Ready or not, generative AI startups are harnessing the power of AI to make ad creative. You’ve probably got some questions.
How can marketers figure out if these startups work for them? What are the hidden costs? The legal risks of using models trained on copyrighted material?
We’ve got answers. Our senior editor Hana Yoo distilled the complex space of generative AI startups for ad creative, and shares her findings on this week’s podcast.
We also discuss the “dueling trends” in AI startups Hana observed: specialization and generalization. Many AI startups focus on just one channel, like video (Runway), audio (AudioStack) or social media clips (Munch). But others are trying to take a marketer’s ad creative and adapt it for all the places it needs to show up (e.g. Typeface, Jasper, Pencil).
Even within certain startup categories, like image generation, some tools will work better at some requests than others. A model trained on classic works of art may logically be better at recreating a medieval tapestry, while a model trained on stock photos may be just the ticket to create more of the same.
Then, we pore over the advertising details revealed in Reddit’s S-1. The startup is preparing for a $5 billion IPO this spring, and it opened its books in advance of going public. While Reddit’s ad revenue makes it large compared to digital publishers, on the social platform side, it’s dwarfed by the giants that are Google and Meta (and even Pinterest and Snap).
Reddit is also a test case for how generative AI and social platforms could align. It just struck a $60 million per year deal with Google that includes access to powerful Google machine learning tools, but also gives Google access to its data to train its generative AI systems.
Plus, Reddit’s own ad platform focuses on using contextual signals over targeting users (after all, there’s a reason you share your real name on Facebook but an anonymous handle on Reddit), and Reddit’s IPO will foreshadow the success of this approach as many digital media publishers double down on a contextual approach as cookies fade away.
