I Asked ChatGPT to Predict the Best Super Bowl Ads

In partnership with

Hi, marketers in the loop,

I had a wild idea while wrapping up my 2024 AI recap: what if a “standard” AI tool could predict the best Super Bowl ad? Some of you agreed in that edition’s poll that it sounded like a plausible challenge.

So this week, I put that theory to the test using ChatGPT - after trying practically every AI under the sun. My experiment included solutions built for video descriptions, real-time screen-sharing analysis, and advanced reasoning. It was a blast to push boundaries, but also more time-consuming than my usual newsletter writing process. This time, I easily spent twice the usual 8–12 hours.

The good news is that I had an experienced “data analyst” working alongside me - Lex, my ChatGPT-based helper. We had fun collaborating on this special project: predicting this year’s top Super Bowl ads. 😁

Enjoy the read!

- Rei

P.S. I would love to hear from you. Please reply and let me know your thoughts on this week’s newsletter. I read and respond to every message.

In this issue:

  • What’s the Best Super Bowl Ad According to ChatGPT?

  • Too Busy for AI? Think again

  • Why Employees Smuggle AI Into Work

  • And much more

Reading time: 7 minutes

MY FAVORITE FINDS THIS WEEK

Must Read

Events

▪️The AI x Marketing Summit 2025 (March 12-13, San Francisco)

Use code "EARLYREI" for a 10% discount.

Industry News

Industry Reports

AI Agents

Practical Tips

Thought Leadership

DEEP DIVE

I Asked ChatGPT to Predict the Best SuperBowl Ads

What started as a far-fetched idea turned into a full-blown project. After running various prediction exercises with ChatGPT, the ad that scored the highest for the 2025 Super Bowl was:

Lay’s, “The Little Farmer” (Note: I have a 10-year-old daughter and loved this ad)

And here’s the rest of our top 10 lineup:

  1. Lay's, "The Little Farmer" - 6.30

  2. Stella Artois, "David and Dave" - 6.27

  3. Hims & Hers, "Sick of the System" - 6.26

  4. Cirkul, "You Got Cirkul" - 6.24

  5. Booking.com, "Get Your Stay Ridiculously Right" - 6.21

  6. Uber Eats, "Century of Cravings" - 6.21

  7. Pfizer, "Knock Out" - 6.19

  8. NFL, "Somebody" - 6.19

  9. Google, "Dream Job" - 6.18

  10. Haagen-Dazs, "Not So Fast, Not So Furious" - 6.18

How We Got Here

I’m no professional data scientist, but I do consider myself pretty data-driven. Since my skills alone couldn’t handle every layer of this project, I enlisted Lex (my ChatGPT “colleague”) for help. We experimented with different AI solutions, exploring multiple linear models, decision trees, and even neural networks. They did the heavy lifting; I was just the human in the loop, offering moral support.

That said, it’s important to remember that this is an experiment, not necessarily a definitive prediction of the actual Super Bowl Ad Meter results. The goal was to push AI’s capabilities, test different methodologies, and see how well an algorithm could anticipate audience reactions.

Why USA Today Ad Meter?

Each year, the Ad Meter serves as a widely recognized benchmark for Super Bowl ads. I replicated its general approach, covering the same ads their panelists review - though I could only get my hands on 38 out of 55 ads by my cutoff time (1 p.m. PST). Once USA Today announces the official scores, I’ll remove any ads I didn’t analyze so we have a fair comparison. 🤞🏼

ChatGPT - Deep Research

My Process: The Journey and Lessons

  • Understanding USA Today’s Methodology
    I first read up on how the Ad Meter works. Then I collaborated with ChatGPT Deep Research to build a project plan that could mimic the diversity of the voting panel. We crafted various “personas” to reflect different audience segments.

  • Figuring Out How AI Could ‘Watch’ Videos
    Current AI models aren’t great at directly analyzing video content, especially when dialogue is minimal - so I had to get creative. Instead of relying on AI to "watch" the ads, I gathered text-based data from multiple sources and used ChatGPT Deep Research to generate detailed, scene-by-scene descriptions. Each summary ranged from 30,000 to 50,000 characters and took about five minutes to compile, pulling insights from 10–20 sources.

    Each ad summary covered the following key aspects:

    • Narrative Overview: Plot summary and key themes

    • Scene-by-Scene Breakdown: Step-by-step breakdown of visuals and actions

    • Emotional Tone and Atmosphere: How the ad makes viewers feel and its comedic, dramatic, or uplifting elements

    • Visual and Audio Elements: Cinematography, special effects, music, and dialogue analysis

    • Brand Message and Intended Impact: What the ad is trying to communicate and how it resonates with the audience

    • Audience Reception and Industry Reactions: How the public and marketing experts are responding

  • Building a Scoring Model
    Next came the real challenge: constructing a reliable system for scoring ads. We started with the o1 pro model, which crashed frequently. The o3-mini models also struggled, so I ended up creating a custom GPT instance that worked surprisingly well. It was the most consistent and reliable model amongst all tested. After several hours of back-and-forth tweaks - using 2024 ad results as a training set - the model settled on a Gradient Boosting approach, scoring ads on:

    • Emotional Impact (EI) – Does it evoke joy, nostalgia, excitement, or other strong emotions?

