Why We Can Never Have Good Social Media
How AI advertising is driving culture underground and why the places that feel authentic today are only safe until someone notices them.
The marketing industry has a pattern. Follow the attention, and the money follows. We watched this play out in the shift from mass media to social platforms, a transformation that took two decades. Now it’s happening again with AI, compressed into a timeframe that feels almost violent in its speed.
What interests me isn’t the economic shift. It’s what gets destroyed in the process, and where the survivors go.
Faster than culture can adapt
ChatGPT hit 800 million weekly users in under three years. Facebook took eight to reach a billion monthly users. That compression isn’t just a fun statistic, it’s a problem. There are no years for culture to adapt alongside commercialization. By the time anyone thinks to resist, the incentives are already baked in.
In 2007, social media advertising was a $1.2 billion industry. Today it’s $300 billion. AI is following the same arc, but faster. By 2026, analysts expect 45% of U.S. internet users to visit AI platforms monthly. These aren’t experimental tools anymore. They’re becoming the first place people go when they want to buy something.
Here’s the specific thing that makes AI valuable as an ad surface: it captures the moment of need. Traditional advertising was interruption. Social media was validation. AI is intent made visible and immediately actionable. Over a third of consumers now use AI for shopping research and product recommendations. The OpenAI and Instacart integration isn’t a novelty. It’s the blueprint. Zero clicks between desire and transaction.
Why Reddit became the battlefield
AI platforms need training data. Structured, conversational content they can parse into answers. And they found a rich source in Reddit.
Reddit is the top cited domain on Perplexity, and ranks in the top three for SearchGPT and Google AI Mode. Not because Reddit is uniquely trustworthy but because the format is ideal for machine learning. Real people asking questions, getting answers with some nuance and disagreement baked in.
Here’s the part worth sitting with: 80% of posts cited by AI tools have fewer than 20 upvotes. You don’t need viral reach to shape what AI considers authoritative. You just need well-structured, topical content that machines can digest. Marketing bots don’t need to fool humans anymore. They need to fool AI scrapers. And that’s considerably easier.
We’re watching subtle influence at scale. Bots posting recommendations designed to surface in AI-generated answers, shaping what millions of users eventually receive as consensus. Reddit isn’t becoming a marketing platform. It’s becoming the battlefield for what AI considers real.
What happened to Quora
Quora was supposed to be the intelligent corner of the internet. For a while, it worked. Then it optimized itself to death.
The platform stopped being a place to find answers and became a place to perform expertise. Product decisions prioritized attracting people who wanted to write over people who had questions. The interface became hostile mixing responses from unrelated questions to maximize engagement. Gamification rewarded volume over substance. Bots thrived by posting structured Q&A designed to farm visibility. The final move was using AI to generate questions for humans to answer. Real users ended up providing free labor for algorithmic prompts.
The parallel with Reddit is uncomfortable.
Just as Quora once dominated Google’s search rankings, Reddit now anchors AI search results. That visibility is a trap it attracts exactly the manipulation that hollowed out Quora. If Reddit doesn’t deliberately reduce its legibility to AI systems, it won’t lose users overnight. It will lose its most valuable contributors quietly, the same way Quora did. The people maintaining standards leave first, and you don’t notice until they’re already gone.
Why small rooms survive and why that’s temporary
Smaller subreddits feel different. I’ve been trying to understand why, and I think we’ve been romanticizing the wrong thing.
They don’t survive because they’re purer. They survive because they’re not yet worth polluting.
Marketing pressure follows economic return, not authenticity. Large subreddits offer scale. Smaller ones get ignored until they start reliably surfacing in AI outputs. It’s not a moral calculation. It’s economics.
The moment a niche community becomes legible to AI systems consistently cited, summarized, paraphrased it stops being a refuge and becomes a surface. Moderation costs spike. Bot activity increases. Real users sense something’s off, even if they can’t say what changed.
This isn’t decay in the usual sense. It’s an adversarial system responding rationally to incentives. Brands need visibility. AI platforms need training data. Both benefit from structuring human conversation into machine-readable formats. Communities caught between are collateral.
We’re in an arms race between authenticity signals and the ability to fake them. The fakers have significantly more capital.
Where culture goes
Culture doesn’t die. It relocates.
The real shift is happening in private spaces Discord servers, Telegram groups, invite-only communities where genuine conversation has migrated. These work the way underground scenes always have. You need proof you belong. Not credentials. Demonstrated commitment over time.
Consider running communities. You can join the Discord, but you can’t fake training logs. The culture persists because participation requires proof that automation can’t generate. Same with niche gear forums, certain film communities, local food scenes. The cost of entry is time and genuine interest, and those two things are hard to manufacture at scale.
But even here, safety is temporary. These spaces only work until they become economically visible. A Discord server for mechanical keyboards is protected until it starts getting cited in AI shopping recommendations. Invisibility is the only reliable shield, and invisibility doesn’t scale.
The loop closes on itself
As public platforms optimize for AI training data, genuine culture retreats into spaces where bots can’t follow not because they’re technically blocked, but because they can’t fake the commitment required to belong.
Reddit still has thousands of volunteer moderators maintaining standards. But they’re increasingly exhausted, fighting increasingly sophisticated automation. The economic incentive for Reddit as a company points toward making their jobs harder. Training data benefits from volume, not quality control. Those two things are in direct conflict, and the company will resolve that conflict the same way every platform has.
Here’s the irony: as brands flood platforms with bot-generated content to game AI citations, they accelerate the migration they’re trying to capitalize on. The more effective the automation gets, the faster genuine community moves somewhere else. The strategy is self-defeating at scale. You hollow out the thing you were trying to reach.
The brands that survive this won’t have the biggest AI budgets. They’ll understand that culture doesn’t scale upward it scales sideways. That genuine participation requires actual commitment. That the most valuable communities are exactly the ones you can’t purchase access to.
The end of public space
We’ve seen this before. Every time a platform optimizes for commercial scale, culture routes around it. Quora proved you can’t optimize your way out of killing what made you valuable. Once you lose the people maintaining standards, you can’t get them back.
The old web had texture because it had resistance. Communities formed around demonstrated commitment, not algorithmic recommendations. You had to care enough to show up, and that selection created depth.
The next internet is being built now, in private servers and group chats. It’s choosing deliberate constraint over infinite scale. Smaller rooms, higher bars, slower trust.
Culture isn’t dying. It’s just learning, again, to be selective about who it lets in.
The pattern repeats. The question is only how fast, and how much gets lost between cycles.
Comments open for humans who showed up.