
Attention Economy: From Pornhub to Twitter, from Pornographic Content to Cryptocurrency, a History of Battles to 'Deceive' Platform Recommendation Algorithms
This excerpt is selected from my operational data records of managing eleven hongkongdoll accounts over the past five years. Some content has been mentioned in my previous posts, some has been blurred and is not replicable. It's long but I find it interesting, though classmates not interested in account operations can skip it. I believe this article is worth reading for those already in the game but not yet aware of what they are doing.
There has always been a lot of content on Twitter about KOLs and traffic monetization. I've read some but never expressed any opinions. I took the entire month of November off. On one hand, I needed quiet time to re-plan the routes for accounts in various tracks and recruit new members; on the other hand, after four years, I truly needed some completely blank rest. While on vacation, I suddenly felt like writing something interesting. I think I have some say in the traffic business. Writing based on interest is my only purpose for long-form articles now. There won't be many more, so enjoy them while they last.
A Pornhub Bug That Changed Destiny
In early 2021, I was just a newbie blogger who had just entered the industry, and few knew the name "hongkongdoll." In April 2021, I accidentally discovered a small bug in Pornhub's recommendation algorithm. At that time, Pornhub was in an extremely special time window. The lawsuit over underage and illegal video content had just ended, payment channels like Visa and Mastercard had terminated cooperation, the platform was severely hit, and the video publishing system underwent a restructuring. All videos needed to be re-verified for age and undergo stricter manual review. During that period, I used Google Sheets to monitor traffic changes for different video subjects in my channel to confirm topic directions. But I repeatedly noticed an anomaly: videos that were re-approved after review would receive significantly abnormal recommended traffic within a short time. After further observation, I gradually confirmed that at that time, once a video was taken down and re-uploaded, it would be given a new time weight while retaining its original view and interaction data. This bug allowed videos to be continuously selected by the system from the traffic pool. In simple terms: if I took down a video that had already gained millions of views and re-uploaded it, the system would mistakenly think it was a new video that had exploded with millions of interactions in a very short time, thus instantly giving it the highest recommendation weight and placing it on the homepage. Repeating this operation could create a terrifying artificial traffic flywheel.
This bug's time window lasted nearly three months. Through this operation, I climbed to the global number one position on Pornhub's amateur chart. Almost all my videos were constantly hung on the homepage, with average views overwhelmingly surpassing the second place. At the same time, through Pornhub's vertical traffic push, my paid subscription count on OnlyFans grew geometrically in a short time. A massive amount of new traffic poured in daily, increasing 40 times within a few weeks. With additional paid content activation and continuous subscription attraction, this level of monthly income, after slow decay, lasted for over three years.
This exploitation of the bug was essentially a recommendation algorithm attack on Pornhub. Using this bug is exactly why I used the word "deceive" in the title. This was also my first attempt at exploiting a bug. Without this bug, I likely wouldn't be in your sight today. Because of this bug, my destiny changed.
Changing Battlefields: Twitter's Crackdown on Sensitive Content and the Breakout
Besides Pornhub, Twitter has always been one of the most important platforms for online sex workers. In 2021, I created my first Twitter account. Through traffic diversion from Pornhub, my follower count broke through one million in a very short time. However, Twitter's restrictions on sensitive content were very clear from the start. Once an account is marked as sensitive, its content is almost impossible to be recommended to non-followers. The recommendation pool and follower pool are completely separated. Under this mechanism, the mainstream practice for adult accounts at that time, and even today, hasn't changed much: mutual promotion and sharing follower pools. The problem with this model is obvious: traffic can only circulate within the existing circle. Without new participants joining, growth naturally caps out. The upper limit is the union of all mutual promoters' followers, making it difficult for traffic to extend to non-follower groups.
Twitter's algorithm once favored likes, and explicit sexual content easily gets likes. But once an account is labeled sensitive, this advantage immediately fails, and the recommendation weight is systematically suppressed. The sensitive label is one of Twitter's core demotion mechanisms that has been retained long-term. To solve this problem, some adult MCNs chose to continuously create new accounts for "jailbreaking," using the window period before new accounts are marked as sensitive to gain traffic from the public recommendation pool, then quickly divert traffic, abandon the account, and repeat the cycle. Although this method is effective in the short term, the conversion rate is extremely low and the cost is very high. Essentially, it's a solution that trades human effort for noisy traffic.