    • Entertainment Value (EV) – Is it fun, engaging, and watchable?

    • Brand Integration (BI) – Does the brand fit naturally into the storyline?

    • Storytelling (ST) – Is there a compelling arc or narrative?

    • Originality (OR) – Does it feel fresh and stand out?

    • Cultural Relevance (CR) – Does it tap into timely trends or social moments?

    • Virality/Social Impact (VF) – Will people talk about and share it?

Each factor was weighted based on the model’s training data. We used around 20 older ads for calibration - far from perfect, but still a neat experiment.

For the data enthusiasts out there, the predicted ad score is calculated using a weighted sum of factors, optimized via gradient boosting:

Predicted Score=w1(EI)+w2(EV)+w3(BI)+w4(ST)+w5(OR)+w6(CR)+w7(VF)+ϵ\text{Predicted Score} = w_1(EI) + w_2(EV) + w_3(BI) + w_4(ST) + w_5(OR) + w_6(CR) + w_7(VF) + \epsilonPredicted Score=w1​(EI)+w2​(EV)+w3​(BI)+w4​(ST)+w5​(OR)+w6​(CR)+w7​(VF)+ϵ

Where w1,w2,...,w7w_1, w_2, ..., w_7w1​,w2​,...,w7​ are the optimized weight coefficients for each factor and ϵ\epsilonϵ represents model error (residual).

ChatGPT - GPT

Key Insights: What Works (and What Doesn’t)

Top Drivers of a Great Super Bowl Ad

  • Humor

    • Always a strong crowd-pleaser.

    • Especially effective when combined with celebrity cameos or relatable scenarios.

    • Playful, absurd, or nostalgic humor scores highest.

  • Emotional Appeal

    • Sentimental or inspirational tones resonate deeply.

    • Overly serious ads can backfire if they become heavy-handed.

  • Brand Familiarity

    • Established brands often outperform newcomers.

    • For a lesser-known brand, a big creative hook is essential.

  • Celebrity Endorsements

    • Works if the celeb is a natural fit and not just a gimmick.

    • Authenticity matters - forced cameos rarely pay off.

  • Surprise Factor

    • Plot twists or unexpected turns keep viewers engaged.

    • Shock value for its own sake isn’t enough; it must tie back to the brand.

  • Strong Storytelling

    • Clear beginnings, middles, and endings perform best.

    • Character-driven “mini-movies” stand out in viewer recall.

Common Pitfalls

  • Overly Sales-Driven Spots
    Hard-sell commercials feel like standard ads, which nobody wants during the Big Game.

  • Confusing or Overly Abstract Storylines
    If the message is unclear, viewers will lose interest.

  • Pure Nostalgia Without a Twist
    Throwbacks can work but need to add something new to keep modern audiences engaged.

  • Being Too Polarizing or Edgy
    Viewers generally want fun content; heavy politics or controversial issues risk backlash.

ChatGPT - GPT

Wrapping Up

These scores are just predictions, of course. Real results will vary, especially since we had limited data and used a modest training set. But even if we only get a few ads right, this has been an invaluable test of how far AI-based experimentation can go.

Big Takeaways:

  1. Expect Roadblocks and Stay Resourceful
    Innovation involves creative problem-solving. If one avenue fails, try another.

  2. Model Selection Matters
    Picking the right AI model isn’t always straightforward. Research, test, and pivot until you find something that works.

  3. New Frontiers Are Accessible
    The tools are out there for anyone curious enough to explore and iterate, even if you’re not a coding guru.

ChatGPT - GPt

Ultimately, this was a testament to the opportunity we have ahead - to push the boundaries of data-driven marketing, creativity, and audience engagement. While the approach isn’t perfect, it’s already proving to be a powerful way to uncover insights, test hypotheses, and enhance decision-making in ways that were nearly impossible just a few years ago.

As marketers, we’re just scratching the surface of what’s possible. Whether it’s predicting ad performance, optimizing creative strategies, or personalizing content at scale, new technologies are evolving from futuristic concepts into practical tools that, when used thoughtfully, can drive real results.

Now, I’m curious - what do you think? Could data and analytics one day predict Super Bowl ad winners with near-perfect accuracy? Or do human instincts and emotions make this an unpredictable art form? Reply and let me know your thoughts!

If you’re interested in a deeper discussion, I’m hosting a live community session this Friday to explore how we can apply this approach to test marketing campaigns, strategies, and creative effectiveness. Click below to let me know if you’d like to join, and I’ll send an invite soon. Hope to see you there!

SUPER BOWL AD POLL

Which single factor do you feel matters most for a standout Super Bowl ad?

Login or Subscribe to participate in polls.

Need a personal assistant? We do too, that’s why we use AI.

Ready to embrace a new era of task delegation?

HubSpot’s highly anticipated AI Task Delegation Playbook is your key to supercharging your productivity and saving precious time.

Learn how to integrate AI into your own processes, allowing you to optimize your time and resources, while maximizing your output with ease.

ICYMI

How was today's newsletter?

Login or Subscribe to participate in polls.

Did you like this email? Share this link with a friend for a virtual high five!