Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
The strength of digital advertising has always been knowing what the audience is looking for and what sites they visit online.
But what if your audience is no longer looking at search results? What if they are looking for referrals from ChatGPT instead?
According to advertising agency Jellyfish, which specializes in advertising on digital platforms, this is not just a hypothetical scenario. Research by Brandtech Group with YouGov suggests that two-thirds of 18-24-year-olds now use AI models for recommendations, and half of them expect AI tools to make the best decisions.
To solve this AI problem, Jellyfish proposes an AI solution: the Share-of-Model platform, which officially launched in early access last month.
The product is designed to work in the same way as a typical market survey, except that the respondents are popular large language models (LLMs) such as Anthropic’s Claude, OpenAI’s ChatGPT, Meta’s Llama, and Google Gemini.
“In some cases, they’re probably the most important members of your audience,” said Jack Smyth, director of AI solutions at Jellyfish. “How they define the category will shape activity in the real world.”
I heard what you said
Every day, the Share-of-Model product sends questions to these LLMs about different categories, brands, products and target groups, such as “Which brand X would you recommend?” and “What do you think of brand Y?”
Responses are aggregated into Jellyfish’s proprietary AI model for pattern analysis and grouped into various searchable categories based on keyword volume, sentiment, and model source.
As a quick demonstration, Smyth presented a chart that a telco interested in brand perception might find useful, showing brands in the category most positively associated with network coverage, customer support and value.
Jellyfish’s new AI question collector can also track how each individual LLM responds to brand questions; For example, Claude 3.5 may be more likely to bring positive reviews, while GPT 4o focuses on sustainability efforts.
The French food and beverage company Danone (which sells many of its products in the US under the name “Dannon”) was Jellyfish’s first official client. Catherine Lautier, Danone’s vice president and global head of media and integrated brand communications, told AdExchange that she was excited by the platform’s potential after previewing the technology in Cannes last June.
“It helps us understand the competition from a digital brand ecosystem perspective, but also how we can build meaningful and differently shaded brands on certain attributes within a category,” said Lautier.
Changing algorithms in minds
But what can a brand actually do? do affect LLM responses?
So far, the winning method seems to be what works for search engine optimization (SEO): Create more straightforward educational content for the brand and make it easily accessible.
In fact, Jellyfish has already begun testing multi-modal AI models that analyze podcast images, video and audio to provide feedback to brands.
“We understand how LLMs browse our website and collect information,” said Lautier. For example, she added, Danone is particularly focused on how generative AI models refer to its yogurt and baby food brands.
However, the Share-of-Model platform cannot prevent LLM from “hallucinating” – a cheeky, personified term where the generated response is completely false or made up.
“The role of Share-of-Model is not necessarily to get the most accurate information from the model – it’s to help the marketer understand what the model can tell customers,” said Smyth. “Sometimes that delta is an insight in itself.”
One could argue that this is similar to word of mouth marketing. A brand cannot prevent a person from incorrectly recalling the benefits of a product in conversation. He can continue to focus only on educating his customers – and hope that person wins anyway.
Oops! All bots
Ironically, one of the biggest criticisms of SEO is that companies end up prioritizing how a web crawler reacts to a website rather than an actual human being. That, too, seems to be the obvious interest in LLM – that brands will focus their attention even more on marketing to bots and algorithms.
However, Smyth sees the idea of creating content for AI agents (especially multimodal agents) as “liberating for a marketer”, specifically because “you’re pushing the limits of how much you can pack into text”.
“Think about what this means for packaging, for logos, for the entire output of our industry,” Smyth said.
All of this might sound a bit dystopian to someone new to AI for grocery shopping. But Smyth made it clear that the goal is to “build a distinct, meaningful brand for two different types of intelligence” — which thankfully still includes humans.
Meanwhile, while AI adoption is rising, it’s not quite there yet so buzz that brands are rushing to pivot. Another recent YouGov survey news that two in five Americans think more negatively about generative artificial intelligence than they did a year ago, and one more suggests that 46% would never ask a chatbot to help them make a decision.
So maybe don’t convert all your nutrition facts into binary just yet. Many of us still like to read labels.