Using AI Assistants to Predict Purchase Intent

Summary

This study examines user interactions with AI assistants to infer purchase intent. By analyzing the text of user-initiated interactions, researchers build a bipartite network of nouns and verbs and measure the distance of specific words to "golden" purchasing words like "purchase," "buy" or "order." The study uses large language models, specifically Chat-GPT4, to annotate data with a measure of purchase intent and validates this method by comparing the results with cost-per-click (CPC) for keywords in Google Ads. The findings suggest that words used in an exchange with an AI assistant can predict purchase intent without customer tracking across interactions.

These findings have implications for using customized small versus large language models and can potentially inform advertising decisions. The study highlights the importance of understanding consumer behaviors in interactions with AI assistants. It provides a method to predict purchase intent based solely on the textual content of these interactions.