For generations, advertisers have obsessed over how best to win the hearts and minds of consumers. But with people increasingly turning to their large language model of choice for product recommendations, the industry has a new audience to get its head around: artificial intelligence, writes Ali Lyon.
David Ogilvy had four golden rules to help advertisers resonate with fickle consumers: make it simple, make it memorable, make it inviting to look at, and make it fun to read.
The late godfather of modern marketing, and founder of his eponymous agency Ogilvy, had a reputation for bon mots and one-liners that almost surpassed his rightful position as one of the 20th century’s most celebrated ad men.
Another oft-cited observation of his was that “the public is more interested in personalities than in corporations”.
But what if, after over a century of painstakingly marketing to them, those corporations no longer viewed ‘the public’ as their priority audience? And that another, more rational, but more enigmatic audience became more important?
That is the question with which a growing number of advertising and public relations agencies like the one Ogilvy founded in 1948 are grappling as customers increasingly consult their favourite large language model for product recommendations.
David Ogilvy looking out over the River Thames in 1978. (Photo by Keystone/Getty Images)
Advertisers turn to AI platforms
“It is one of the biggest shifts we’re seeing in how brands show up in the world,” Ellie Tuck, creative director and partner at Fleishmanhillard, a PR agency, tells City AM.
“While the fundamentals of trust, quality content [and] authority remain the same, how they are brought to life in an LLM-powered world, is… undergoing massive change.”
People consulting a second opinion before a big purchase or trip is nothing new. For as long as they have been published, readers have thumbed through newspapers, lifestyle magazines or simply sought the advice of friends, on the hunt for opinions and reviews of products and places.
Consequently, third-party validation has played an outsized role in marketers’ decision-making and allocation of resources.
Companies channel vast amounts of intellectual and financial capital into rolling out product review programmes with media outlets, or to ensure they perform well in the kind of test environments made popular by Which? and the Good Housekeeping Institute.
But those well-worn consumer habits are rapidly changing. A recent McKinsey poll found that over a quarter of consumers now use generative AI tools to help them choose products. And so, where consumer-facing firms once had to worry about persuading newspapers or ‘influencers’ to try out their latest release, they now have to ensure they appear in answers regurgitated by the likes of Chat GPT and Perplexity. It’s not just direct questions about big purchasing decisions, either.
It is one of the biggest shifts we’re seeing in how brands show up in the world
“People are interacting with LLMs in different ways than you might expect,” says Natasha Wallace, chief strategy officer at digital marketing firm Jellyfish. “They’re very conversational, and will ask it things like, ‘It’s festival season, what should I be thinking about for my summer outfit, and what should be on my Spotify playlist?’”
General Engine Optimisation (GEO)
Having good visibility in those answers has been given a name – General Engine Optimisation (GEO) – and advertisers are increasingly turning to specialist firms like Jellyfish to ensure they don’t become invisible as customers embrace this new frontier.
Recently, the Amsterdam-based agency developed its own ‘Share of Model’ software, which allows it to dig into how different LLMs regard their clients and determine their respective likelihood of being recommended by them.
“What we do is configure it to that client – adjusting it to their audience segment and category personas – and then fire these models thousands and thousands of prompts in, to extract the full view of how it understands a brand,” says Wallace.
To make matters more complicated, the answers that Wallace’s Share of Model product – or just regular consumers – receive to the same prompt can vary enormously by LLs.
To prove the point, City AM gave Gemini, Chat GPT and Perplexity exactly the same question – ‘Recommend me the five best washing machines for under £300’ – and just one product made the top three in every AI model.
“They’ve been trained on totally different kinds of data sets,” says Alex Dalman, head of AI at UK advertising juggernaut VCCP. “It’s very hard to keep track of which data is coming in at what point… it can be a nightmare for marketers.”
Sam Altman’s OpenAI has produced a large language model, ChatGPT, that has a proclivity for citing information from Reuters, Wikipedia and the Financial Times (
A recent paper from media analysis firm Muckrack sought to trace the favoured sources of each major LLM, giving them over 1m prompts to map their sources. Chat GPT – the paper discovered – loves Reuters and Wikipedia for general questions, but Claude – the LLM operated by Anthropic – preferred CNBC and Good Housekeeping.
Sources varied by sector, too. For prompts relating to retail and pharma, Claude tended to cite information from official sources like the US Food and Drug Administration and the National Library of Medicine. OpenAI’s Chat GPT, on the other hand, looked to Wikipedia and the Associated Press.
“It’s most obvious with [Google-owned] Gemini and [Meta’s] Llama, which are biased to their own ecosystem,” says Jellyfish’s Wallace. “Gemini will be looking at Youtube videos. Llama lends itself to Instagram, Facebook and its own ecosystem as well.”
A fledgling discipline
But despite the speed at which consumer behaviour is changing, Dalman believes GEO remains somewhat of a “black box” for marketers.
“I think the rubber will hit the road in the next three months, as we get into the biggest sales period of the year,” she says. “We’ve got Black Friday, Christmas, and everyone will be going, ‘What are the best headphones for my son for Christmas?’ or what are the latest toys that I need to buy?’”
By that stage, tech firms may have opened up their models to paid recommendations, an “inevitability” according to Dalman and Wallace.
But until they’re able to buy their way to the top, global brands are turning to agencies like Jellyfish or resorting to other, slightly off-the-wall solutions.
“One piece of advice we’re seeing a lot is for companies to make their web pages more machine readable, as opposed to human readable,” says Peter Gasston, innovation lead at VCCP. “When we go to a website it’s formatted for us. For something to be machine readable, you need to strip away all of that visual data, and reduce it down to – basically – a spreadsheet of data.”
It needs all the information in one place, clearly delineated, he adds, and doesn’t have to “look good or be readable for humans”.
You wouldn’t have caught David Ogilvy recommending that.