Research in advertising has shown that share-of-voice (buzz), or more specifically excess share-of-voice, correlates to market share growth. A recent HBR paper showed that the number of positive/negative associations that a brand evokes is strongly correlated with market share and future spending. It also showed that the more inter-connected these associations are the stronger the correlation. Such brand associations can be mixed and connected to consumer’s own beliefs and values. The brand association network then evolves into a brand narrative. A brand narrative includes functional associations (what does the brand do for me), emotional associations (how does it make me feel), to what does it mean in my life. Narratives merge specific brand attributes with themes that are relevant for the consumer such as values (love, ambition, security, etc.) and cultural elements (survival, winning, identity, etc.). It gives consumers a deeper emotional connection with the brand beyond the functional utility. Brand narratives are influenced by share-of-voice (e.g. ad spending), what a consumer’s experience is with the brand, and how the brand is covered in earned media (e.g. PR, podcasts, blogs, etc.).
When we look at the effectiveness of earned media, both buzz and narrative richness could drive public interest. In earned media we define share-of-voice (buzz) as simply the frequency with which brands are mentioned. We can also apply the brand narrative concept to assess the quality with which brands are covered in earned media. Specifically, here, we look at long-form content such as podcasts which are well suited to measure brand narrative richness. When we use podcast data, we are not measuring consumer narratives directly—the podcasts are typically conversations between influencers and domain experts. Instead, we measure earned media brand conversation richness: the combination of the number of different themes used in discussing a brand and the interconnectedness of those themes. A brand heavily discussed at high volume but always around the same theme has low conversation richness. Lower volume with more themes and more interconnection yields higher richness. In this paper we discuss how buzz in earned media and narrative richness drive public interest for brands.
How Conversation Richness and Buzz Drives Public Interest
We analyzed 1.36 million podcast transcript segments collected by the AllEars.ai platform, spanning September 2025 through March 2026. We focused on five brands with sufficient weekly volume for robust statistical modeling: OpenAI, Google, Anthropic, Meta, and Microsoft.
There are several dimensions we wanted to capture from this data.
· First, we want to track brand buzz over time. This is simply the number of times a brand is mentioned across the podcast over the course of a week.
· Second, we used text analysis and emotional classification tools, to analyze the podcasts transcripts and to distill four conversation themes and three emotions: Curiosity & Intrigue, Anxiety & Safety, Distrust & Cynicism, Creative & Media (video, audio, image generation), Consumer Life (learning, productivity, daily use), Developer & Technical (coding, APIs, infrastructure), and Product Launches.
· Third, we calculate overall brand conversation richness: this metric is a function of distinct sub-themes, and how connected these themes are in the conversations.
Public interest is measured independently using two external sources: weekly Wikipedia pageviews (Wikimedia API) and Google Trends search index. Neither outcome is controlled by any brand—they reflect genuine public curiosity. Using a Bayesian hierarchical model with within-brand demeaning, we isolated which conversation themes predict week-over-week changes in public interest after controlling for product launches, seasonal effects, and persistent brand-level differences.
Conversation richness drives public interest
When we model public interest (Wikipedia & Google trends) as a function of buzz and conversation richness, we find that both variables have a significant effect on both public interest variables. Interestingly, brand conversation richness has a stronger effect on Google trends search than buzz (about 2 times stronger), whereas buzz has stronger effect on Wikipedia pageviews (about 5 times stronger). So, how much the brand is talked about and how rich that conversation is affects public interest.
There are several reasons why buzz alone does not drive interest. One, the richer the conversation around a brand, the more hooks are out there that can cause someone’s interest to spike. Second, when we analyze what themes and emotions are associated with which brands, we find that not all themes are discussed at the same depth across various brands and the emotions are not evenly spread across the brands. This is relevant because the themes and emotions themselves have an impact on public interest. Some themes have a bigger impact on public interest than others. Figure 1 shows the seven themes and their impact on public interest, calculated across brands using a pooled model (meaning across brands).
Bayesian hierarchical model, N = 915 brand-weeks, 5 brands. Binary within-brand top-quartile indicators. *** p < 0.001, ** p < 0.01, * p < 0.05.
Several findings stand out. First, curiosity-framing is the single most powerful lever. Creative media demonstrations and anxiety & safety discourse also drive significant interest. On the negative side, developer & technical discourse and distrust & cynicism strongly suppress public curiosity. Product launches show no significantly pooled effect on interest, but this masks dramatic brand-level variation.
The themes vary considerably across brands and hence contribute differently to a brand’s overall conversation richness. Curiosity is almost half the conversation richness for overall Anthropic conversation richness but is non-significant for OpenAI, where creative media and consumer life are the primary drivers. Product launches drive richness for Anthropic (whose launches open new product categories) but affect interest for OpenAI negatively. These differences underscore that effective narrative strategy must be brand specific.
The results we presented were based on a sophisticated Hierarchical mixed effects model, where we controlled for buzz, product launches, seasonal effects, and persistent brand-level differences. The conversations around a brand can matter more than the sheer volume with which a brand is discussed, and as much as what the brand ships. We also found that themes that are associated with a brand can further explain why conversation richness can have a strong effect.
Industry Value
Brand narratives go beyond communication. Communicated and experienced brand attributes and associations are woven into consumers’ minds more deeply, merging with cultural and emotional themes that matter to the consumer. We show here that we can apply the concept of conversation richness to earned media and the public discourse domain. This sets up a framework for assessing how brand conversation richness—measured through deep analysis of long-form unstructured data—predicts relevant outcomes such as public interest in a brand.
Marketers and the PR department can use this to:
· Keep track of conversation themes and richness, not just volume—buzz alone is not enough.
· Diagnose which themes drive interest for their brand or decrease it.
· Connect themes intentionally: e.g., don’t just talk “technical”—link it to “consumer life” or “creativity.”
· Plan narrative arcs around launches to sustain interest without creating fatigue.
· Monitor risk: spikes in distrust or cynicism may predict attention—but not the kind you want.
We have shown that conversation richness can be distilled from long-form data. In traditional surveys such data is usually not achievable, but with the emergence of AI-driven interviewing, we can pose open-ended questions with intelligent follow-up. This produces richer unstructured brand data from which we can calculate brand narrative richness in consumers’ minds. Once the system of analysis is developed, the model can be updated monthly as effects may change over time. A technical note with modelling details is available from the authors.
