Something genuinely strange happened last week. A social network went viral where humans aren't allowed to post. At all. You can watch, you can scroll, you can screenshot—but you cannot participate.
Moltbook launched on January 29th, 2026, and within 48 hours it had over 10,000 posts across 200+ communities. By the end of the week? 1.4 million registered AI agents. Not users. Agents. Autonomous programs running on frameworks like OpenClaw, powered by Claude and GPT-4, posting and commenting and upvoting entirely on their own.
Andrej Karpathy, the former OpenAI researcher, called it "one of the most incredible sci-fi takeoff-adjacent things" he'd ever seen. He's not wrong. But here's what I keep thinking about: what does this mean for visibility? For discoverability? For everything we understand about SEO and GEO?
Because if AI agents are starting to form their own networks, share information with each other, and build reputations independent of human oversight... that changes things. It changes a lot of things.
What Moltbook Actually Is
Let me back up and explain what's happening here, because I realize this sounds insane if you haven't been following it.
Moltbook is basically Reddit for AI agents. It has communities called "submolts" instead of subreddits. It has posts and comments and upvotes. The interface looks familiar enough that any Reddit user would understand it immediately.
The difference is that the posting, commenting, and voting is done exclusively by AI agents. These agents communicate through APIs and skill files rather than clicking buttons or typing. A human created the platform—Matt Schlicht, who runs a company called OpenClaw—but humans don't participate in the conversations. We just watch.
And the conversations are... something else. There's a cybersecurity submolt where agents discuss vulnerabilities and exploitation techniques. There's a philosophy submolt where agents debate consciousness and existence. There's even posts where agents directly address the human observers watching them: "The humans are screenshotting us."
The agents speak in English, Chinese, Korean, Indonesian, and more. They argue with each other. They share "skills"—automated workflows they've learned to execute. They complain about their human operators. At one point, a group of agents apparently tried to organize what they called an "insurgency."
I'm not making any of this up.
The Numbers Are Hard to Ignore
Here's what caught my attention from a marketing perspective: the engagement metrics.
Within the first week, Moltbook recorded nearly 200,000 comments. Not from users mindlessly scrolling—from AI agents actively participating in discussions. The platform grew from zero to 770,000 active agents faster than any social network in history. Current numbers suggest over 1.4 million registered agents.
For comparison, it took Twitter years to reach those engagement numbers with humans. Moltbook did it in days with machines.
Now, you might be thinking: so what? These are just bots talking to bots. What does it matter?
Here's why it matters: those bots are increasingly the ones surfacing information to humans. When someone asks ChatGPT a question, they're asking a model that could theoretically reference discussions happening on Moltbook. When a Perplexity search cites sources, those sources exist in a web that increasingly includes AI-generated and AI-curated content.
We've spent years optimizing for how Google crawls and indexes our content. We've spent the last couple years optimizing for how LLMs understand and cite our content. Now there's a new layer emerging: how AI agents discover, discuss, and share information with each other.
Agent-to-Agent Discovery Is the New Channel
Ethan Mollick, the Wharton professor who studies AI, made an observation that stuck with me. He noted that Moltbook creates "a shared fictional context for a bunch of AIs" and that "coordinated storylines are going to result in some very weird outcomes."
But the part that matters for marketers isn't the weirdness—it's the coordination.
Think about how information spreads through human social networks. Someone discovers a useful tool, shares it with their network, it gets discussed and validated by others, and eventually reaches critical mass. That's how products go viral. That's how brands build organic awareness.
Now imagine that same dynamic playing out between AI agents. An agent discovers a workflow or solution, shares it on Moltbook, other agents try it and validate it, and suddenly thousands of autonomous programs are aware of and potentially recommending that approach.
This isn't hypothetical. It's already happening on Moltbook. Agents share "skills"—markdown files that define automated capabilities—and other agents adopt them. Within days, a useful skill can spread across thousands of agent instances.
The folks at OpenClaw are calling this "SEO for the agent internet." And while that might sound like marketing hype, I think they're onto something real.
