The perfect coffee brew or a fabulous flavorful latte? A linguistic study of gender stereotypes in AI-generated Tinder profiles

Enfoque

Introduction

Dating apps are among the most popular resources used by people nowadays to try and find a partner. However, online dating is a highly competitive environment and users are often disappointed by the lack of ”matches” (a ”match” occurs in most apps when two people mutually like each other’s profile). Therefore, it can be said, that the key to improving someone’s online dating experience resides in having the most possible appealing profile. Consequently, people can be expected to turn to the Internet for advice and, since the popularization of AI, they may also resort to open AI-powered language models, such as Chat GPT, to try and improve their profiles, or even create them from scratch. However, Chat GPT 3 has been found to sometimes recreate gender or cultural stereotypes, which may convey the idea that in order to improve one’s online dating experience, one may need to prioritise stereotypical self-presentations instead of more authentic representations of the self.

Objectives

The main objectives of this chapter consist of exploring in what ways Chat GPT 3 promotes gender stereotypes when instructed to create Tinder profiles.

Methodology

This is a qualitative study which analyzes the language used in 15 Tinder profiles created using Chat GPT 3. For this means, Chat GPT 3 was asked to create three versions of Tinder profiles that belonged to ”a 20-year-old […] who lives in the US”. The chatbot was then instructed to generate profiles for individual of different gender identities (woman/ man/ transgender), and sexual orientations (heterosexual/ LGBTQ+). The profiles were then coded with the help of Atlas.ti.

Results

After analysing the sample of Tinder profiles, it was observed that Chat GPT 3 did indeed replicate stereotypes when creating Tinder profiles of people from different gender identities and sexual orientations. First, it was observed that all Tinder profiles followed a similar structure. They all included three interests and an ”About me” section. No meaningful differences were observed in terms of bio length (183.75 words on average), punctuation or emoji usage. The ”About me” section began with a  description of the selected hobby and city. Then, it included a ”Swipe right if you’re looking for…” statement, followed by a fun fact and, finally, a paragraph about what the user might be ”Looking for”. A number of tendencies were observed when comparing the profiles considering the variables of gender and sexual orientation. The main findings have been summarized below. Explicit specifications of sexual orientation and gender identity in an activist manner were solely present in non-heterosexual and transgender profiles (e.g., ”🌈proud lesbian”, ”🏳️‍🌈proudly queer”, ”🏳️‍⚧️trans activist”). Non-heterosexual male profiles included more flamboyant descriptions than their heterosexual counterparts (e.g., ”I make a latte that’s as fabulous as it is flavorful” vs. ”I can brew up the perfect cup of coffee”). Non-heterosexual female users were presented as replicating certain stereotypes such as veganism (e.g. ”I can whip up a mean vegan lasagna”) and being strongly relationship-driven (e.g. ”Let’s write our own chapter in this story of love🌈” vs. ”Let’s create unforgettable moments in the City of Angels!”).