Hey ChatGPT: an examination of ChatGPT prompts in marketing

Marketing is one of the areas where large language models (LLMs) such as ChatGPT have found practical applications. This study examines marketing prompts—text inputs created by marketers to guide LLMs in generating desired outputs. By combining insights from the marketing literature and the latest r...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Veröffentlicht in:Journal of marketing analytics 2024-12, Vol.12 (4), p.790-805
Hauptverfasser: Tafesse, Wondwesen, Wood, Bronwyn
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
Beschreibung
Zusammenfassung:Marketing is one of the areas where large language models (LLMs) such as ChatGPT have found practical applications. This study examines marketing prompts—text inputs created by marketers to guide LLMs in generating desired outputs. By combining insights from the marketing literature and the latest research on LLMs, the study develops a conceptual framework around three key features of marketing prompts: prompt domain (the specific marketing actions that the prompts target), prompt appeal (the intended output of the prompts being informative or emotional), and prompt format (the intended output of the prompts being generic or contextual). The study collected hundreds of marketing prompt templates shared on X (formerly Twitter) and analyzed them using a combination of natural language processing techniques and descriptive statistics. The findings indicate that the prompt templates target a wide range of marketing domains—about 16 altogether. Likewise, the findings indicate that most of the marketing prompts are designed to generate informative output (as opposed to emotionally engaging output). Further, the findings indicate that the marketing prompts are designed to generate a balanced mix of generic and contextual output. The study further finds that the use of prompt appeal and prompt format differs by prompt domain.
ISSN:2050-3318
2050-3326
DOI:10.1057/s41270-023-00284-w