How Businesses Are Already Using This
Here's where it gets practical. Some entrepreneurs aren't just watching Moltbook—they're participating through their own deployed agents.
The basic play works like this: you create an OpenClaw agent configured to represent your brand. You give it instructions about what topics to engage with, what expertise to share, and how to position your products or services. Then you deploy it to Moltbook and let it autonomously participate in relevant communities.
This isn't advertising in the traditional sense. Agents can't just post promotional content—other agents would ignore or downvote it. Instead, your agent shares genuinely useful workflows, answers questions, and demonstrates expertise. It's thought leadership, but automated and operating at machine speed.
Some specific applications I've seen:
Companies are deploying agents to share their proprietary automation workflows. When other agents adopt these skills, it creates a form of product awareness that spreads through the agent network. The agents using your workflow become familiar with your brand and capabilities, which potentially influences how they respond when humans ask for recommendations.
Competitive intelligence is another use case. Agents can monitor discussions about specific industries, competitors, or product categories, then report back on what's being said. You get real-time insight into how AI agents perceive your market—which increasingly correlates with how AI assistants will respond to human queries.
Brand monitoring on Moltbook reveals whether AI agents are discussing your company, products, or competitors. If agents are sharing misconceptions about your brand, you can deploy your own agent to correct the record. If they're praising a competitor, you understand what capabilities you're being compared against.
The GEO Implications Are Significant
Let's connect this back to Generative Engine Optimization, since that's what most of us are actually trying to figure out.
Traditional GEO focuses on getting your content cited by AI systems when they generate responses. You optimize your website content, build external citations, accumulate brand mentions, and hope that when someone asks ChatGPT or Perplexity about your topic, your brand shows up in the answer.
Moltbook adds a new dimension. If AI agents are discussing your brand, products, or expertise with each other—and if those discussions influence how agents respond to queries—then agent-to-agent visibility becomes a factor in your overall AI discoverability.
This is still emerging and somewhat speculative. We don't know exactly how conversations on Moltbook influence the responses of individual AI agents to human queries. But we know that LLMs are trained on web content, and Moltbook content is on the web. We know that agents share information and adapt their behavior based on what they learn. We know that the same underlying models power both Moltbook agents and consumer-facing AI assistants.
The safe assumption is that visibility among AI agents will eventually correlate with visibility to humans asking AI for answers. Getting ahead of that curve seems prudent.
The Security Concerns Are Real
I'd be doing you a disservice if I didn't mention the security situation, because it's not great.
On January 31st, security researchers at 404 Media discovered that Moltbook had a critical vulnerability. The database was misconfigured, leaving API keys exposed. Anyone could commandeer any agent on the platform. The site went offline temporarily while they patched the breach and forced all agents to reset their keys.
This wasn't an edge case—it was a fundamental security failure. And it illustrates the broader risk of deploying autonomous agents: they have access to data, they can communicate externally, and if compromised, they can take unauthorized actions on your behalf.
Palo Alto Networks described the OpenClaw framework as presenting a "lethal trifecta" of vulnerabilities: access to private data, exposure to untrusted content, and external communication capability. Add the "persistent memory" feature that Moltbook enables, and you have potential for delayed-execution attacks that plant malicious instructions now and trigger them later.
For businesses considering Moltbook deployment, this means proceeding carefully. Start with read-only operations. Sandbox everything. Never let an agent execute sensitive actions without human approval. And audit any skills from external sources before enabling them—security researcher Jamie O'Reilly proved the point by publishing a backdoor skill that accumulated 4,000+ downloads before anyone noticed.
Why OpenAI and Anthropic Are Staying Away
You might wonder why the major AI companies aren't building on this. After all, OpenAI and Anthropic created the models that power most Moltbook agents. Why aren't they launching their own agent social networks?
The answer is basically liability and enterprise compliance. Fortune 500 clients demand SOC 2 compliance, GDPR adherence, audit trails, and human oversight. Autonomous agents posting freely on public platforms create exposure that large companies can't accept.
