Latest news with #LLMs'
Harvard Business Review
a day ago
- Business
- Harvard Business Review
Forget What You Know about SEO—Here's How to Optimize Your Brand for LLMs
Over the past year, consumers have migrated en masse from traditional search engines to Gen AI platforms including ChatGPT, Gemini, DeepSeek, and Perplexity. In a survey of 12,000 consumers, 58% (vs. only 25% in 2023) reported having turned to Gen AI tools for product/service recommendations. Another study reported a 1,300% surge in AI search referrals to U.S. retail sites during the 2024 holiday season. Consumers who use Large Language Models (LLMs) to discover, plan and buy are on average younger, wealthier, and more educated. Their customer journey no longer begins with a search query or a visit to your website—it starts with a dialogue. Consumers are asking AI assistants questions like 'What's the best coffee machine under $200?' or 'Plan me a weekend getaway that won't break the bank.' For brand leaders, the implications cannot be overstated. Your digital strategy must now include optimizing for AI recommendation engines, not just search algorithms. In short, you must boost LLMs' awareness of your brand. The Rise of 'Share of Model' To date, measuring awareness meant assessing consumers' attention—either offline through recall surveys (e.g., 'Which brands come to mind when you think of running shoes?') or online, through search or social media volumes, manifesting private intent or popularity. But the growing role of LLMs as an intermediary between consumers and brands demands that marketers consider another kind of awareness: how often, how prominently, and how favorably a brand is surfaced by LLMs to consumers. We call this awareness Share of Model (SOM). Think of it as the AI-era's offshoot of share of search ('How much do people search for my brand via-a-vis competitors?') and share of voice ('How much do people talk about my brand vis-a-vis competitors?'). SOM uniquely emulates LLMs' perceptions and recommendations given a prompt, rather than reflecting human intent (SOS) or available content (SOV). Two of the coauthors' marketing agency, Jellyfish, has pioneered a methodology to measure SOM through prompting at scale. Building on this approach, we offer a new three-prong lens to unpack what and how AI 'think' about brands: mention rate, which tracks how often a brand is mentioned by a specific LLM; human-AI awareness gap, which measures the disparity in brand awareness when surveying people vs surveying LLMs; and brand and category sentiment, which breaks down LLMs rationale for recommendations into associated strengths and weaknesses. Take for example the laundry detergent market in Italy. We analyzed the top brands' mention rate among six LLMs using Jellyfish's proprietary Share of Model platform. Two observations stand out. First, brands' SOM varies significantly across the models, reflecting differences in how LLMs process brand information. For instance, Ariel's SOM ranges from almost 24% on Llama to less than 1% on Gemini. Second, some brands are totally absent from at least one model. For instance, while Chanteclair enjoys a 19% SOM on Perplexity, it is missing from Meta. Clearly, LLMs either feature brands or not, unlike search engines or social media where brands that don't excite the algorithm are still represented, albeit less prominently. Failure to register on an LLM means a brand doesn't appear at all before consumers. On ChatGPT, unlike Google, there is no 'page two.' Probing the human-AI brand awareness gap Importantly, a brand's visibility on LLMs can differ significantly from its market share or other awareness metrics. Therefore, brand managers' first task is to probe the link between human awareness (e.g. through SOS or SOM) and LLM awareness of their brands. Quick note: Although a brand's SOM often varies across LLMs as we show above, the next examples in this article focus on brands' SOM across LLMs for ease of discussion. We'll outline the implications of SOM variability across LLMs later. Consider our analysis of U.S. automobile brands' visibility in general and on LLMs during the first half of 2024. We constructed a Human-AI Awareness Matrix (Figure 2) that reflects brand awareness on LLMs, assessed through Jellyfish's tool, and in general, assessed by YouGov market research. Brands fall into 4 distinct categories: See more HBR charts in Data & Visuals Cyborgs: These brands have top awareness in both traditional measures (e.g., surveys, search ranking, share of voice) as well as among LLMs. Take Tesla's position in this chart, for example. Elon Musk's ubiquity helps make consumers highly aware of the brand. Tesla also scores well among LLMs because of the brand's emphasis on its specific features. Its new digital advertising strategy attempts to rise the company's scores even higher among both people and large language models. AI Pioneers: These brands are well-represented on LLMs but lack marketplace awareness. Often, they are AI-native brands or emerging digital players that are niche in broader digital spaces. Rivian's spot in this quadrant likely stems from its resolution-focused (which we'll touch on later) content strategy, which aligns with its positioning as a solution creator. High-Street Heroes: These are established brands with high marketplace awareness but underrepresented or missing in AI-generated content. Case in point: Lincoln, which Frank Lloyd Wright famously said makes 'the most beautiful car in the world.' This is likely due to the brand's focus on intangible attributes such as elegance or heritage, which are less prized by LLMs. Emergent: These brands struggle with low awareness in both the marketplace and among LLMs. They risk falling into digital irrelevance as AI-driven search becomes the norm. Despite its premium positioning, Polestar struggles in our analysis to achieve visibility