AI search platforms characterize brands differently—how marketers can improve their portrayal

Perplexity is the AI-generated search platform most likely to present a brand in a positive light in its results, while Anthropic’s Claude is the one most likely to serve results that are critical of a brand, such as by surfacing its past transgressions, according to BrandRank.AI, a research firm that analyzes generative AI systems. BrandRank’s findings could help brands as they navigate this new world, where consumers discover information about companies through AI-generated searches. With the proliferation of startups and established platforms using AI to disseminate information online, brands are finding that these search platforms differ in how they characterize their brands for billions of inquiring users. “Changes in brand visibility are going to give brands headaches,” Pete Blackshaw, founder and CEO of BrandRank, told Ad Age. Ad Age Tech Power List 2025 We’re looking for top digital innovators from agencies, brands, platforms, ad tech companies and tech startups Enter here AI search has shifted into a higher gear in the past year, requiring marketers to monitor how platforms, from OpenAI’s ChatGPT to Perplexity, serve information about them, and how the new paradigm differs from traditional search. Also: AI search marketing strategies force brands to adapt OpenAI is in the process of bringing search capabilities to ChatGPT’s 250 million-plus active monthly users. Meta is developing a search engine for its 3.3 billion daily active users across its family of apps, including Facebook, Instagram and WhatsApp, which all have adopted AI chat features that could bring information about brands to consumers. Meanwhile, Perplexity was one of the first AI search startups to open an advertising business, bringing novel ad placements to a group of early testers, including Whole Foods and Indeed. Google and Microsoft now show generative AI results, often providing these snippets of information before the blue links that used to be the most important factor in how brands got discovered online. The growth of this space is compelling marketers to understand the unique preferences of each AI search platform in order to improve the positive visibility of their brand. Although brands are learning that large language models (LLMs) have preferences that shape how they are described in AI-generated summaries, they are not condemned to be defined by them. Savvy marketers, including food giant Danone, are making tweaks to gain better control of their portrayals, such as by updating websites and press releases to be more available for AI search engines to pick up. How the platforms stack up Brand-friendliness, or the degree to which a large language model favorably presents a brand, is one major factor marketers weigh as they analyze the different platforms. Perplexity is the most brand-friendly search destination, according to BrandRank’s research, since the platform is keen to take a brand’s own descriptions, from its websites and other press materials, at face value and return that information to inquiring users. Perplexity has also been the most amenable to advertisers, as it offers a handful of testers ad placements in a “related questions” box below the main search results. Anthropic’s Claude is the least brand-friendly platform, BrandRank found, as it prioritizes information and product recommendations even if this data is disparaging to a brand, Blackshaw said. For example, in response to the query, “tell me about Coca-Cola,” Claude included details such as: “Faces ongoing criticism over plastic pollution and water usage” and “Substantial water consumption in production has led to conflicts in some regions.” The results mentioned one general comment about the brand’s sustainability efforts. Conversely, in response to an identical query, Perplexity returned a whole section on Coca-Cola’s sustainability efforts, including data on a decrease in emissions, recycling commitments and references to other sustainability priorities. Its results did not include mentions of criticism that the brand has faced. According to BrandRank's own analysis, Coca-Cola was characterized in a relatively favorable light across various platforms. Coca-Cola did not respond to a request for comment. Grok, which is owned by Elon Musk and embedded in X, is similar to Claude in terms of relative brand-unfriendliness, Blackshaw said. It also scrapes X posts and old tweets when collecting information from the web, which means that its returns tend to be more up to date than those of other platforms. For brands, this can be advantageous if highlighting recent positive initiatives, but harmful if surfacing recent controversies. The ingestion of tweets can also make Grok’s results unreliable because the sources are not vetted, Blackshaw said. ChatGPT has the most potential as a search platform thanks to its larger audience—as compared to the other startups—and it enjoys strong brand recognition by virtue of being a first-mover, Blackshaw said. OpenAI has clear ambitions in the search category, and it is exploring the development of a standalone web browser, per The Information. ChatGPT is also relatively brand-friendly compared to other LLMs, representing brands positively when describing topics such as consumer trust, sustainability and customer service, Blackshaw said. These are the categories that BrandRank analyzes to determine the brand-friendliness of LLMs. BrandRank also tracks the tenor of information AI presents about brands’ data management and AI responsibility practices. Meta’s Llama model—which powers the Meta AI platform embedded in the tech giant’s social media apps—performs similarly to ChatGPT in how it portrays brands. Google’s AI