Gen-AI is coming, but marketers need to remember that efficiency should never be at the expense of effectiveness, say Kantar’s Ecem Erdem, Global Creative Thought Leadership Manager, and Vera Sidlova, Global Creative Thought Leadership Director.

Generative AI (GenAI) brings the potential to unleash creativity as well as create advertising content with unparalleled efficiency and scale. Kantar has been exploring and experimenting with these new tools, as have marketers and agencies.

We wanted to explore how involving GenAI in the creation process could potentially improve the effectiveness of digital advertising. We did this by testing some fully AI-generated ads, along with others where GenAI played a partial role (i.e., writing the script or creating imagery), using Link AI for Digital – our AI powered ad-testing tool which uses data from 250,000 real-world ad tests to give indicative effectiveness results in as little as 15 minutes. In other words, we let our machines score the output of their fellow machines.

But ads created with GenAI are not that easy to find. There are still relatively few ads created with at least some level of GenAI involvement, so we selected nine ads to explore, from high production value ads such as Masterpiece for Coca-Cola, to experimental ones such as Pepperoni Hug Spot that many will remember by its infamous tagline, “Like family, but with more cheese.”

Lesson 1: GenAI ads performed strongly, but quality was variable

Kantar’s Demand Power Contribution score measures the long-term potential of an ad to drive meaningfulness, difference and saliency for the brand. Of the nine ads we tested, six performed better than average by this metric and two of the three fully AI-generated ads were in the top 30%. That said, there’s a wide spread of results, with no particular pattern as to whether the ad was entirely or partially AI-generated.

Demand Power Contribution percentiles

Demand Power Contribution percentiles

Lesson 2: There’s more than one route to success

As with any new tool, the key to success is how you use it. Of the ads we tested, three stand out as good examples of the different approaches that can be taken to using AI: they had potential to build the brand in the long term. Pepperoni Hug Spot is a fully AI-generated ad for a hypothetical Pizza chain that is a product of the creator’s efforts to explore the world of AI-driven video. Masterpiece for Coca-Cola features a human-created script showing the journey of a bottle of Coca-Cola from canvas to canvas in an art gallery, where AI brought these famous paintings alive. Driven by intuition for Lexus has a story created by AI which was trained with previous award-winning luxury adverts, and brought to life by Oscar-winning director Kevin Macdonald. All three had Demand Power Contribution scores in the top 30% of ads tested by Kantar.

Lesson 3: AI can assess different creative routes and variants as well as generate content

Among the ads we tested, two were for the same brand, from the same campaign, and both fully AI-generated. We tested both routes in context for Facebook, and the results highlight how AI-powered ad testing could quickly help marketers select the route with more potential for certain online platforms. Our Link AI for Digital solution found that one of the ads performed much better than the other in terms of enjoyment, persuasion and interaction – it’s easy to see how this could save time and budget in a real-world setting.

Considering the budget and time pressures around digital campaigns, AI can again prove to be a great ally in asset creation here, making it possible for agencies to explore and produce different versions of the same creative execution at speed and at scale. This is where AI-powered testing can also play a significant role – helping marketers ensure machine-created ad versions are optimised for the right channel.

Lesson 4: The technology is good, but it’s not perfect

Some examples point to AI’s limits: Ryan Reynolds experimented with a ChatGPT script in his recent ad for Mint Mobile. Delivered with Reynold’s characteristic charm, leaning into the uncanny valley-ness of the script, it appears to be a good example of incorporating AI in a knowing way. But when tested using Link AI for Digital, the ad was rated as among the bottom 30% for enjoyment. On the other hand, the machine-generated slightly nonsensical conversational humour in Beans are back, an experimental fake ad that was fully created by generative AI, was placed top 25%. This suggests AI is not able to parse nuanced humour – at least, not yet.

Some fully AI-generated ads with odd imagery were also high performers, suggesting these executions are good content in theory. But our human eyes get rather stuck on odd visual details, such as people smashing pizzas into their faces. Clearly at this point machines do not know if the people depicted have extra fingers or monstrous smiles, but just recognise that they are portrayed as enjoying the product.

The common consensus on the use of AI in advertising is that we need to have humans at the centre of the creative process. And it’s not hard to see why. There is no doubt that AI will continue to be a celebrated tool in advertising, one that brings efficiency and enhances creativity.

However, a word of warning: producing content more easily will also bring the temptation to produce high quantities of stuff, and marketers will need to keep their eye on quality. Producing a lot more advertising could also mean running the risk of just adding to clutter and not delivering on business objectives.