The real turning point came in 2022. Elon unexpectedly acquired Twitter, leading to large-scale layoffs, system streamlining, and frequent iterations, putting the platform into a period of obvious maintenance instability. During this time, Twitter's content recommendation showed rare loosening. As long as users turned on the sensitive content switch, unsensitive content from unfollowed accounts would directly appear on the For You page. From mid-2022 to the end of the year, which I call the Summer of Adult Twitter, my main sensitive account broke through 2.5 million followers during this phase. Not only that, but various opportunities to create artificial traffic flywheels emerged. In 2023, Twitter made the core logic of its recommendation algorithm public. Interactions, especially comments, became key weight indicators. New jailbreaking methods emerged: not posting sensitive content in the main post, but creating highly dense interactions through the comment section. The account itself wouldn't be marked sensitive, but the recommendation weight would be continuously amplified. I'm sure you've heard of the specific method: "Daily Contest" or "Find the Code," along with various similar approaches. An account not marked as sensitive could gain massive interactions and exposure by soliciting sensitive content from sensitive accounts in the comments. Many accounts in various tracks bypassed Twitter's sensitive content restrictions this way. The essence was still "deceiving" the platform algorithm to bypass restrictions and artificially create massive traffic flywheels, thereby gaining huge follower growth. Many well-known Twitter users rose to prominence during this time.
After understanding that the core of Twitter's traffic flywheel is increasing comment interactions, I combined this with the previously known adult account retweet and MCN new account jailbreaking methods to create reverse traffic diversion—a method I've only seen myself using. I added a Chinese lottery account that doesn't post sensitive content. Using just one trick, "comment 1," I gained nearly one million followers. The specific method was using lotteries to increase interactions, combined with having a marked sensitive big account with millions of followers retweet. Initially, I used existing follower interactions to boost the new account's lottery, then entered the public traffic pool, generating a large number of non-follower interactions. Combined with the unmarked sensitive account, this created an artificial traffic flywheel. Then, through the non-sensitive account, I diverted traffic to the sensitive big account. By the end of 2023, my main sensitive account's follower count reached three million. At this point, my accounts had a clear division of labor: the main account served as a reservoir and interaction amplifier, while the subsidiary accounts were solely responsible for triggering the recommendation algorithm, taking the jailbreaking to the extreme.
However, starting this year, this "jailbreaking war" for sensitive content will gradually become ineffective as AI is integrated into the algorithm. More and more AI content recognition actively adds labels to account content. The difficulty of jailbreaking has shifted from simply deceiving the algorithm to figuring out unpredictable AI, almost putting an end to this battle. Although I made a series of plans for my accounts in 2023, as the online adult traffic economy slowly returned to stability or even declined after peaking during the pandemic, the account that ultimately bore the commercial conversion was this subsidiary crypto account, which has now become the only main account @hongkongdoll.
Why I Came to Crypto: Success and Failure Both Due to Twitter
After talking so much about adult content, it's time to return to crypto. But before entering this topic, it's necessary to explain one thing: why I am here. The reason is not complicated; I was brought in by traffic generated by "deceiving" the platform's recommendation algorithm. When NFTs rose in 2021, traffic tactics on Twitter were still very primitive: lotteries, retweets, tagging three friends, leaving addresses. My sensitive big account was almost one of the most tagged accounts in Chinese Twitter at that time. It was during this phase that I first clearly realized that a significant proportion of my followers were naturally distributed crypto users. So I converted one of my subsidiary accounts into one specifically for observing the crypto market, which is this account now, and began to pay attention to potential opportunities in this track. Back then, many NFT project parties approached me for collaboration, hoping I would directly launch an NFT project. But most were low-quality, and the contracts were performance-based, which I didn't accept. At that time, the cost of asset issuance was still high, and you had to go all out. At the same time, I gained a very preliminary understanding of market making and market manipulation. To be honest, I didn't need to open a scam project and get scolded just to earn a hundred thousand U. Unlike today, where you can issue a meme coin with a small wallet in one click, buy at the bottom while shouting and selling, and no one cares about a dozen of them a day. If there's one commonality in the attention economy across different tracks, it's that it always evolves toward shorter, faster, and denser dopamine stimulation. In the end, I didn't participate in that wave of NFT entrepreneurship. But I did play around with it myself, hand-drew my avatar at the time, and even made a small series, originally intended for a lottery. Someone offered 5E for it, but I neither sold nor gave it away. I wonder if anyone still remembers this story.
Later, in 2022, I cashed out a total of several hundred E from Azuki. I was also lucky enough to hedge against ETH, avoiding the Luna crash and UST depeg. This led to rumors on Twitter that I had achieved financial freedom. But in reality, the money I made from luck only amounted to less than half a year's worth of my actual income. The real traffic monetization still came from paid subscriptions. The subsidiary account itself didn't generate any revenue, so it was dispensable to me. Entering the crypto circle at the end of a bull market laid the inevitable groundwork for going to zero during the hellish phase of the subsequent bear market. Everything that seemed beautiful changed because I started browsing Twitter and got involved in short-term high-leverage futures trading. From September to October 2023, I followed a signal provider and blew up my last 100,000+ U. The signal provider euphemistically called it "being good at 20x manual DCA." In November of the same year, I began the story of "borrowing falsehood to cultivate truth" that I wrote about before. Success and failure both due to Twitter.