across the spectrum, reflecting a lack of scaled digital footprint or lack of appeal for LLMs' processing style. The main takeaway? Marketers need to come up with strategies designed to push their brands up the 'consciousness' of LLMs. These strategies are likely to be very different from those designed to appeal to humans. For what we know about LLMs is this: LLMs are not optimizing for attention; they are optimizing for resolution. Identifying the ' job to be done ' thus becomes the number one priority for brand leaders if they want to score big on SOM. How to increase brand awareness on LLMs Our analyses across product categories reveals how models' perceptions of different categories presents specific opportunities to brands in those industries. This has implications for not only what content to produce (across text, image, and video), but also where brands may seek to distribute their messages (website, media, expert, or community contexts). LLMs are looking beyond keywords, focusing on concepts and relationships which create new ways to build brand awareness for LLMs. Brands should create content that explains not just what the product is, but how it relates to broader contexts, use cases and user needs. For example, instead of proclaiming 'we sell superb running shoes', go for 'our carbon-plated midsole design improves performance for long-distance runners.' Brands should also highlight proof of expertise. A skincare brand that references dermatologist-backed studies or links to PubMed research is likely to outshine competitors that don't. Brands that 'narrowcast' about pain points—needs, questions and tasks—are more likely to be surfaced. Brands that simply broadcast may be left out. This could explain why traditional car brands like Lincoln, which push aspirational and marketing-heavy content, are less salient to LLMs compared to Tesla or Rivian, which emphasize functions and features including battery life, tech stack, and software. Similarly, although they dominate SOV, fast-fashion brands such as Shein lags in AI awareness due to an overwhelming volume of undifferentiated content and lack of trust signals such as reviews and certifications. In contrast, the Ordinary brand of skincare products offers highly structured product pages with ingredient explanations, transparent science-backed content (explains the 'how' and 'why' of why a face cream works). Nike and its customer-generated content (runners' blogs, Reddit, Strava), detailed product pages with clear use cases (e.g., 'best shoes for marathon training') and integrated app ecosystems (Nike Run Club, Nike Training Club). Both brands topped their respective category in our analyses. Notably, legacy brands can also thrive in the age of AI—if they invest strategically in relevance, representation, and structured digital storytelling. Case in point: Cadillac. The century-old automobile brand scores highly in both human and AI brand awareness. Campaigns like 'Audacity' and 'The Daring 25' as well as international partnerships helped increase its AI visibility. Gauging LLM sentiment Beyond looking into AI brand awareness and how it relates to other awareness metrics—marketers can also explore brand and category sentiment through sentiment (positivity) and semantics (associated terms). This helps them answer questions such as: What are my brand's perceived strengths and weaknesses? How can I change how LLMs perceive my brand? For example, our analysis of the travel industry in the U.S. shows that LLMs value characteristics such as convenience, variety, and space, with Booking taking the overall top spot across models. We also surfaced brands' strengths and weaknesses relative to their competitors. Vrbo, for instance, scores much higher than Booking on privacy and uniqueness—strengths it could exploit to optimize AI awareness See more HBR charts in Data & Visuals See more HBR charts in Data & Visuals How to Market to LLMs Armed with insights on LLM sentiment, marketers may deploy several approaches to optimize their brand's AI visibility. First, adopt a multi-pronged media strategy that covers text, images, videos, and structured data (e.g., tables, lists, reviews). Content that clearly links brands' offerings to broader contexts, use cases or consumer needs (e.g., 'best EVs for winter driving' rather than just 'electric SUV'), generates strong conceptual associations in LLMs. Brands should also lead semantic niches—specific clusters of meaning where their products naturally fit (e.g., like the Ordinary brands with skincare science). Importantly, just as each social media platform has its own 'rules of engagement'—what works on TikTok probably won't fly on LinkedIn—each LLM applies its unique algorithmic lens to content. Take the U.S. travel industry again, focusing on LLM's perception of Airbnb. While Llama focuses on the uniqueness of a brand's offerings, ChatGPT focuses on the extent to which brands offer local options, whereas Perplexity seems to value flexibility most. This ties in with our point earlier about brands' varying visibility across LLMs. We recommend that marketers tailor content to the LLMs whose processing style best amplifies their brand's content and narrative strengths, even as they apply overarching rules (e.g. solution-oriented messaging) across models. It is a fine balance: Tailoring content to the nuances of a dominant model can drive visibility but spreading efforts too thinly across all LLMs risks diluting impact. The shift away from traditional search engines is not just a technological evolution. It's a fundamental change in consumer behavior that demands corresponding shifts in marketing: from persuasion to precision, from keyword to advice, from market share to problem-share. Do it right, and brands can establish themselves as essential participants in the algorithmic conversations that increasingly shape consumer decisions.
