Published in March 2023, this article examines early generative-AI advertising use cases and limitations. The brand examples and model screenshots belong to that period. For a creative team evaluating a tool today, the practical questions are whether the output represents the product accurately, can be used with the necessary rights and helps test a useful message.
Table of Contents
What is Generative AI?
Let’s start with a quick explainer. Generative AI is a type of machine learning that creates new outputs (content, ad copy, images, etc.) based on the input parameters provided – usually named prompt. For a generative AI model to do this, it needs to be trained on an extensive data set to understand the input and produce something original based on training data. Generative AI has been making headlines recently, with tools like Midjourney, ChatGPT, and Stable Diffusion going viral on social media. Major brands have also been utilizing these tools. For example, the real estate marketplace Zillow uses the generative AI tool Jasper to help produce blog posts and other content assets. When Heinz launched its Draw Ketchup campaign, it used DALL-E 2 to create a range of AI-generated ketchup bottle images.
The clothing company Stitch Fix even uses generative AI to create visualizations of clothing based on consumer preferences to inform new product development. For example, shoppers can enter their preferred color, fabric, size, and other attributes and see a visualization of the product.
How Generative AI Will Change the Advertising Industry
The best way to see the impact is to look at the different stages of the advertising process.
Ideation
One of the most obvious use cases is coming up with ideas and concepts for campaigns and ad creative. Using generative AI, you can automate the ideation process by using a model to create new ideas or variations based on previous winning campaigns or competitor campaigns. Generative AI tools like Stable Diffusion generate new campaign ideas and speed up the ideation process. The text-to-image model creates realistic images based on text input.
Creation
Content and ad copy creation is another area where generative AI will play a significant role. Instead of needing a large team of copywriters and designers, a small team working with generative AI tools can produce an extensive range of creative assets every week. It’s important to note that human creativity and intervention will still be needed in the creation process. Across all industries, companies typically achieve optimal results when AI is used to augment rather than completely replace human intervention. With advertising creative, you will still need to make sure branding is consistent, and copy and concepts are original.
Testing
AI can suggest test ideas from prior work, but a suggestion is not evidence that a creative will perform better. Record the hypothesis and what changed, then assess the result with the available audience and downstream outcome data. Human review still needs to check the product claims, brand fit and the limits of the evidence.
Cutting Through The Hype
Generative AI could be a game changer for the advertising industry. But, like all emerging technologies, there is a lot of hype surrounding it. With the current generation of tools, generative AI may not be ready for mass adoption. There are still challenges that need to be overcome. For example, the OpenAI language model ChatGPT sometimes generates inaccurate information in response to user inputs. In a test, Moz used ChatGPT to generate local SEO guidance. But any business owner following the advice would have actively hurt their local SEO performance. The steps ChatGPT outlined would have likely resulted in the permanent removal of a Google Business Profile due to guideline violations. Instead of providing accurate and up-to-date information, ChatGPT generated disinformation that would lead business owners to waste their time and marketing budget. One of the issues with generative AI is that people need help to spot the difference between human and AI-generated text. A recent study found that untrained humans could only identify text generated by AI at a rate consistent with random chance. With image creation tools like Midjourney, many generated images look more like graphic images than real photos. The model also experiences issues with specific inputs. A Twitter post showing the capabilities of Midjourney went viral earlier this year. At a glance, the AI-generated images appear very realistic. The level of detail is impressive.
But if you take a closer look, you’ll notice some strange features. Almost every person depicted in the images has too many teeth. And if you look closely at the bottom left image, the woman holding the camera appears to have far too many fingers on her left hand. The human hand is complex and can take on many different shapes in images. In their current state, AI tools struggle to make sense of the anatomy.
The Bottom Line
For a production team, the useful question is which task a tool improves after review, editing and rights checks. Compare usable output and total effort with the existing workflow. The examples in this 2023 article explain early possibilities; they do not establish near-zero production cost or a guaranteed advantage for every advertiser.
