• by Laurie Sullivan  @lauriesullivan, December 4, 2024

Jellyfish, a Brandtech Group company, developed a platform to analyze how different large language models (LLMs) perceive a company’s brands, products and services.

The technology -- which the company believes is the first -- identifies whether content is correctly optimized to serve queries and recommendations related to the brand in generative AI (GAI) models such as OpenAI’s ChatGPT, Google’s Gemini and Meta’s Llama.

Knowing the signals that serve up in GAI's based on the LLM used to train the bots can also support ad targeting in ways that become more competitive across multiple types of media. 

“We reverse engineered a marketing approach, picking up signals from an entirely new audience, who’s seen everything you’ve ever made as a marketer and is showing you where the gaps are in your messaging,” said Rebecca Sykes, Brandtech Group Partner and Head of Emerging Tech.

Sykes believes technologies like this one will disrupt marketing and advertising in positive ways. The technology will be prompted to make content when the signal strength is weak for a benefit of the brand through value and data.

The Share of Model Platform connects LLM research with the company’s digital marketing strategy to support the brand’s keyword strategy by adjusting website text and images that adapt to each GAI model.

In early trials, the platform successfully leveraged competitive analysis to understand how AI models perceive the competition as well as identify gaps to take advantage of previously unforeseen opportunities.

"We tracked the top 10 beer brands across the world's most popular models. Their share of LLM recommendations differed from model to model, with major implications given the one billion people interacting with LLM generated content every month," John Dawson, Jellyfish VP of media strategy, told MediaPost in an email. "This is key input to any brand strategy as share of recommendations will determine share of market."

The technology also can analyze the specific language the models use to recommend different beers. One brand might stand out for premium in one model but be perceived as innovative or trendsetting in another. The team can use this specific language to optimize every type of asset - text, image, audio and video - to be more relevant to multimodal models. The more relevant the content, the more greater chances of recommending our clients.

Catherine Lautier, Danone vice president and global head of media, believes the tool provides insights on how its brands compare with competitors, and what drives the interest of consumers online in specific categories.

The tool specifically provides insight on how to develop content on different platforms analyzed by LLM, so when it serves up to consumers in GAIs the content appeals more to consumers.

Results from a YouGov study suggest a need for brands to understand how people use AI tools and their impact and influence on purchase behaviors. Surveying 1,000 U.S. consumers between the ages of 18 and 65, research shows that 66% of 18- to-24-year-olds report that they ask AI models for brand, product and service recommendations.

Fifty-one percent of 25- to-34-year-olds turn to AI models -- a percentage that drops to 42% for those ages 35 to 44, and 31% for those ages 45 to 54.

Half of 18- to-24-year-olds expect AI tools to guide them to the best brands, products or services for them -- a percentage that drops to 47% among 25- to 34-year-olds, and 35% for 35- to-44-year-olds. The decline continues in the 44-and-older category.