When OpenAI and Google both used the Super Bowl stage in 2025 to tell an “AI story,” they took noticeably different creative directions. OpenAI’s spot, ChatGPT The Intelligence Age, presents a broad, conceptual narrative about technological progress. Google’s Pixel Dream Job tells a much more personal story, anchored in everyday family moments and a clear emotional arc — a classic example of emotional advertising.
That contrast made the two ads a compelling case for comparison. Both in terms of which ad was liked more, and in how viewers responded while watching, scene by scene. Those moment-to-moment reactions help explain changes in brand perception.
To explore this, we ran a structured A/B-style study in FaceReader Online, combining survey questions with facial coding. This allowed us to compare what viewers reported after each ad with how their facial expressions evolved during viewing. This type of mixed-method approach is common in neuromarketing research, where researchers use explicit ratings and implicit measures together to strengthen ad effectiveness measurement.
Facial Coding Study Setup
FaceReader Online supports the full research workflow, which allowed us to apply a clear and consistent study design across both ads:
- Participants watched both videos in randomized order.
- Participants rated brand favorability before and after each ad on a 7-point scale.
- Ad liking was rated after viewing each video.
- Facial expressions analysis was continuous during playback using webcam-based facial coding.
In total, 105 participants were recruited via FaceReader Online’s built-in Prolific workflow. The total sample includes, after automatic quality checks, 99 unique participants, resulting in 195 successful recordings (98 for the OpenAI ad and 97 for the Google ad).
Survey Results: What Viewers Reported
The survey results already revealed a clear difference between the two ads. On ad liking, Google Pixel Dream Job scored 5.53 on the 7-point scale, while ChatGPT The Intelligence Age scored 4.72. That gap is large enough to suggest a meaningful difference in ad experience.
We also examined how brand favorability changed from before to after each ad within the same viewers. On average, Google’s favorability score increased by 0.53 points, while OpenAI’s increased by 0.05 points. Looking at distributions tells the same story. OpenAI started from a higher baseline, with 68% of viewers rating the brand above neutral (>4) before the ad and 73% after. Google started lower, at 56% above neutral before the ad, but rose to 72% after. In practical terms, OpenAI began ahead, while Google made up most of the difference after viewers had seen the ad.
Facial Coding Results: Sentiment And Timing
Facial coding adds an important layer of context to these outcomes. When looking at overall valence, both ads showed negative averages, with OpenAI more negative (-0.094) than Google (-0.056). This is not unusual for ads with a serious or reflective tone. Average valence tends to be dominated by neutral and focused expressions, while positive or negative responses appear as brief peaks tied to specific scenes.
Those scene-level dynamics are where the two ads diverged most clearly. When we examined happy expression over time, Google’s ad showed several pronounced peaks, while OpenAI’s timeline was comparatively flat. In this context, “happy” does not only capture laughter; it also reflects warm smiles, empathy, and positive social moments. For Google, the strongest peaks aligned closely with the narrative structure of the ad: early scenes around the parent–child relationship, and later scenes centered on growing up and letting go, including a strong hug and a college goodbye. OpenAI’s ad showed fewer moments where positive expression clearly stood out.
Audience Differences: Age Attitude
To understand whether these patterns were consistent across audiences, we looked at age and brand attitude as segmentation variables.
Using the platform’s automated age analysis, we grouped viewers into a younger segment (Gen Z and Millennials, late teens to early forties) and an older segment (Gen X and Baby Boomers, mid-forties and up). For the Google ad, younger viewers showed higher average happy responses than older viewers. For OpenAI, the pattern was reversed, with older viewers showing slightly more positive facial responses than younger viewers. Importantly, this doesn’t mean one group liked one ad and disliked the other; rather, it shows that the emotional journey of each ad played out differently depending on who was watching.
This also showed up in the automatically detected scene highlights. For older viewers, the strongest positive response occurred in an earlier moment of the Google ad, around the reading scene, while younger viewers showed their strongest positive response later in the ad, during the goodbye and growing-up scenes. That’s a useful reminder that the same ad can have different “best moments” depending on who is watching.
Audience Differences: Brand Attitude
We also segmented viewers based on their attitude toward each brand before the ad. Participants were grouped into those with a positive brand attitude (ratings 5–7) and those with a neutral or negative attitude (ratings 1–4). Viewers who were already positive toward a brand tended to like that brand’s ad more and showed more positive facial expressions during viewing. Facial coding doesn’t replace this survey insight, but it reinforces it by showing the difference in response and can show how those differences unfold across the timeline of the ad.
From Outcomes to Insights
Overall, the results point to a consistent pattern. Google’s Pixel Dream Job performed better on ad liking and showed a larger within-sample increase in brand favorability. Facial coding helps explain why: the ad contains a small number of scenes that reliably generate positive emotional responses, and those moments resonate particularly strongly with younger viewers. OpenAI’s The Intelligence Age started from a more favorable brand position, but its more abstract framing produced fewer standout emotional peaks and resulted in only a modest shift in favorability.
Key Survey and Facial Coding Takeaways
Taken together, this comparison shows how combining survey data with facial coding helps move from simple outcomes to clearer creative insights. Surveys tell you what changed; facial expression analysis helps explain when and why those changes occurred. This is especially relevant for emotional advertising, where a small number of scenes can carry most of the impact.
- Survey measures capture outcomes such as ad liking and changes in brand perception.
- Facial coding adds diagnostic value by showing when emotional responses occur and which scenes drive them.
- For serious or narrative ads, average sentiment can be misleading; scene-level peaks are often more informative.
- Audience segmentation matters: age and prior brand attitude can change which moments resonate most.
FaceReader Online supports this workflow end to end, from setting up studies with video stimuli and questions, to recruiting participants, filtering and segmenting results, and automatically highlighting the scenes that drive emotional responses. That combination makes it possible to move beyond scorecards and toward a clearer understanding of why an ad works the way it does — and where its impact really comes from.