Celebrity Just Might Be Meta’s Edge in the AI Chatbot War

Photo illustration of a robot holding a mask of Snoop Dogg in front of his face
Illustration: Variety VIP+: Adobe Stock; Snoop Dogg: JC Olivera/Getty Images

Last week at its annual Meta Connect 2023 event, Meta presented its set of 28 AIs, interactive chatbot characters based on real celebrities, including Tom Brady, Paris Hilton, Kylie Jenner, Snoop Dogg, MrBeast and Charli D’Amelio.

Each will interact with users and offer advice on the star’s specific area of expertise, such as Brady’s persona, “Bru,” on sports; D’Amelio’s “Coco” on dance; and Jenner’s “Billie” with “ride-or-die” advice. The chatbots will be supported by Meta’s large language model (LLM) Llama 2 and made available across Meta’s family of social and messaging apps including Instagram, WhatsApp and Facebook — and could further be extended into VR experiences coming to Meta’s Quest 3 mixed-reality headset.

No specifics were given as to Meta’s business model for these AIs or payment models for celebrities involved, but it suggests a new kind of creator monetization opportunity.

So far, efforts to allow talent to extend their personal brand at scale have been only marginal, with a few celebrities or creators cloning themselves to create and launch AI replicas as interactive experiences for fans.

In May, a popular Snapchat creator launched CarynAI as an AI-powered girlfriend, powered by OpenAI’s GPT-4 and created by startup Forever Voices, offering fans personalized interactions at $1 a minute. Celebrities have also been exploring their own options, with some bringing their likenesses to AI-fueled customer experiences in partnership with brands.

Meta’s integration of AI chatbots also participates in an emergent category of conversational AI-based apps, positioning to offer unique services or user experiences and types of engagement. OpenAI’s ChatGPT and Google’s Bard are chiefly positioned as question-answering services or productivity aids.

But alongside them, one-to-one conversational chatbots have arisen to offer users forms of entertainment and even companionship — whether that’s as famous people living or dead (e.g., Character.ai), therapists offering advice or emotional support or romantic partners offering connection or just “someone” to talk to (e.g., Replika).

There’s also precedent for chatbots to appear natively in social media apps: In April, Snap launched its chatbot MyAI for all users and had about 150 million as of its Q2 earnings. TikTok is testing its own, called Tako.

In launching its own native AI characters, Meta will rival the likes of Character.ai and Replika. The company has failed more often than succeeded in launching new features, often clones of rival services. (Case in point: Threads.) But when Meta succeeds, it succeeds big. Stories and Reels, copied from Snapchat and TikTok respectively, have massively scaled across Instagram and Facebook’s user bases, added incremental engagement, and gradually account for greater shares of platform ad revenues.

AI chatbots may be just another opportunity for Meta to increase user engagement on its many platforms across social and, increasingly, VR. But AI chatbots are still relatively nascent, starting as web applications and only more recently introducing mobile app versions for iOS and Android. Web traffic to popular chatbots including ChatGPT, Character.ai and Replika has tapered slightly in recent months, per Similarweb data shared with VIP+.

But even as user traffic has waned, time spent per visit has held steady or grown. Notably, engagement time with Character AI has been consistently higher than even for ChatGPT, measured as average minutes per visit on desktop and mobile web.

Despite app versions being newer, time spent improves considerably for Character AI on its mobile app version versus web. In August 2023, Android users spent an average of 64.6 minutes per visit versus almost double session times on desktop and mobile web (33.4 minutes).

While generative AI has been positioned for many uses, users are predominantly interested in using generative AI tools for fun.

But despite some positive signals surrounding “personified” AI chatbots, AI-powered virtual romantic partners, virtual friends and AI-generated social media personalities still rank lower among consumer wants for generative AI-powered media experiences. Further, percentages of those citing interest haven’t shifted since February 2023.

Most also wouldn’t pay for such applications of gen AI if they were available. Just 10% said they’d be absolutely certain or very likely to pay for an AI-powered virtual romantic partners, followed by AI-generated social media personalities (16%) and virtual friends (19%).

Taken together, Meta faces some headwinds as it launches AI chatbots. Virtual friends and personalities on social media are still in the minority of user interests among AI-generated media experiences. Naturally, chatbots come with their own set of content risks that make the outlook somewhat murkier, including the unpredictable nature of LLMs, even when they’re trained to interact in prescribed ways. Snap’s My AI has received negative feedback, despite over 150 million users engaging with it. Users might also have concerns about how Meta intends to use data from their conversations with AI.

But some factors support Meta’s effort. It’s clear that those who already engage with these chatbots spend a lot of time doing so, particularly on mobile devices. Likewise, most gen AI users already prefer engaging with the technology to have fun.

In contrast to “blank-slate” AIs, basing its chatbots on popular personalities could be the critical factor that makes Meta’s AIs trustworthy and compelling AI companions, ones users are curious to at least try and more comfortable to engage with regularly over time.

See our full catalog of artificial intelligence articles ...

Read the Report