Privacy regulations and platform policies reduce access to behavioral tracking data that many firms rely on to profile, target, and acquire customers across digital channels. As these trackingbased signals become increasingly constrained, firms must rely on privacy-compatible inputs to operate their customer-acquisition processes. We study how search engine advertising (SEA), which can be run solely on search queries, can be used not only for intent-based performance optimization but also for audience-specific targeting under privacy constraints. Our focus is on search queries—the text that consumers voluntarily enter when seeking information—as probabilistic indicators of whether the searcher belongs to a strategically defined target audience. Using a custom-built search environment as an operational testbed, we collect over 4,000 queries from 2,778 consumers across three privacy-sensitive contexts (weight loss, online dating, and investment) and link each query to detailed consumer characteristics. We extract semantic topics and linguistic features from queries and estimate Bayesian logistic regressions that map these signals to target audience membership, thereby identifying “indicative queries” that define a query-based targeting policy. On held-out data, this policy yields mean hit-rate uplifts of 7-18% over random selection; 72-81% of characteristics show positive uplift, and upper-quartile uplifts range from 15-31%. To assess the cost-quality trade-off in a real-world setting, we implement the policy in a seven-month field study with a retail bank seeking experienced traders. An audience-specific search campaign based solely on indicative queries acquired 21% more new customers and generated 98% higher trading volume per active account than the bank’s established performance-based campaign, at roughly 50% higher cost-per-click—a favorable tradeoff for the bank. Together, our findings show how firms can redesign customer-acquisition operations to maintain audience-targeting capabilities under stringent privacy constraints by repurposing search queries—voluntarily provided, non-tracking data—as a privacy-compliant source of audience signals.

Many brands leverage authenticity to connect with consumers by embodying values that resonate with them, fostering a sense of shared purpose. Yet, despite its advantages, such a values-based positioning might entail substantial risks for demand. We argue that when consumer-brand relationships are premised on a brand's consistent commitment to its values, demand-side authenticity beliefs carry a binding liability. If a brand action disrupts these beliefs, it may cause non-compensatory consumer reactance. To document a possible impact of such an authenticity violation on consumer demand, we leverage a recent brand acquisition in a mainstream consumer goods category (spices). Using difference-indifference and synthetic control methods, we estimate that the acquisition-widely perceived as an authenticity violation-led to a 22.74% decline in brick-and-mortar sales and a 32.02% demand decline online. Content analysis of social media records confirms that this decline was indeed driven by a loss of consumers' beliefs in brand authenticity. We further suggest that the violation's impact was transmitted through the brand's reliance on influencers: many influencers publicly withdrew their support following the acquisition. Our findings demonstrate how authenticity, while a valuable marketing asset, can also expose brands to severe demand risks when violated.

Foundational marketing knowledge, from theories of consumer behavior to segmentation frameworks and pricing models, supports effective marketing decisions, yet managers often struggle to access and apply it in practice. This paper evaluates whether large language models (LLMs) can function as scalable marketing knowledge systems that make such knowledge broadly accessible. Specifically, we assess what LLMs “know” about marketing and how effectively they can reason with and apply that knowledge. We construct a comprehensive dataset of over 30,000 questions drawn from instructor materials of 25 widely used marketing textbooks and assess LLM performance across three dimensions: domain knowledge, reasoning capabilities, and user interaction. Across models and providers, current LLMs achieve high accuracy (84%- 94%), demonstrate near-complete coverage of marketing topics, and show strong performance on recall and conceptual understanding. Performance declines on higher-order reasoning tasks, though newer reasoning models close these gaps substantially. Experimental manipulations of these questions suggest that performance reflects conceptual understanding rather than simple memorization. In a human benchmarking study, LLMs substantially outperform respondents across all training levels, while showing complementary strengths relative to more advanced human reasoning. These findings indicate that LLMs have crossed an important capability threshold: they now provide reliable, on-demand access to much of the discipline’s foundational knowledge, with important implications for how marketing knowledge is disseminated, taught, and applied.