Experimental Design Details
There are three experimental conditions in our experiment:
Original introduction. We collect the original campaign introductions from a large crowdfunding platform and use them as our experimental materials.
AI-augmented introduction. Leveraging the insights from previous literature (e.g., Karlan et al. 2011) and empirical analysis, we deploy generative AI models to rewrite the campaign introductions, hopefully improving the likelihood of generative AI extended description.
AI-extended introduction. Since introductions augmented by artificial intelligence (AI) commonly exhibit an increase in expressive elements compared to their original counterparts. Consequently, the length of these AI-augmented introductions tends to be significantly greater, rendering them not directly comparable to the original versions in terms of length. To mitigate potential confounding factors and ensure a more controlled comparison, we introduce a subset of introductions termed 'AI-extended introductions.' In this approach, we employ generative AI to paraphrase the original introductions, thereby increasing their length while deliberately avoiding the introduction of new content. This method allows for an examination of text length augmentation effects, independent of content variation.
For each campaign, we will generate three pairs of comparisons, namely, the original introduction versus its AI-augmented version, the original version versus its AI-extended version, and the AI-augmented version versus the AI-extended version.
In this study, each participant will be presented with two random pairs of campaign introductions. Specifically, for each participant, we first randomly draw one campaign that was originally funded and another campaign that originally had no funding secured. For each campaign, we will randomly draw one pair (from the three types of combinations). The two campaigns will be presented in a random order, and within each pair of comparisons, the two variations of the same original campaign will be shown in a random order side by side. After each pair, we will collect participants' preferences and their predictions of other people's general preferences towards which campaign will be more likely to receive financial support.
Experimental procedures
To summarize, if a participant opts to take the survey and passes the attention check, they will conduct a one-time online survey with the following steps:
Step 1: Attention check
Step 2: The first pair of campaign descriptions is presented side by side.
Step 3: Participant will indicate their potential donation preferences and their estimation of to which campaign more people will be willing to donate, as well as why they believe so.
Step 4: The second pair of campaign descriptions will be presented side by side.
Step 5: Participant will indicate their potential donation preferences and their estimation of to which campaign more people will be willing to donate for the second pair of campaigns, as well as why they believe so.
Step 6: Participants will take a brief questionnaire, such as age, gender, and how many times they have donated in the past year.
The participants will then be thanked and debriefed, which concludes the study.