its predicted visibility is relatively strong or weak for its role.
The broader Score Evaluation then assesses the composition across Element Visibility, Visual Hierarchy, Attention Focus, Contrast, Visual Complexity, and Text Load.
Starting Point: A Visually Competing Banner
The initial banner contained all the information users needed, but several elements competed for attention.
The headline, date and time, CTA, cyan glow, purple mesh graphic, and additional supporting text all contributed visual weight. Although the design was visually engaging, the attention path was less controlled than we wanted.
The initial analysis produced a Visual Performance Score of 54.
The predicted attention distribution across the three key AOIs was:
| Area of Interest | Attention |
|---|---|
| Main Message | 43% |
| Offer Details | 8% |
| Primary CTA | 8% |
The headline was already attracting substantial attention, but both the offer details and CTA were comparatively weak.
The broader evaluation supported the same conclusion. Element Visibility, Visual Hierarchy, Contrast, and Visual Complexity were marked Critical, while Attention Focus still needed improvement.
The challenge was therefore not simply to make the headline stronger. We needed to improve the supporting information and CTA without destroying the hierarchy that was already forming around the main message.
Iteration 1: Improving CTA Visibility
The first round of optimization focused on reducing unnecessary competition around the key conversion elements.
We simplified the composition, adjusted the supporting graphics, and gave the central content more breathing room. The aim was to make the CTA and event information easier to process without adding more visual noise.
The result improved the Visual Performance Score from 54 to 60.
Attention was distributed as follows:
| Area of Interest | Attention |
|---|---|
| Main Message | 35% |
| Offer Details | 8% |
| Primary CTA | 17% |
The most significant improvement was the CTA.
Predicted attention increased from 8% to 17%, more than doubling its share of attention. Visual Hierarchy and Contrast also passed in the broader evaluation.
However, the Offer Details remained weak at only 8%.
This illustrates an important point when optimizing display ads: strengthening one element can redistribute attention rather than simply adding attention everywhere.
Iteration 2: Strengthening the Supporting Information
The next iteration focused more directly on the event details while preserving the improved CTA visibility.
We refined the date and time treatment, adjusted spacing and hierarchy, and continued reducing visual competition around the core message.
The Visual Performance Score increased again, reaching 62.
This version produced:
| Area of Interest | Attention |
|---|---|
| Main Message | 38% |
| Offer Details | 19% |
| Primary CTA | 16% |
At the AOI level, this looked extremely promising.
The Offer Details jumped from 8% to 19%, while the CTA retained 16% of predicted attention. All three AOIs now performed strongly.
However, the broader design evaluation revealed a different picture.
Despite the strong individual AOI results, the design still showed weaknesses in Visual Hierarchy, Attention Focus, and Visual Complexity.
This version was therefore an important intermediate step rather than the finished design.
Strong AOIs Do Not Automatically Mean a Stronger Design
It can be tempting to optimize until every Area of Interest produces the highest possible percentage.
But visual design does not work that way.
Attention is finite. When one element becomes more prominent, attention may shift away from another. Increasing the visibility of multiple elements can also make the composition more competitive and weaken the overall hierarchy.
Predictive eye-tracking results therefore need to be interpreted in relation to the intended hierarchy of the design, rather than treated as independent scores that should all be maximized.
This iteration demonstrated that clearly: its individual AOIs performed well, yet the complete design was still less balanced than it could be.
That led to one final round of refinement.
Final Design: Optimizing the Overall Attention Hierarchy
For the final version, we focused less on maximizing every individual AOI and more on creating the strongest overall attention structure.
We refined the headline treatment, adjusted the date and time block, maintained a clearly separated CTA, and reduced unnecessary visual competition. The decorative wave remained toward the bottom of the banner, supporting the composition without competing heavily with the main content.
The result was the strongest overall version.
The Visual Performance Score increased to 70, moving the design into Strong Visibility.
The final predicted attention distribution was:
| Area of Interest | Attention |
|---|---|
| Main Message | 41.9% |
| Offer Details | 16.1% |
| Primary CTA | 19.8% |
Compared with the previous iteration, Offer Details received slightly less attention, but the CTA increased and the broader design became substantially stronger.
Most importantly, five of the six evaluation categories passed:
| Evaluation Category | Final Result |
|---|---|
| Element Visibility | Critical |
| Visual Hierarchy | Passed |
| Attention Focus | Passed |
| Contrast | Passed |
| Visual Complexity | Passed |
| Text Load | Passed |
Only Element Visibility remained Critical because the Offer Details were still below the system’s average visibility benchmark.
That did not mean the final design was unsuccessful.
In fact, the hierarchy was now much closer to the intended communication sequence:
Main message → supporting event details → primary CTA
The headline commanded the largest share of attention, while the CTA became the second-strongest actionable area.
Using AI Recommendations and Generate Design Variant
Attention Insight’s AI Recommendations identified the remaining weakness directly: Offer Details had below-average visibility.
The recommendations suggested increasing the prominence of the date and time and adjusting its visual relationship with the CTA.
Importantly, this was a relatively isolated issue. In the final 70-point design, Visual Hierarchy, Attention Focus, Contrast, Visual Complexity, and Text Load had all Passed. Rather than suggesting that the entire composition needed to be redesigned, the results pointed to one remaining area that could potentially be improved.
The platform’s Generate Design Variant feature could then visualize a possible solution based on those recommendations.
