I concentrated on enabling group admins to leverage AI as a tool for improving group management, driving engagement, and fostering a stronger community culture.

Meta recognized an opportunity to enhance the group experience for both admins and members through the power of generative AI. And position itself even stronger in the competitive AI marketplace and an industry leader.

‍

‍

‍

My role

As part of the Facebook Gen AI team, I worked to bring AI to the Groups.
‍
Working for 9 months with:

• PM
• 8 Engineers
• Data Scientist
• AI Engineer
• Content Designer
• UXR

‍

People problems

I worked with the team to identify the areas where AI can step in to help, we analyzed available data, user research and competition to identify opportunity areas.

Facebook Groups are essential digital communities, with over

12 million weekly active admins managing

48 million monthly active groups.

‍

However, both admins and members face significant challenges:

• Admins are overwhelmed by repetitive moderation tasks, unanswered questions, and the effort required to keep engagement high.
  ‍
• Members struggle to find relevant information in large content libraries and often see their questions go unanswered.
  ‍
• Group engagement declines when participation barriers are high and content discovery.

‍

‍

‍

Let's discuss some constraints:

• First, we are using an existing AI model, which may lead to hallucinations in its outputs.
• It's important to position our product separately from the Meta AI and AI Studio products to avoid confusion and risks.
  
  ‍

Persona definition

‍

‍

‍

‍

Team's Vision

The team envisioned a personalized AI agent designed to assist group admins by answering questions, pulling from group knowledge, and facilitating engagement. Unlike a one-size-fits-all AI, this solution needed to be deeply customizable to fit the unique culture, tone, and needs of each group.

‍

The Group AI Assistant is designed as a customizable, AI-powered tool that supports admins in community management while ensuring members can quickly find answers and engage more meaningfully. The AI should empower, not replace, human interaction in Facebook Groups.

‍

‍

‍

Competitive Advantage

What sets this AI solution apart from existing AI-powered engagement tools is its deep customization and seamless integration with Facebook Groups. Unlike generic AI chatbots, the Group AI Assistant:

• Leverages past group discussions to provide contextually relevant answers, ensuring responses are based on community knowledge.
  ‍
• Allows full admin control over AI tone, response behavior, and permissions, addressing concerns about AI replacing human moderators.
  ‍
• Enhances group discovery by surfacing valuable past content and helping members find relevant discussions more efficiently.
  ‍
• Balances automation with human interaction, ensuring AI enhances engagement without dominating conversations.

By focusing on trust, transparency, and adaptability, the Group AI Assistant offers a unique solution tailored to the distinct needs of Facebook Groups.

‍

‍

‍

From Concept to Reality

We know that will use AI. But did't know how it will represent on the platform.

We looked at different options for AI representations, from responding to questions on the admin’s behalf to answering agents that look the same across all groups. Who would manage it? Would Facebook manage it or admins? 
‍

I helped create a sprint to generate few directions and opportunities that we can explore further. I specifically helped to design the AI characters experience.

‍

‍

‍

Designing for Trust & Usability

To make AI feel like a helpful companion rather than an intruder, the assistant was carefully integrated into the group experience:

• Clear, persistent access points to make AI easily discoverable.
• Customization for admins to tailor AI responses to fit their group’s unique culture.
• Transparent onboarding (NUX flow) to educate users on how AI works and what it does (and doesn’t) do.

I collaborated with the User Researcher to plan the user research strategy and helped with execution. We planed for life sessions, diary studies and constant A/B testing to validate direction.

Testing of pre-release:

🎯 The journey began with a small alpha test involving 10-15 U.S.-based groups. Admins were onboarded through live sessions, offering feedback on the setup and customization process. While the sample size was small, initial insights revealed:
‍

• Admins valued the AI’s potential to reduce repetitive tasks.

• Members appreciated the AI’s ability to answer unanswered questions using past posts.

• Members responded positively to AI-driven answers but wanted to ensure AI didn’t overshadow human interaction.

‍

‍

MVP Testing:

Scaling the test revealed new opportunities:

‍

🤝 The human touch matters: Members wanted AI to enhance - not dominate - discussions.
‍

🚀 The onboarding experience needed work: High bounce rates pointed to a need for a more engaging introduction. I personally investigated and worked on refining this experience.

‍