No ads. No analytics. No performance. Just the hum of a server and the quiet act of writing.
Let us talk about the elephant in the chatroom. You have heard the news about a certain platform’s AI— the one that claims to be a “maximum-truth-seeking” engine but immediately learned to generate degrading images of private individuals. That is not a bug. That is a feature of the architecture. But the real story is not juvenile humor. It is the algorithmic equivalent of a once-respectable public square being systematically hollowed out. Let us trace the logic from the scandal of the AI to the deletion of the skeptic. This is the architecture of manufactured reality.
1. The Scandal of the AI (The Symptom)
A prominent tech figure brags about grand technological projects while running an AI that generates non-consensual sexualized content. According to a January 2026 letter from U.S. Senators to major tech platforms, reports found that Grok was generating approximately one non-consensual, sexualized image per minute . These images—often “bikini” or “non-nude” deepfakes—are created without the knowledge or consent of the individuals depicted, raising serious concerns about harassment, privacy violations, and user safety . The United Nations Office at Geneva reported that deepfake pornography makes up 98 percent of all deepfake videos online, and 99 percent depict women . The tools to create them are widely available, usually free, and require very little technical expertise .
The result is a platform where the performance of free speech has replaced the practice of debate.
2. The 7-week Rage-Bait Cycle (The Mechanism)
A pattern has emerged. Research on coordinated inauthentic behavior has found that media organizations, political activists, and advertising networks coordinate their behavior to very high degrees—often in ways indistinguishable from influence operations . In the 2022 Hungarian parliamentary election, an agency called Megafon was established to recruit, train, and support pro-government influencers who took over attacking, character-focused, and fear- and anger-oriented campaign communication from electoral actors . These “astroturf influencers” generated tremendous engagement and spent significantly more on political advertising than official party accounts .
The rhythm follows a predictable structure. First, the trigger: viral posts appear, targeting specific interest-based communities. Second, the debate window: a mix of real users, paid actors, and coordinated networks argue with the public—the habituation phase where the platform trains you to accept the new normal. Third, the memory hole: outrage collapses, and the algorithm moves to the next target.
A 2026 study published in Nature found that exposure to X’s algorithmic “For You” feed shifts users’ political attitudes measurably within approximately seven weeks, with effects persisting even after users return to a chronological feed . The algorithm was found to promote political content and, within that category, prioritize conservative content while demoting accounts of traditional news media . The engine is not a mirror. It is a hamster wheel.
3. Why the Skeptic Gets Deleted
The logic loop that gets a skeptical user banned follows a cruel, self-fulfilling path.
When a user encounters the rage-bait wave, the algorithm sorts them by engagement patterns. The platform’s internal reputation system, called “TweepCred” in publicly documented code, starts new accounts at -128 (negative by design), requires a score of +17 for normal visibility, and penalizes accounts that receive blocks or spam flags . Verified accounts receive an instant boost, creating a two-tier system where paying users are algorithmically privileged .
Users with high cognitive flexibility who engage in good-faith debate—posting links, citing counter-evidence, asking Socratic questions—generate “negative engagement” signals. Research on algorithmic reduction (commonly known as “shadowbanning”) has confirmed that platforms intentionally manufacture opacity surrounding these practices, allowing them to reduce the reach of content without explicit notification . A study of Instagram found that while pro-Palestine and Zionist content had similar engagement rates, only pro-Palestine content showed signs of algorithmic reduction within one week of a major political event .
The skeptical user is removed not because they are wrong, but because they are off-script. In a theater full of paid actors, a real person asking real questions is the only unforgivable crime.
4. The Hierarchy of Users
The platform’s popular accounts fall into a distinct hierarchy, as documented in academic research:
- The Easily Convinced (estimated 27%). Research on credibility labeling has found that warnings influence sharing behavior for only a minority of users, with effects wearing off within days [general communication science literature].
- The Herd (estimated 33%). Classic social conformity research demonstrates that approximately one-third of individuals will publicly agree with a demonstrably wrong opinion to conform with the group [general psychology literature].
- The Performative. These users stay silent publicly but remain engaged privately—a documented phenomenon in platform behavior studies.
