Like some malfunctioning robot that’s crashed a swish party, broken the furniture and been sick on the shagpile, AI is once again causing a stir across adland.

But while the ad industry’s AI reckoning dominated chatter across the Croisette last week – from slick panel sessions to cosy rosé-fuelled debates – the waves were not coming from the warm, yacht-speckled waters of the South of France. They came from a basketball court in the heart of middle America.

During Game 3 of the NBA Finals, a brash, bizarre, entirely AI-generated ad beamed out to millions of viewers via YouTube TV. It was part Grand Theft Auto trailer, part late-night cheese-induced dystopian fever dream: a man riding an alligator in a kiddie pool; aliens pounding beers; people betting on hurricanes.

The AI-generated ad, aptly titled “World Gone Mad”, was for prediction market Kalshi, a platform that lets you “trade on anything, anywhere in the US”, including Oklahoma City and Indiana, where both NBA Finals teams play. Its “prime-time” debut sparked a flurry of headlines proclaiming that AI ads had finally gone mainstream. 

But while betting on anything from the price of eggs to the weather may feel like another unwanted symptom of the current political and cultural climate, the same could not be said for its rather predictable reception.

Scan the comments under the video – which AI filmmaker PJ Accetturo claims he created in just two days for only $2,000 – it’s fair to say many viewers were less than impressed. “Garbage made by garbage”, “This doesn’t look real at all” and “AI slop” are just some of the responses. 

However, others were a lot more positive, suggesting the campaign could be a game-changer – “Pandora box is opened” and, “just like that, my job is void.”

So which is it? Well, loud, chaotic, and unapologetically weird, the ad is definitely a slight departure from the bloodless AI sausage we’re usually served up by our machine overlords-in-waiting. 

But to try and put some numbers behind what we are saying and compare it to other ads, we used our human-trained, AI-powered creating testing platform to measure its effectiveness, including predicted emotions, attention and brand impact. 

Now, I know what you are thinking: AI testing AI. One machine grading another’s homework. What kind of dystopian nightmare is this? Well, our algorithms are trained on tens of millions of human responses to ads from over the years. So while it sounds like a Black Mirror subplot, it’s a robust way to benchmark creative performance.

Anyway, let’s take a look at the insights.

One of the biggest criticisms aimed at AI-generated ads in the past has been that they feel a little clean, leaving viewers feeling a little cold. 

However, the same charge could not be levelled at “World Gone Mad”, which generated intense emotional responses from viewers, although admittedly mostly negative.

Negative emotions were 18% higher than the industry average, while of the 17 emotions that overindexed, 13 were negative. For example, viewers were 66% more likely to feel awkward than the average ad, 63% more likely to be disgusted, 42% more likely to feel intense guilt and anger, 31% more likely to feel distrustful, 27% more likely to feel embarrassed and 26% more likely to feel fear. Other negative emotions that were above the norm include anxiety (+18.5%), shame (+14%) and horror (10.9%).

Not exactly a recipe for ad success. But at least people felt something and weren’t bored (12% less likely to be bored than average, to be exact). 

Predictably, on the flipside, positive emotions were slightly below the norm (-5.6%), with amusement (+27%) – an incredibly difficult emotion to get right – one of the standout emotions, alongside excitement (+9%), entrancement (+2.6%) and knowledge (+1.1%). Feelings of warmth and trust were both way below the industry average. 

With so much going on on screen, it’s no surprise that feelings of confusion were also high – 54% more than the average ad – which in turn affected attention levels, which were slightly below the norm throughout. People switched off because they found it hard to understand what they were watching. This in turn meant people were less likely to act after watching, with purchase intent and brand recommendations all below average. 

But at least they knew which company was behind the ad – with correct brand recall 25.7% higher than the norm.  And here’s the uncomfortable truth: for all its glitchy weirdness, the ad outperformed 54% of the ads we’ve tested.

Using our Creative Effectiveness Score – a composite metric that measures overall effectiveness – it scored 5.97 out of 10. Not bad for something made in someone’s bedroom in a fraction of the time.

It may be easy for the industry to dismiss campaigns like “World Gone Mad” with sniffy “AI slop” comments while sipping Chablis in the French sun and handing out trophies to itself, high on the perfume of its own success. But behind all the eye-rolls is a nagging fear: these scrappy, glitchy AI ads are already outperforming a lot of the high-budget work we celebrate.

We’re still a long way from AI-generated ads being spoken about in the same hallowed breath as Grand Prix winners – but the gap is closing. And, let’s be honest, the bar isn’t always as high as we like to think it is. We recently ran some analysis of thousands of social ads from the last 12 months to see how they performed against industry norms – and the results were quite shocking.

Over half (53%) were more boring than the industry average, while nearly half (47%) overindexed for negative emotions. In another test, we compared some of the most boring videos we could find on the internet with industry averages. What we found was 28% were less interesting than a video of paint drying and 14.5% generated less attention than a 10-minute video of just a black screen – with no sound.