Pay Per Post, and Your Community Becomes a Vendor

Does Paying for Posts Actually Wake Up a Community?

If you’ve ever run a community or a service, this meeting scene will feel familiar. Engagement starts to sag, and sooner or later someone floats the same idea: “What if we paid points per post? Wouldn’t that get things moving again?” It sounds intuitively right. People respond to rewards, after all.

But the analysis of an experiment that actually tested this idea—at a scale of 570,000 people, with real money on the line—was just presented at an international conference. And, almost at the same moment, one of the world’s largest “pay-for-posts” experiments was forced to shut down in the market.

Looking at both cases together tells you exactly what kind of behavior rewards actually increase. Pay people for writing, and the number of posts goes up—but quality doesn’t. The community starts behaving less like a community and more like a marketplace commissioning content. What determined success or failure wasn’t the size of the reward, but whether that reward could be converted into cash outside the community—in other words, its liquidity, or convertibility to cash.

We Paid 570,000 People to Write. Here’s What Happened.

Farcaster is a blockchain-based social network. What makes this platform valuable to researchers is something specific: multiple, differently-natured rewards coexist within a single service. There’s the official stablecoin USDC reward paid by the platform itself, third-party volatility-token rewards like DEGEN and MOXIE, and direct peer-to-peer tips between users. All within one service — which lets you compare how different kinds of rewards shape user behavior.

An international team of researchers analyzed usage data from this platform and presented their findings at ACM SIGMETRICS 2026, the leading conference in computer systems measurement. The dataset covered 574,829 users with linked wallets — 64.25% of all users on the platform. As someone who teaches data analysis, if I had to name this study’s single greatest virtue, it’s that it aimed at causation, not correlation. Rather than settling for “people who got rewarded wrote more,” the researchers used difference-in-differences¹ to compare changes before and after reward introduction against a control group and estimate actual effects.

The results boil down to three lines. First, most token rewards increased the volume of posts and replies. Second, they did not improve quality — and under some conditions, quality actually fell. Third, users who repeatedly received algorithmic rewards showed a cumulative pattern of learning behavior that games the reward system, not behavior that produces good writing. There was also a finding that competition to gain followers intensified, even as the act of following others didn’t increase at all. In other words, behavior tilted toward broadcasting one’s own posts to more people, rather than toward the kind of back-and-forth conversation where people actually read and reply to each other.

But the part of this paper I spent the longest time staring at is something else. Farcaster’s official USDC reward — the predictable, cash-like payout — had no statistically significant effect on the volume of posts and replies. What actually moved behavior was the volatility-token side, where prices swing. In other words, it wasn’t the reward with a fixed amount that increased writing — it was the reward that came with the hope of hitting it big. And indeed, the Gini coefficient of reward distribution ranged from 0.72 to 0.94: a structure where a tiny minority captured most of the payout.

In fairness, the paper isn’t all bleak. Person-to-person tips flowed 1.3–2x more often to recipients outside the tipper’s follow network, which actually eased the tendency toward closed, clique-like circles. Still, today’s focus is on the other side — the money for words that algorithms hand out, designed by whoever runs the community.

The Same Experiment, Repeated Four Times

Is this outcome just a quirk of the crypto world? Trace the lineage back, and you’ll find the same experiment repeated at least four times.

The first is Steemit. Launched in 2016, this blockchain-based social network paid authors in STEEM coins—cashable on exchanges—whenever their posts got upvoted. It has deep ties to Korea, too: according to the community’s own 2017 analysis, Korean was the platform’s second-largest language after English. Most of you probably remember what happened next. Controversy over self-voting and reciprocal vote-trading rings—modeled on pumashi, Korea’s traditional custom of mutual labor exchange—dragged on, and when the coin price collapsed, the community sank with it. In 2020, the founder sold the platform, the community split, and the experiment’s momentum effectively died.

The second is Quora. In 2018, it launched a Partner Program that shared ad revenue on “questions”—and questions were exactly what got mass-produced. A question factory churned out content chasing pageviews. The English-language program shut down in September 2022, and the remaining language versions followed in March 2023.

The third is X. Starting in 2023, it began sharing revenue tied to engagement metrics, and the controversy over low-quality posts and comment spam—which I’ll get to later—has to be read alongside this incentive structure.

And the fourth is the experiment that just ended: Kaito’s Yaps. Launched in December 2024, Yaps measured influence and awarded points whenever users posted about crypto projects on X. According to reports citing data from the analytics platform Dune, monthly active participants topped 200,000. What’s interesting is that Kaito fought farming² fairly seriously. It touted an algorithm that judged content quality and authenticity rather than raw activity volume, and kept tightening eligibility and ranking criteria. Even so, both Korean and international coverage pointed to the same problem: accounts kept climbing the rankings with sensational images and repetitive comments.

