Back in July, digital media verification firm DoubleVerify published a blog article about the importance of recipe sites posting real photos and not AI-generated images. The firm argued that while cooking and recipe sites offer valuable advertising opportunities, advertisers should consider the low-quality signals AI imagery displays and that there is value in talking with partners to ensure there’s brand alignment.
Since the blog post was published, we have seen the launch of OpenAI’s new video generation tool SORA 2, which furthers the question about how people engage with AI generated imagery.
In talking with DoubleVerify’s Chief Innovation Officer Jack Smith, there was one question that was top of mind:
Do most people actually care if an image or video is fake?
This is a question especially relevant when it comes to food photography, which has a long-held reputation of images being manipulated and manufactured by photographers. Is AI generation images of delicious-looking food actually a problem?
“I’ve said for a long time that even if you go back to Ansel Adams, you’re looking at [a photo of] Yosemite, and you go to Yosemite, it doesn’t look like that. So there’s always been this notion that photographers would manipulate images,” Smith said, evoking the name of one of America’s earliest, most iconic, landscape photographers.
Ansel Adams photography of Yosemite
“[In regards to food photography] I think it’s a little bit different here where there are pros and cons, right? It’s the ability for non-professionals to create visual content of high quality. On the other side of that, it’s also created an explosion of AI slop.
“Really terrible content that might look on the surface as being good, but it’s just more garbage. And the ability for someone to scale that up has become easier and easier. As SORA 2 or SORA 3 or SORA 10 comes out, the quality of the images are going to get better, but that doesn’t mean that the script is necessarily going to get better at that rate, or that it won’t be just more AI slop trying to grab the attention of consumers.”
While images that look immediately like an AI generation might frustrate media-savvy people like Jack Smith and the editor of Mediaweek, there are a lot of people who aren’t bothered by the fakery of an AI-generated image. For advertisers who are equally concerned, by avoiding sites and channels that publish images and videos with AI slop, is there a risk of losing access to audience segments who just don’t care?
“You’ve got to remember, there’s a value exchange between the publisher and the consumer: I will give you my attention if you give me quality content,” Smith said.
“It becomes a question, if you’re a marketer, of how much you’re willing to pay for the difference in consumer experience. And I would draw a line between AI-generated content and AI slop. It doesn’t mean that it’s necessarily bad, but there is a consumer experience issue there where I might be willing to put up with it, and an advertiser might also be willing to put up with it.
“The pure scale of it and the awfulness of some of that content means that many consumers are starting to reject it because they can see the difference. Maybe the issue isn’t so much that you’re avoiding AI-generated slop, but it’s about the value that’s being attributed to the sites.”
Jack Smith – Double Verify
With the launch of SORA 2, media consumers are about to be challenged by all manner of ethical concerns when it comes to discussing goods within a marketing context. When YouTube began to go mainstream, pretty early on we saw the rise in content like unboxing videos and product reviews. How does that change when content creators start publishing similar content with products they have never actually physically laid hands on?
“Fake reviews have happened since the early days of the internet, and the changes in technology just allowed to scale that,” Smith said.
The public acceptance of content like AI generated product review videos will again come back to media literacy, which may be easier for younger consumers who grow up in the world of AI generated video. This raises a pretty big question that all marketers need to start considering…
If we are in a wild west where we are all relying on an audience to have strong media literacy and to be able to instinctively pick good from bad content, we’re heading into an ecosystem where former considerations of quality no longer apply and brand safety is less of a concern
“Yeah, it’s a good question. I think there’s still brand safety and suitability, which largely you could call quality, right? And I think the definition of that’s expanded.
“Brand safety to me would be things that no advertiser wants to run on and really shouldn’t be monetised – like terrorist videos or certain kinds of violence. And then you’ve got suitability, which is more about trying to understand how to match your brand’s values with the environment that you’re running in. It used to be that it was all about… is this site medium risk sexual or low risk sexual? Now it’s become quite a bit more about that.”
As we consider media literacy and the audience understanding of the content they are consuming, hip hop music becomes an easy way to think about it. It’s not a new idea to talk about artists sampling and remixing former recorded music to create new works. The audience knows that these are remixed works, but the authenticity comes in part from the narrative of the artist. And so the music fan is engaging with not just the music itself, but all the additional textual elements to it.
Smith, a music fan, is quick to jump on this idea and bring it back to the role of AI generated work in recipes. “The judgement comes down to: if you make the recipe, does it taste good or not? Many recipes are adapted from other recipes. You take an existing recipe, and this is before digital even, you take an existing recipe and you modify that. You take your mother’s recipe for meatballs, and you modify that, and that continues to be modified over time.”
When considering whether you are buying media against channels using AI generated works, that’s ultimately the question one should be asking: “Does it taste good or not?”