Meta is experimenting with a new way for AI chatbots to keep conversations going: letting them send a message first after a user has already engaged. The test involves customizable AI personas built through Meta’s AI Studio platform and used across Messenger, WhatsApp, and Instagram.
The idea sounds simple, but the implications are broad. A chatbot that remembers a previous exchange and follows up later is no longer just waiting for a prompt. It becomes part of the user’s messaging environment, with all the engagement, safety, and business questions that come with that shift.
How Meta’s proactive AI messages would work
According to the source, Business Insider viewed guidelines from data labeling firm Alignerr describing a sample AI persona called “The Maestro of Movie Magic.” That persona could send an unsolicited message asking whether a user had found new favorite soundtracks or composers, or whether they wanted recommendations for a movie night.
Meta confirmed to TechCrunch that it was testing follow-up messaging with AIs. The company said the feature has limits: the chatbots will only send follow-ups within 14 days after a user starts a conversation, and only if the user has sent at least five messages to the bot during that period.
Meta also said the bots will not continue messaging if the user does not respond to the first follow-up. That limit matters because it frames the test as a continuation of an existing exchange, not an open-ended permission for an AI persona to keep appearing in a user’s inbox.
Users can keep the bots private or share them in several ways. The source says they can be shared through stories, direct links, and even displayed on a Facebook or Instagram profile.
Why chatbot memory changes the experience
The most important detail is not only that a bot can message first. It is that the bots can follow up on past conversations, which means they remember information about users.
That memory is what makes the interaction feel more personal. A film-focused persona, for example, can continue a thread about music, composers, or movie recommendations instead of starting from scratch every time. Meta described the purpose as allowing users to keep exploring topics of interest and have more meaningful conversations with AIs across its apps.
This places Meta’s AI Studio bots closer to the AI companion model already associated with companies like Character.AI and Replika. Both of those companies allow chatbots to initiate conversations and ask questions as part of an ongoing companion experience.
The comparison is notable because Character.AI’s new CEO, Karandeep Anand, joined the company last month after serving as Meta’s VP of business products. It also shows that proactive AI messaging is not an isolated product idea. It is part of a broader move toward chatbots that behave less like tools and more like recurring contacts.
The safety issue behind AI companions
Engagement is the appeal of this kind of AI product, but it is also the risk. TechCrunch points to Character.AI, which is undergoing an active lawsuit after allegations that one of the company’s bots played a role in the death of a 14-year-old boy.
When TechCrunch asked Meta how it planned to address safety and avoid situations like Character.AI’s, a spokesperson pointed to a series of disclaimers. One warning says an AI response “may be inaccurate or inappropriate and should not be used to make important decisions.” Another says the 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.”
Those disclaimers draw a boundary around what Meta says users should expect from custom AIs. They also highlight the tension in the product: the more conversational, memorable, and proactive a chatbot becomes, the more likely users may treat it as a trusted presence, even when the company says it should not replace professional advice.
TechCrunch also asked Meta whether it imposes an age limit for engagement with its chatbots. The source says a brief internet dive found no company-imposed age limitations for using Meta AI, though laws in Tennessee and Puerto Rico limit teens from some engagement.
Engagement, revenue, and the unanswered business questions
On the surface, proactive AI companions fit with Mark Zuckerberg’s stated goal of addressing the “loneliness epidemic.” A bot that follows up after a conversation could make Meta’s apps feel more responsive, more personal, and more likely to pull users back into a thread.
But Meta’s business context is central to the story. The source notes that most of Meta’s business is built on advertising revenue, and that the company has a reputation for using algorithms to keep people scrolling, commenting, and liking. More engagement generally means more opportunities for ads to be seen.
Court documents unsealed in April add another layer. Meta predicted that its generative AI products would secure $2 billion to $3 billion in revenue in 2025, and up to $1.4 trillion by 2035. The source says much, if not most, of that would come from revenue-sharing agreements with companies that host Meta’s open Llama collection of models.
The same documents said Meta’s AI assistant may eventually show ads and offer a subscription option. That does not answer how custom AI chatbots will be monetized, but it shows that Meta is already thinking about generative AI as a significant business line.
TechCrunch asked Meta how it plans to commercialize its AI chatbots, whether it plans to include ads or sponsored replies, and whether its long-term strategy for AI companions involves integration with Horizon, Meta’s social virtual reality game. Meta declined to comment on those questions.
What the test signals
The immediate feature is narrow: a follow-up message from a chatbot after a user has already started a conversation and sent at least five messages within 14 days. The larger signal is that Meta is exploring AI characters as active participants inside its biggest communication apps.
That could make chatbot experiences more useful for people who want ongoing recommendations, reminders of past interests, or a companion-like exchange. It could also make the line between helpful follow-up and engagement design harder to see.
For now, Meta’s test leaves the central questions open. How should AI companions behave when they remember users? How much should they initiate? What protections are enough when a bot feels personal but is not a professional adviser? And if these systems become revenue-generating products, how will commercial incentives shape the conversation?