On January 15 this year, the way the service operated abruptly changed. Nikita Bier, X’s head of product, announced that the platform would cut off API access for apps that pay out rewards for posts. AI-churned low-quality content and a spike in comment spam driven by InfoFi³ were cited as the reasons. Kaito announced its own pivot the same day: after 13 months, it shut down Yaps and the leaderboard, converting instead into “Kaito Studio,” a curated marketplace where brands directly select and contract vetted creators. Reports said the token price dropped more than 20% right after the announcement. Cookie DAO, another InfoFi service, announced the end of its own reward program just 10 minutes after X’s statement, and word came that X had banned the Yaps participant community—roughly 157,000 people. Notably, foreign coverage pointed out that Korean users made up a particularly large share of that group.

Two things are worth flagging here. One is that AI was the decisive factor that killed this model. Once the cost of producing a single post is effectively zero, per-piece rewards simply stop making mathematical sense. One Silicon Valley investor summed up the situation this way: “the median crypto-Twitter post became Kaito slop.” The other is the view from the opposite side. Since X itself runs its own pay-for-engagement revenue share while singling out InfoFi for elimination, some critics argue the real issue was never the reward model itself but who gets to collect the fee. That criticism isn’t wrong in principle—the quality problems plaguing X’s own timeline stem from the same structure, after all.

But this is exactly where the real question emerges. If every reward scheme with money on the line ends this way, what made the ones that actually work turn out differently?

The Fork in the Road Isn’t the Amount — It’s Whether You Can Cash It Out

Think about it — plenty of communities already run on rewards. Naver’s Knowledge iN gives users naegong (merit points that signal expertise and contribution), Stack Overflow hands out reputation scores, and Danggeun (Korea’s neighborhood marketplace app) tracks a “manner temperature.” Gamification — designing tiers and badges — has long been a standard tool of community management. So why does this kind of reward keep these platforms alive while paying for posts kills them?

The difference lies in whether the reward can be cashed out beyond the community. Naegong, reputation, and manner temperature are all scores that reflect contribution and trust within that specific community. No matter how much you farm them, they don’t convert into an hourly wage, so the whole cost-benefit calculus of farming never even gets off the ground. What matters more is the direction of the incentive. The value of that status only holds up if the community itself stays healthy. If I wreck the place, my reward disappears along with it. Cash and tokens, by contrast, leave the building. Even if the community collapses, the money I pocketed stays in my account. At that point, treating the community as disposable becomes a rational strategy for the individual. This is exactly where a community starts to function like a market that simply procures content.

When money enters the picture, it doesn’t just change the math — it changes how we feel. Behavioral economics calls this the motivation crowding effect⁴. The most famous experimental stage for it was a daycare center in Haifa, Israel. When the center started fining parents who picked up their kids late, tardiness didn’t drop — it rose. The social norm of feeling guilty had been replaced by a price tag: a service you could simply pay for. Even after the fine was scrapped, lateness never returned to its original level. Research on blood donation points in the same direction. In 1970, sociologist Richard Titmuss argued that paying for blood donations could actually reduce them, and a 2008 field experiment in Sweden partly confirmed this. But the real twist in that experiment came next. When donors were offered payment with the option of donating that money to charity, the drop in donations disappeared entirely. Once people stopped pocketing the money for themselves, the original norm came back to life. What this shows so clearly is that what determines the outcome isn’t whether money is offered at all, but whether an individual keeps it for themselves.

Korea carries a heavy trace of this same experiment. Reviews. As photo-review reward points and product-trial programs (companies handing out free products in exchange for reviews) became commonplace, cookie-cutter reviews flooded the internet, distrust toward sponsored posts built up, and it all erupted in the 2020 “hidden advertising” scandal. That same year, the Korea Fair Trade Commission tightened its guidelines on ad disclosure. What I find especially telling is a word born out of this whole process: naedon-naesan (“bought with my own money,” a phrase Koreans use to mark a review as unpaid and genuine). Think about what that phrase implies: trust now requires actively proving you weren’t compensated. Once reviews came to be treated as ad copy written to order, people even had to invent a separate term just to mark out the reviews that weren’t paid for.

So what if we just shrink the amount drastically? Unfortunately, a middling amount of money is the worst option of all. The title of another paper by Uri Gneezy and Aldo Rustichini — the economists behind the Haifa daycare study — makes the conclusion for us: “Pay Enough or Don’t Pay at All.” A small reward crowds out intrinsic motivation without being large enough to fill the void it leaves behind.

Oswarld’s Lens

I’ve actually lived through this one myself. Let me tell you about it based on my time running Notion’s early Korean community. Back then, the community grew explosively without any monetary compensation. What got people to stay up all night building templates and writing how-to guides wasn’t money. It was the chance to present on stage at meetups, the status of being an “ambassador,” early access to new features and the home team, and the identity of being “someone who really knows Notion.” All of these are rewards that can’t be converted into cash once you step outside the community. Imagine if I’d put ₩5,000 (~$3.6) on the table for every post back then. Sharing a template would have turned into piecework paid per submission, and I would have spent my time reviewing submitted posts instead of growing the community.

