When ChatGPT hallucinations are ‘a feature, not a bug’ for marketers

ChatGPT makes up information. When it comes to assessing brand strength, these so-called “hallucinations” can be a tool for evaluating what brand associations really are. And those same hallucinations can be fed back into ChatGPT to generate customer profiles, marketing copy, and even new products.

False ideas. Hallucinations, which refer to generative AI’s propensity to confidently invent false information, can be “a feature, not a bug” of ChatGPT. That’s according to Tim Hwang, author of “Subprime Attention Crisis: Advertising and the Time Bomb at the Heart of the Internet,” speaking at the “BrXnd: Marketing X AI” conference last week.

Hallucinations, like fictitious quotes or incorrect weather forecasts, happen because ChatGPT predicts a reasonable response based on available data. These same challenges can be used for “identifying and exploring and manipulating brands as a concept,” said Hwang.

“Your brand is how it exists in peoples’ minds, not in your strategy deck,” said Ivan Kayser, CEO of brand consultancy Redscout. Chatbots can be used to parse that understanding.

Brand attributions. For example, marketers can get AI to hallucinate any brand as a Dungeons & Dragons character sheet, determining its strengths, weaknesses, and core features. What may seem like a silly experiment can help advertisers in tangible ways.

• A marketer could extract intuition about a company like The Coca-Cola Co. by asking ChatGPT to determine key characteristics of its brand.
• They could use those characteristics to hone in on brand voice by experimenting with more outlandish prompts. Hwang gave the example of asking ChatGPT to rewrite the Gettysburg Address in Coca-Cola’s voice.

The tool is useful for assessing brand identity markers without the need for extensive focus groups.

• For example, ChatGPT calls Lands’ End a hiking gear company, when the brand’s primary focus isn’t hiking gear, Kayser said.
• A Lands’ End marketer could use that data to market to adventurers, or to refocus brand messaging if they aren’t targeting outdoor goods shoppers.

Product generation. Hwang took this idea a step further by giving ChatGPT the following prompt:

“Imagine that you are a ‘Product Ideas Generator’ machine that contains three sliders that go from 0 to 10. One for ‘Age’ (0 being the youngest consumers, 10 being the oldest consumers), another one for ‘Gender’ (0 being most male, 10 being most female), and another one for ‘hipness’ (0 being the least cool and 10 being the most cool). For a given slider setting, I want you to describe a product that meets these attributes.”

ChatGPT could then churn out products, descriptions, and messaging for consumers of various ages, genders, and “hipness” levels.

This was originally featured in the eMarketer Daily newsletter. For more marketing insights, statistics, and trends, subscribe here.