# Can AI Generate Useful Messages for Smoking Cessation Campaigns? A Test with Different Emotional Appeals and Source Attribution

## TL;DR

This study tests if AI-generated messages can help quit smoking by comparing emotional appeals and source credibility.

## Contribution

It explores the use of AI-generated messages in smoking cessation with factorial design and source attribution effects.

## Key findings

- AI-generated messages increased risk perceptions but not quit intentions.
- Human sources were perceived as more credible than AI sources.
- No interaction was found between message appeals and source attribution.

## Abstract

This study investigates the viability of using ChatGPT 3.5 to produce smoking cessation messages featuring different emotional appeals. The effect of source attribution to Artificial Intelligence (AI) vs. human experts is also examined. A sample of current smokers (N = 480) was recruited from Prolific and randomly assigned to read one of five ChatGPT-generated messages reflecting a 2 (appeal: threat vs. humor) × 2 (source: AI vs. human experts) factorial design plus an irrelevant message control condition. Exposure to the smoking cessation messages led to a pattern of cognitive and emotional responses largely consistent with expectations based on previous research. Compared to control, the smoking cessation messages generated greater risk perceptions on the featured health effects but did not produce significantly stronger intentions to quit. Human experts as the source produced greater perceived source credibility than AI, but there was no source effect on other outcomes. No interaction between message appeals and source attribution was observed. Implications of the findings for tobacco education campaigns are discussed.