This extends the workflow beyond simply identifying a problem:
Insight → Recommendation → Generate Design Variant → Designer evaluation → Retest
The generated variant provided another direction to consider, but we treated it as a suggestion rather than an automatically improved final design.
AI-generated recommendations can be useful for exploring changes to hierarchy, contrast, positioning, size, or prominence. However, the output still needs to be assessed using design judgment.
In some generated directions, elements can become exaggerated simply to increase visibility. A larger or more prominent element may improve one predicted metric while making the overall composition less balanced.
The most useful approach is therefore not:
AI recommendation → accept
but:
AI recommendation → evaluate → adapt → test → compare.
Testing the Recommendation: Why a Higher Individual Metric Wasn’t Better
To test whether the remaining Element Visibility issue should actually be corrected, we created another variation in which Offer Details were made slightly more prominent.
The targeted metric improved:
| Area of Interest | Final Design | Additional Test |
|---|---|---|
| Main Message | 41.9% | 39.5% |
| Offer Details | 16.1% | 16.5% |
| Primary CTA | 19.8% | 18.4% |
Offer Details increased from 16.1% to 16.5%.
But the overall result moved in the opposite direction.
The Visual Performance Score fell from 70 to 68. Visual Hierarchy moved from Passed to Needs Improvement, while Attention Focus became Critical.
So although the specific element we were targeting improved slightly, the overall design became weaker.
We therefore rejected this additional test and retained the 70-point version as the final design.
This was an important validation step. Without retesting, it would have been easy to assume that increasing Offer Details visibility automatically made the banner better.
It didn’t.
This demonstrates why optimization should not become an attempt to force every individual metric into the green.
Comparing the Four Google Display Ad Versions
Looking across the four main versions shows how the banner evolved:
| Version | Visual Performance Score | Main Message | Offer Details | Primary CTA |
|---|---|---|---|---|
| Original | 54 | 43% | 8% | 8% |
| Iteration 1 | 60 | 35% | 8% | 17% |
| Iteration 2 | 62 | 38% | 19% | 16% |
| Final Design | 70 | 41.9% | 16.1% | 19.8% |
The progression was not linear across every metric.
Iteration 2 (Google Banner 3), for example, achieved the highest Offer Details percentage at 18.9%. But its broader visual evaluation still identified weaknesses in hierarchy, focus, and complexity.
The final design reduced Offer Details attention slightly to 16.1%, while achieving the strongest overall Visual Performance Score, passing five of the six evaluation categories, and increasing Primary CTA attention to 19.8%.
Looking only at individual AOI percentages could therefore have led us to choose the wrong version.
What We Learned About Optimizing Google Display Ads
This experiment demonstrates why Google Display Ad optimization should involve more than simply making important elements bigger, brighter, or more visually dominant.
The strongest design was not the version in which every individual AOI reached its highest value.
Instead, it was the version in which the elements worked together most effectively.
1. Attention is a hierarchy, not a collection of independent scores
Increasing attention to one element can reduce attention elsewhere. That redistribution is not automatically negative.
The relevant question is whether attention is being distributed according to the intended communication hierarchy.
2. Individual AOI results need broader design context
The 62-point iteration produced strong AOI percentages, particularly for Offer Details, but the wider evaluation still identified problems with hierarchy, focus, and complexity.
The final version had slightly lower Offer Details attention but performed substantially better as a complete composition.
3. More attention is not always better
Our additional optimization test increased Offer Details attention from 16.1% to 16.5%, but the overall score dropped from 70 to 68.
A successful ad does not need every element to attract maximum attention. Some elements should dominate; others should support them.
4. Every optimization should be validated
Even a logical design change can have unexpected consequences.
Without retesting the final adjustment, we might have assumed that improving the remaining weak AOI automatically improved the banner.
Predictive attention testing showed otherwise.
From a 54 to 70 Visual Performance Score
Across four main design versions, the Google Display Ad improved from a 54 Visual Performance Score to 70.
The final design achieved:
- 41.9% attention on the Main Message
- 19.8% attention on the Primary CTA
- 16.1% attention on the Offer Details
- Passed Visual Hierarchy
- Passed Attention Focus
- Passed Contrast
- Passed Visual Complexity
- Passed Text Load
The remaining Element Visibility warning was useful rather than something that needed to be eliminated at any cost. When we tried to push that metric further, the overall design became weaker.
That is perhaps the most important lesson from the experiment:
Predictive attention optimization is not about turning every metric green. It is about using data to make better-informed design decisions while preserving the hierarchy required by the communication goal.
Using Predictive Attention to Optimize Google Display Ads Before Launch
Google Display Ads operate in crowded visual environments where advertisers have very little time to communicate.
Predictive eye-tracking gives designers and marketers a way to evaluate likely attention patterns before spending media budget on a campaign.
With Attention Insight, we could test multiple design iterations, define specific Areas of Interest, compare predicted attention percentages, evaluate broader visual performance, receive AI Recommendations, explore an AI-generated design alternative, and validate subsequent changes.
That last step is crucial.
AI can identify weaknesses and suggest solutions, but design optimization still requires interpretation. The best-performing result may not be the version with the highest score for every individual element.
For this banner, the strongest result came from balancing message visibility, supporting information, CTA prominence, and overall visual hierarchy.
The final result was not the banner in which every element received the most attention. It was the banner in which attention worked together most effectively to support the goal.
And that is ultimately what optimizing a Google Display Ad should achieve: the right attention, in the right place, in the right order.