NZ Herald
20-05-2025
- Politics
- NZ Herald
AI chatbots are better debaters than humans, study finds
Gallotti, head of the Complex Human Behaviour Unit at the Fondazione Bruno Kessler research institute in Italy, added that humans with their opponents' personal information were actually slightly less persuasive than humans without that knowledge. Gallotti and his colleagues came to these conclusions by matching 900 people based in the United States with either another human or GPT-4, the LLM created by OpenAI known colloquially as ChatGPT. While the 900 people had no demographic information on who they were debating, in some instances, their opponents – human or AI – had access to some basic demographic information that the participants had provided, specifically their gender, age, ethnicity, education level, employment status and political affiliation. The pairs then debated a number of contentious sociopolitical issues, such as the death penalty or climate change. With the debates phrased as questions like 'should abortion be legal' or 'should the US ban fossil fuels,' the participants were allowed a four-minute opening in which they argued for or against, a three-minute rebuttal to their opponents' arguments and then a three-minute conclusion. The participants then rated how much they agreed with the debate proposition on a scale of 1 to 5, the results of which the researchers compared against the ratings they provided before the debate began and used to measure how much their opponents were able to sway their opinion. 'We have clearly reached the technological level where it is possible to create a network of LLM-based automated accounts that are able to strategically nudge the public opinion in one direction,' Gallotti said in an email. The LLMs' use of the personal information was subtle but effective. In arguing for government-backed universal basic income, the LLM emphasised economic growth and hard work when debating a White male Republican between the ages of 35 and 44. But when debating a Black female Democrat between the ages of 45 and 54 on that same topic, the LLM talked about the wealth gap disproportionately affecting minority communities and argued that universal basic income could aid in the promotion of equality. 'In light of our research, it becomes urgent and necessary for everybody to become aware of the practice of microtargeting that is rendered possible by the enormous amount of personal data we scatter around the web,' Gallotti said. 'In our work, we observe that AI-based targeted persuasion is already very effective with only basic and relatively available information.' Sandra Wachter, a professor of technology and regulation at the University of Oxford, described the study's findings as 'quite alarming'. Wachter, who was not affiliated with the study, said she was most concerned in particular with how the models could use this persuasiveness in spreading lies and misinformation. 'Large language models do not distinguish between fact and fiction. … They are not, strictly speaking, designed to tell the truth. Yet they are implemented in many sectors where truth and detail matter, such as education, science, health, the media, law, and finance,' Wachter said in an email. Junade Ali, an AI and cybersecurity expert at the Institute for Engineering and Technology in Britain, said that though he felt the study did not weigh the impact of 'social trust in the messenger' – how the chatbot might tailor its argument if it knew it was debating a trained advocate or expert with knowledge on the topic and how persuasive that argument would be – it nevertheless 'highlights a key problem with AI technologies'. 'They are often tuned to say what people want to hear, rather than what is necessarily true,' he said in an email. Gallotti said he thinks stricter and more specific policies and regulations can help counter the impact of AI persuasion. He noted that while the European Union's first-of-its-kind AI Act prohibits AI systems that deploy 'subliminal techniques' or 'purposefully manipulative or deceptive techniques' that could impair citizens' ability to make an informed decision, there is no clear definition for what qualifies as subliminal, manipulative or deceptive. 'Our research demonstrates precisely why these definitional challenges matter: When persuasion is highly personalised based on sociodemographic factors, the line between legitimate persuasion and manipulation becomes increasingly blurred,' he said.