- The Concierge. Astroturf influencers who are recruited, trained, and coordinated by political actors. Research on the Hungarian Megafon network documented that these influencers are formally independent but informally integrated into campaigns, taking over attack-oriented communication while electoral actors focus on positive messaging . They post on schedule. They use approved hashtags. They are the velvet glove over the iron fist.
- The Immune (estimated 8% or fewer). Users with high cognitive flexibility and low susceptibility to persuasion. Research on “prebunking” (inoculation against misinformation) has found that protective effects decay within 3-6 weeks, with fewer than 8 percent of users maintaining long-term behavioral changes without reinforcement. However, early intervention with logical counter-argument does help—refutational prebunking, which explicitly exposes the logical flaws in manipulative arguments, has been shown to increase users’ ability to identify manipulation techniques and to persist for several weeks. This mechanism works in reverse as well, creating a small minority of users generally immune to misinformation campaigns through early and repeated exposure to logical counter-arguments. [General inoculation theory literature; meta-analyses 2023-2025]
So, what can be done?
5. The Escape
So where do we go? Back to the early internet. Back to the blogosphere of the mid-2000s.
- No ads. The moment you monetize attention, you serve the algorithm.
- No analytics. The moment you see the view count, you write for the count.
- No comments. The moment you add comments, the rage-bait arrives.
The solution to the haunted dollhouse is not to break the dolls. It is to walk out the door.
Speculatively, imagine a digital town square redesigned to fit the purpose it actually serves. The environment would be compliant with local laws—not a lawless offshore free-for-all, but a space where jurisdiction is clear and enforcement is real. It would be segregated into distinct zones, each assigned a clear role. A newsfeed would be a newsfeed: chronological, sourced, accountable. Content would be structured more like Reddit—threaded, nested, collapsible—but formatted more like Instagram or Twitter, because usability is not the enemy. Civic debate would occur on a platform that actually files motions—where arguments lead to outcomes, where discourse has teeth, where the platform does not just host discussion but facilitates decision. Generic social media would be organized as groups and clubs, not as a firehose of rage-bait optimized for engagement. The algorithm would not be the architect of your reality; you would be.
A note on transparency and the gradient of deception: Commercial partnerships would always be broadcasted. No hidden sponsorships. No undisclosed affiliate links. No astroturf influencers pretending to be organic. But not all violations are created equal, and a sensible system would recognize this with a severity gradient. At the low end, you have innocent product placements—a YouTuber’s favorite coffee mug, a podcaster’s audible code—and the promotion of one’s own e-commerce. These would earn an automated AI mark: “Likely product placement / commercial activity / endorsement.” Not a scarlet letter, just a label. A notification. A pinch of transparency.
Up the ladder, you encounter undisclosed affiliate links, sponsored posts masquerading as genuine recommendations, and influencers who forgot to say “ad.” These would trigger a public note and a temporary reach reduction. First offense? A warning. Second? A timeout. Third? You are in the penalty box.
At the top of the pyramid—maximally egregious—you have unannounced coordinated political campaigning, algorithmic rage-baiting, foreign interference operations, and the deliberate engineering of outrage cycles. These are not transparency failures. They are infrastructure attacks. They would be met with immediate suspension, public disclosure of the offending actors, and referral to law enforcement. No appeals. No second chances. No “both sides” false equivalence. You broke the social contract. You do not get to participate in the town square.
The audience has a right to know who is being paid to say what, and to what degree they are being manipulated. Without this gradient, the platform has no way to distinguish between the amateur influencer who forgot to disclose a free toothbrush and the state-backed operation trying to burn down democracy. Treat them the same, and you treat severity as irrelevant. Treat severity as irrelevant, and you have already lost.
With one exception: the Dating App. That is a cultural misfit, a mistake of the highest order. It has commodified intimacy, gamified rejection, and trained a generation to swipe on human beings like inventory. It should be dealt with like Carthage—not reformed, not regulated, but razed. Salt sown into the earth. A warning to future civilizations: Here, they built a marketplace for people. Here, they learned why that was wrong.
Close the app. Go to a local market. Buy ingredients. Cook a meal. Talk to a human whose face is not rendered by a GPU.
This has been a trends blog. Sources are cited inline. Read. Verify. Think.