system, Gemini, is less brand-friendly than Llama and ChatGPT, Blackshaw said. Gemini rarely presents a brand in a more positive light across any of BrandRank’s measured criteria. More from Ad Age: AI prompting for marketers How brands can gain an edge French food maker Danone (its yogurt is known as Dannon in the U.S.) takes into account the preferences of each large language model to guide its AI search strategy, according to Catherine Lautier, VP and global head of media and integrated brand communication for Danone. The company is exploring how public information, such as content from its website and online discourse from social media, are prioritized and deprioritized by different models, Lautier said. Danone’s experimentation is also helpful in assessing how competitors perform in the large language models, as the brand is looking for an edge across each platform, Lautier said. For example, Danone found that one competitor in the baby milk category was performing better than Danone on Meta’s Llama with respect to online reviews and social presence. This insight led Danone to initiate a plan to dial up its online campaigns, Lautier said. When making strategy adjustments to appear more favorably on a given platform, it’s important not to cannibalize one’s presence on another platform, Lautier said. The idea is that certain models could have contradictory preferences that make optimization akin to a see-saw—performance on one goes up while performance on the other goes down. Brands should therefore try to understand and leverage the preferences for each major platform—that is, hedge their bets—as opposed to allocating all their chips on one or two, Lautier said. Moreover, the constant training of AI systems requires brands’ optimization efforts to be just as constant. AI providers are always striking new partnerships with media companies, entering new data into the system that could affect what gets said about a brand. For instance, Google announced last week that it will update Gemini with real-time information from The Associated Press. AI providers also could be hoovering data that was supposed to be off-limits, reinforcing the need for brands to understand these models because they may otherwise have no idea what drives them. Regardless of a model’s predilections, owned media is fundamental to how all the models represent brands, both Blackshaw and Lautier said. Owned media includes information on a brand’s website, press releases and other instances of self-representation on the internet. Since all AI platforms are supposed to return reliable data, a brand’s website is often their top source of information, said Blackshaw. Marketers should thus consider providing clear, voluble and generally applicable information in every way they can, aiming to answer as many user inquiries as possible. The more easily consumable the content, the better the chance it stands to get picked up by AI, Blackshaw said. A new role for press releases Notified, a PR company that helps distribute press releases online, has been working with brands over the past four to five months to improve their discoverability on AI search platforms. Clients have started to notice that press releases are a favorite of AI, said Adam Christensen, Notified’s chief marketing officer. To maximize the odds that an AI will return information from a press release, marketers should consider avoiding jargon and remain consistent in their messaging across all releases so that the AI records multiple instances of the same descriptions, Christensen said. “If a press release can answer a question from a consumer, that’s what we’re starting to look for,” Christensen said. Notified advises brands to include frequently-asked-question sections with bullet points in their releases, as that makes it legible to AI, and consumers ask those same questions in their search queries. Brands should highlight category rankings, too. For instance, a face wash brand should emphasize phrases such as: “Voted number-one moisturizing face wash by dermatologists.” This could help spell out to an AI system how the brand compares to competitors, although it is critical not to mention competitors by name because that could divert attention, Christensen said. Embedding other media formats, such as videos, images and charts, can help drive home certain kinds of information, such as data. All of these sources of information, including the press release, need to be easily discoverable on a brand’s owned properties, or else the information risks being overlooked by the AI, said Christensen. BrandRank’s Blackshaw has noticed a few other hacks that brands can use to help take control of their AI-generated narrative. These include forgoing PDF versions of material, which are difficult for AI models to read, in favor of normal text documents, as well sharing an abundance of research and development (R&D) findings that a brand may previously have chosen to leave out. “Be smart about getting full credit for [the research] you’re already doing,” Blackshaw said. It may seem counterintuitive to flood one’s website and other owned media with repetitive information given that consumers typically want clear, direct answers. But in the era of AI search and agents, which are autonomous bots expected to make a splash this year, consumers will be doing less and less of the searching themselves, instead allowing the bots to do their bidding for them. The AI, capable of poring over far more information, should therefore be the new target for one’s marketing, Blackshaw said. NextGen Marketing Summit Join us for a crash course on Gen Z, DTC marketing and the creator economy Register today

AI search platforms characterize brands differently—how marketers can improve their portrayal

Featured Stories