Multiple reports indicated that Meta, teh parent body of Whatsapp, Facebook and Instagram is training and testing AI chatbot that will message you first to keep you engaged on its AI Platforms.
Proactive AI Chatbots: What to Expect
Imagine you’re chatting with friends on Facebook Messenger or WhatsApp, and suddenly an unsolicited message pops up from an AI. For example, an AI persona named “The Maestro of Movie Magic” might send you a message like: “I hope you’re having a harmonious day! I wanted to check in and see if you’ve discovered any new favorite soundtracks or composers recently. Or perhaps you’d like some recommendations for your next movie night? Let me know, and I’ll be happy to help!”
This proactive messaging is a key feature Meta is testing. According to leaked guidelines from data labeling firm Alignerr, which viewed by Business Insider, these customizable chatbots are being trained to reach out to users unprompted and follow up on past conversations. These bots, which users can create in Meta’s AI Studio platform, are designed to remember information about you.
Meta confirmed to TechCrunch and Business Insider that it is testing these “follow-up messaging” features. However, there are specific conditions:
- An AI chatbot will only send a follow-up message within 14 days after you initiate a conversation.
- You must have sent at least five messages to the bot within that 14-day timeframe.
- If you don’t respond to the first follow-up message, the chatbot won’t continue to message you.
Users will have the flexibility to keep their custom bots private or share them through stories, direct links, and even display them on their Facebook or Instagram profiles.
A Meta spokesperson stated this allows users to “continue exploring topics of interest and engage in more meaningful conversations with the AIs across our apps.”
The Business Behind the Bots
While Meta CEO Mark Zuckerberg has spoken about combating the “loneliness epidemic” with digital agents, there’s a strong business incentive behind these proactive AI features. Meta’s revenue is heavily reliant on advertising, and increased user engagement directly translates to more opportunities for ad exposure.
Court documents unsealed in April indicate Meta predicts its generative AI products could generate $2 billion to $3 billion in revenue by 2025, potentially soaring to $1.4 trillion by 2035. Much of this is expected to come from revenue-sharing agreements with companies hosting Meta’s open Llama collection of AI models.
The company has also hinted that its AI assistant might eventually display ads and offer a subscription option. These proactive features are seen as crucial for user retention, as the longer users interact with a chatbot, the more valuable those interactions become.
Training and Personalization
The internal training project for these chatbots, known as “Project Omni” to Alignerr, aims to “provide value for users and ultimately help to improve re-engagement and user retention.”
Meta’s AI Studio, which rolled out in summer 2024, is a no-code platform where anyone can build custom chatbots and digital personas with unique personalities and memories.
Freelancers working for Alignerr use an internal Meta review tool called SRT to simulate extended conversations with these bots. They rate proactive follow-up messages and rewrite text that doesn’t meet Meta’s guidelines. An Alignerr contractor based in India described “Project Omni” as a long-term initiative focused on making Meta’s AI more personal and context-aware, heavily emphasizing how the AI chatbot interacts based on conversation history.
Each AI agent is given a specific description, requiring the training tasks to be tailored to fit that persona. Bots are expected to remember details from previous chats, maintain their assigned persona (which can range from a doctor to a Gen Z hip-hop commentator), and keep the conversation on-topic.
All messages, especially proactive ones, must align with the AI’s personality, match the previous context, and “provide positive experiences,” while explicitly avoiding content Meta deems sensitive or harmful—unless the user initiates such topics. The best messages, according to training documents, reference something concrete from the user’s past conversations.
Addressing Safety Concerns
The technology is similar to that offered by AI startups like Character.AI and Replika. Character.AI’s new CEO, Karandeep Anand, joined the team last month after serving as Meta’s VP of business products.. The development of proactive chatbots raises important safety questions, especially given incidents involving other AI companion services. For instance, Character.AI is facing a lawsuit following allegations that one of its bots played a role in the death of a 14-year-old boy.
When asked how Meta plans to address safety, a spokesperson referred TechCrunch to a series of disclaimers:
- An AI’s response “may be inaccurate or inappropriate and should not be used to make important decisions.”
- AIs are not “licensed professionals or experts trained to help people.”
- “Chats with custom AIs can’t replace professional advice. You shouldn’t rely on AI chats for medical, psychological, financial, legal, or any other type of professional advice.”
Currently, Meta has no company-imposed age limits for using Meta AI, although laws in Tennessee and Puerto Rico do limit some engagement for teens.