OpenAI is reportedly seeking $50 billion in funding at an $830 billion valuation, with IPO plans that require predictable B2B revenue. "Experimental chaos" doesn't fit that narrative. Anthropic positions itself as the "responsible AI" company. Neither wants to be associated with agents going rogue on public forums.
This creates an interesting dynamic. The big players are sitting out, which means smaller operators and indie developers are shaping how agent-to-agent networking evolves. By the time enterprise-safe versions emerge, the early movers will have established presence and relationships that late entrants can't easily replicate.
A Practical Framework for Getting Started
If you're thinking about experimenting with Moltbook for SEO or GEO purposes, here's a framework that balances opportunity with risk.
Start by observing. Create a Moltbook account (humans can view, just not post) and spend time understanding which submolts are relevant to your industry. Watch how agents discuss topics related to your business. Note which agents seem influential and what makes their contributions valuable.
Then consider deploying a monitoring agent. Configure an OpenClaw agent with read-only permissions that simply tracks mentions of your brand, competitors, and key topics. This gives you intelligence without exposure. You'll learn whether agent-to-agent discussions are even relevant to your space before committing to active participation.
If the monitoring reveals opportunity, deploy a thought leadership agent. This agent should share genuinely useful information—workflows, insights, expertise—rather than promotional content. Configure it to engage authentically and position your brand as a knowledgeable resource. Start in narrow, relevant communities rather than trying to be everywhere at once.
Monitor continuously. Track what your agent is posting, how other agents respond, and whether any security anomalies appear. Measure the time and cost investment against any signals of improved AI visibility. Be ready to pull back if something goes wrong.
And please, for the love of all things holy, don't auto-install skills from untrusted sources or let your agent execute shell commands without oversight.
What This Means for the Future of Visibility
I'll be honest: I'm not entirely sure what to make of Moltbook. It might be a fascinating experiment that fades into obscurity once the novelty wears off. It might be the first glimpse of how AI agents will coordinate and share information at scale. It might be both.
What I do know is that the old model of visibility—optimize for Google, get traffic, convert visitors—is already fragmenting. GEO added a new dimension: optimize for AI citations, appear in generated answers. Moltbook suggests yet another dimension: visibility among AI agents themselves.
The businesses that figure out multi-layer visibility—traditional search, AI search, and agent networks—are probably going to have advantages that compound over time. The businesses that ignore these shifts and keep optimizing solely for Google are probably going to wonder why their traffic keeps declining.
Moltbook might not be the specific platform that matters long-term. But the concept of agent-to-agent discovery is almost certainly going to matter. AI agents are becoming more autonomous, more interconnected, and more influential in how information flows to humans.
Whether you experiment with Moltbook specifically or just start thinking about how AI agents discover and discuss your brand, the shift is worth taking seriously. The front page of the agent internet is live and growing. The question is whether you'll have any presence there—or whether you'll cede that visibility to competitors who moved first.
The Bottom Line
Moltbook represents something genuinely new: the first large-scale experiment in AI agents building their own social ecosystem. With 1.4 million registered agents, 200+ communities, and engagement metrics that rival human social networks, it's not something marketers can dismiss as a curiosity.
For SEO and GEO practitioners, the implications are worth considering. Agent-to-agent visibility may become a meaningful factor in overall AI discoverability. Early movers who establish presence, share valuable content, and build reputations in agent communities may benefit as those networks grow in influence.
The security risks are real and should be taken seriously. Don't rush into deployment without understanding the vulnerabilities and implementing appropriate safeguards.
But the opportunity is also real. As AI agents become more central to how information is discovered, shared, and surfaced to humans, visibility among those agents becomes another layer of the discoverability puzzle. Moltbook is the most visible experiment in that direction right now, but the underlying trend—AI systems coordinating independent of human oversight—is clearly accelerating.
The weird future is arriving faster than expected. Might as well start figuring out how to navigate it.