In the initial traffic diversion phase of 2023, I followed a concise tutorial I had written earlier about creating artificial traffic flywheels. I practiced it for about half a year, and almost every tweet had millions of views. The principle was to include interaction operations and exposure operations in every tweet. Lotteries, asking for resources, Q&A, creating opposition—these interaction operations increased interactions. High interactions allowed entry into Twitter's recommendation feed, increasing exposure. Adult content, quick money, tutorials, arguments, gossip—these exposure operations increased passive traffic-grabbing interactions. By combining one from each category, you could create a self-reinforcing tweet that entered Twitter's recommendation feed.
I deliberately chose this most efficient traffic path for two reasons. First, my energy was indeed limited in the early stages, so I used the least time investment to create as many artificial traffic flywheels as possible, focusing on speed. Second, my persona itself wasn't suitable for posting content that didn't match it. I didn't post complex content while consolidating the existing persona of "Sister Doll, rich but naive," continuously accumulating principal through advertisements. At a time when I had no principal, this was indeed effective. Novelty traffic was at its peak, and advertising prices were indeed high. But soon, I found that almost all these ads were the last straw—no compounding, no positive feedback, only one-time monetization. The account had no potential for development. Over time, negative feedback began to show: the larger the traffic, the thinner the trust. Actual advertising revenue began to decline significantly after lacking positive effects. The truth is that the pure advertising revenue in the first year was less than one-tenth of the income I earned from simply holding, arbitrage, and new token launches in the past two years. It wasn't until mid-2024 that I fully realized this problem and began to fully dedicate myself to crypto research, gradually reducing photo content and switching to pure text output. I no longer used stories to drive the narrative, only recording my actual operations. During this process, traffic continued to decline, but advertising revenue—especially long-term contracts and opportunities to build positive feedback collaborations—increased.
Is Controversial Traffic and Story Traffic Effective for the Crypto Circle?
Simple conclusion: Sometimes yes, but most of the time no. After two years of practice, I am convinced that in a track centered on strength and judgment, whether the content itself meets the real needs of the track always takes precedence over any traffic tricks. Having content, combined with traffic tricks, is the icing on the cake. Without content, forcibly applying traffic tricks will only backfire. However, emotional value is indeed value, but emotional value cannot effectively become a tool for monetization in the crypto circle.
In the early days, I relied on the pornographic appeal of the name "Winnie the Doll" to generate massive traffic without account restrictions, using that to run ads and accumulate capital. After a period of deeper involvement, I still had to constantly adjust strategies to improve efficiency. If someone consistently posts on Twitter but only depends on ads as their crypto income source, with traffic but no valuable content output to create positive feedback, from that person's perspective, it might be acceptable if they're not investing time. But once they go full-time, they're essentially trading time for a text-based salary that lacks compounding—barely above average pay. I gave this state a name: the底层黑奴 (bottom-tier slave) of the ad distribution agency. In an industry where you could seize opportunities and take off, using the most precious time for the least certain income is truly somewhat sad. This is my take on those who teach others to be KOLs without adding the caveat, "How much you earn depends entirely on your cognition."
Back to traffic itself. Information asymmetry and traffic are indeed necessary conditions for wealth transfer, but they have completely different effects on different forms of wealth transfer. Sometimes traffic even stands in opposition to wealth effects. Additionally, traffic that doesn't match the track is almost always irrelevant and ineffective—we can skip that entirely. Focusing only on track-matched traffic, I once wrote an article about the impossible triangle of crypto traffic. From a retail investor's perspective, you have to step into countless pitfalls to understand it. Many opportunities are one-bite-less-for-everyone deals; if you haven't even finished your own share, how could you actively tell others? By the time an opportunity is pushed to you via your feed, what kind of opportunity could it be? Some people even browse crypto Twitter not for useful information, but for gossip, arguments, or eye candy—so what's the point of spending time in crypto? That's why filtering information sources is so crucial. Retail investors are often compared to the tail end of the crypto chain. If you constantly stand at the exhaust outlet of the traffic tail and look in, you'll indeed prove the saying: "Even eating shit, you can't get it while it's hot."
Postscript:
This battle around recommendation algorithms will likely continue for me, but I still have a choice: to get lost in the energy drain of blindly chasing traffic, or to master my own time and goals. Ultimately, whether it's a loophole dividend, a traffic dividend, or a cycle dividend, they are always temporary. The P站 bug lasted only three months, the summer of porn tweets less than half a year, the NFT feast barely a year, and the bull market cycle just three years. I never expected to arbitrage permanently and solve everything once and for all. Anti-fragility isn't about adding one more track or running one profitable path. But as a qualified speculator, I can do four things: quickly judge within the window → quickly bet → quickly harvest → quickly retreat. Writing this, an afternoon has passed. I took a sip of tea, looked at my bag of BTC that I didn't sell at the top, and fell into deep thought. Though I followed through on my earlier decision, maybe I should still use that picture as the cover for this article—to punish myself and stay clear-headed.