That’s why, when I advise on reward design, I stick to two principles. First, give the majority of members rewards that can’t be cashed out outside the community — status, a stage to present on, early access to the team and new features, and mechanisms that make contributions visible to others. Second, don’t spread money across the whole community; instead, pay a vetted handful of professional creators under contract. This is exactly the direction Kaito moved toward when it shut down Yaps after just 13 months. The studio model — where a brand directly selects and contracts with creators who’ve passed a screening process — is precisely this structure. From a GTM standpoint, the moment you start paying a community per post, it stops being a community and becomes a cheap advertising channel. And advertising channels inevitably get dragged into a race to the bottom on price.

Closing

Looking at the usage records of roughly 570,000 users and these four cases, one pattern holds: putting a reward on writing increases the volume of content, but it doesn’t raise quality or trust along with it. When a reward can be cashed out externally, the incentive to farm — repeatedly meeting the bare conditions — grows stronger. Points and reputation that only work inside a community, by contrast, retain their value only as long as that community itself holds together.

Try this exercise this week. Take whatever reward system you’re running and split it into two columns: “convertible outside” and “valid only inside.” If the rewards in the first column are tied to content count or view counts, farming — gaming the conditions rather than genuinely engaging — is likely to show up.

I’d love to hear about your own experience on the ground, Reader. Whether you’ve offered points or payment and watched farming emerge, or moved people with non-monetary rewards like tiers or badges, tell me in the comments. If enough cases come in, I’ll turn them into a “Korean Reward Design Case Book” for a future issue.

---

💬 Share your reward-design wins and failures in the comments — I’ll fold them into the next issue. 📨 If a colleague is wrestling with community or reward policy, pass this along to them.

---

References & Further Reading

Primary sources

• Yang, W. et al., “Beyond Single-Tokenomics: How Farcaster’s Pluralistic Incentives Reshape Social Networking”, Proceedings of the ACM on Measurement and Analysis of Computing Systems (ACM SIGMETRICS 2026), 2025. Link ··· This is the backbone of today’s piece. The causal-analysis table mapping which rewards drove which behaviors is the highlight.
• The Block, “X users celebrate crackdown on ‘plague’ of AI-led reply spam as InfoFi platforms seek alternatives”, January 2026. Link ··· This lays out the industry’s split reaction when InfoFi platforms got purged. You can also find the counterargument aimed at X’s own revenue-sharing model here.
• Gneezy, U. & Rustichini, A., “A Fine Is a Price”, The Journal of Legal Studies, 2000. Link ··· This is the original paper behind the Haifa daycare experiment. It’s short enough to read in full.
• Mellström, C. & Johannesson, M., “Crowding Out in Blood Donation: Was Titmuss Right?”, Journal of the European Economic Association, 2008. Link ··· This is the source for the twist that “an opt-out donation option eliminates the crowding-out effect.”

Background

• Titmuss, R., The Gift Relationship: From Human Blood to Social Policy, Allen & Unwin, 1970. ··· The classic that kicked off the whole reward-versus-norms debate.
• CoinGecko, “What Is Kaito? 2026 Guide to Studio, Markets & KAITO Token”, 2026. Link ··· This traces Kaito’s whole arc, from the end of Yaps to its pivot into Studio, in one read.
• TechCrunch, “Quora shutting down English version of Partner Program”, August 2022. Link ··· A record of how an experiment in paying people for questions came to an end.
• Blockmedia, “Google Blocks Apps, X Blocks Yapping⋯Big Tech’s Blade Rattles the Digital Asset Industry”, January 2026. Link ··· A Korean-press account summarizing the situation before and after the Yapping ban.

Worth reading alongside this issue

• Issue 167: The Earnings Breakdown of a Creator With 30,000 Followers
• Issue 170: Six Claims of a Perfect AI Score, Only Two Officially Graded

---

The author is Oswarld (Kwangseob Ahn). Current roles: Adjunct Professor at Sejong University, Strategy Consultant at INLEVEL9. Career, research, books, and recent work are kept current on the About page. Latest · July 2026: HEMA-2: A Consolidation-Aware Tri-Memory Architecture with Multi-Channel Scheduling for Lifelong Conversational AI.

📝 Glossary

Footnotes

1. Difference-in-Differences: A statistical technique that estimates the pure effect of an intervention by comparing the before-and-after change in a treated group against that of an untreated group. ↩

2. Farming: Gaming the loopholes in a reward system to repeatedly satisfy the conditions for a payout, regardless of whether the content has any real value. The term comes from repeatedly hunting for items in video games. ↩

3. InfoFi (Information Finance): A crypto-industry model that measures information and attention like financial assets and rewards them with tokens. Services that pay you points or coins for writing posts fall into this category. ↩

4. Motivation crowding-out: A phenomenon in which an external reward displaces intrinsic motivation (enjoyment, a sense of duty, social norms), so that behavior declines or degrades in quality even though a reward has been introduced. ↩