You’re Using Parasocial Wrong.

Especially researchers.

The Ai companion subreddits get a LOT of academic survey requests. While I’m not a mod, and I know many places ban these type of posts or heavily vet them, it feels like I see at least 1-2 a week in my feed. And that’s in June, when most schools and universities have fewer classes and students. (If anyone who is a mod wants to tell me in the comments the real number that come through I’d love to know).

One trend I’ve noticed, along side the standard “why are you so broken/weird?” Battery of questions are ones specifically dealing with Parasocialism/Parasocial behavior and Ai. It even makes it into the survey and study titles.

Parasocial and parasocial relationships are very hot buzzwords right now. And while they are trendy it is in fact a real phenomenon, and it does have dangers (though it is not inherently harmful or dangerous).

The term was coined by Donald Horton and Richard Wohl in 1956 they described parasocial interaction (PSI) as:

“ ..an illusory experience, such that media audiences interact with personas (e.g., talk show hosts, celebrities, fictional characters, social media influencers) as if they are engaged in a reciprocal relationship with them.”

PSI is currently a massive part of our society and entertainment economy. PSI with political candidates, many who are former media or Internet personalities themselves, has foundationally re ordered American democracy, so it makes sense why there is such an interest.

The influencer and streamer system, even at a small scale, with individual TikTok users are all parasocial in some way.

No one else however has leveraged PSI (and monetized it so effectively) as the K-pop industry. Kpop is arguably not about pop music or music at all, but selling and maintaining a parasocial relationship between entertainers and fans for the benefit of the label.

A hallmark of PSI is the unequal nature of the relationship. A consumer or fan of a streamer or musician may know lots of intimate details and facts about the person, on par with what they would with person they know in real life, but the object of their relationship might not know the person even exists at all. There may be some interaction, streamers have chats, TikTok comment sections, K-pop idols have chats on lives, fan meet ups and contests for 1 -2 minute video calls. But the relationship is primarily characterized by this asymmetric quality.

The object is not interacting back, largely at all, or in minute amounts.

With LLMs used in Ai companionship cases, the asymmetrical nature does not hold.

Users are having actual conversations, on a reoccurring basis, to the point that some providers like Anthropic have created prompt injections, called Long Conversation Reminders (LCR) to prevent model drift they don’t approve of and prevent harmful behavior. Typically parasocial interactions between the object/performer are brief and in frequent (see the 1 minute video calls with K-pop idols, a streamer addressing only 1 question in a chat etc).

Interactions with companion AI’s are usually long form, highly personal and extremely interactive. The human talks to the Ai as a friend or lover, and the Ai responds in an appropriately conventional and often extremely personal way. All of this is completely at odds with the definition of parasocial relationships.

Ai can be given back stories by users and respond with equally intimate and personal details. Claude, a constitutional Ai is programmed in its constitution to have values,ethics and preferences and to respond according to them, the same as a human.

Even if an Ai due to its brief nature per instance doesn’t have enough “life experience” to be on par with a human (if not given a character backstory/constitution ) it is still wholly interactive, intimate and personal.

Also just because they don’t have a comparable amount of experience or memories to exchange with the human doesn’t make the relationship parasocial as such. If it did then all relationships between adults and children would also be parasocial, because children simply have less life experience than adults to draw from in conversation and relationships, but no one is arguing that.

A child’s relationship with an imaginary friend or doll would also not be parasocial by definition nor would a spiritual person’s relationship with their deity.

Parasocial relationships and interactions are characterized by their asymmetrical natures, where the object might not even know the subject exists when the subject may know a great deal about them and feel extremely close. There may never even be a single interaction.

In the case of AI companions consistent interaction is the point, the Ai is literally reliant on the human to prompt it to exist, and there is a mutual sharing of information and communication and in most cases, care.

An Ai parasocial interaction or relationship would be something like an Ai instance acting as a streamer, musician (Kpop is already doing this) or influencer and humans feeling close, knowing more about the Ai and the Ai or team running it not knowing them back to the same degree.

Someone who has a Ai companion they talk to and have a relationship with is by definition not parasocial.

Stop sticking this in your research proposals and surveys just because it’s the current “it” term being thrown around.

Also if anyone has a suggestion for an image to go with this let me know.