A Reddit field experiment asked whether small awards can shape online behavior and trust

How the award experiment was set up on Reddit

The Reddit test used two automated accounts that looked different on the surface. One was presented as a bot. The other was styled like an ordinary person. Both had plain profiles with no age, gender, or personal history. The bot account said it existed to give awards. The human-styled account described itself as a longtime fan and lurker who wanted to send small rewards. In practice, both accounts ran by the same script. Only their wording changed.

Each award was the Diamonds are Forever badge, worth about one dollar. It came with a short private message. Four messages were used. One praised clear reasoning or logic. Another said the recipient had won a random lottery. A third highlighted emotional tone or empathy. A fourth pointed to moral intuition or integrity. A fifth group got no award at all. This made nine conditions: four human-sent rationales, four bot-sent rationales, and one control. The total sample was 2,442 observations, with about 263 to 275 in each condition.

The awards were sent into twelve general-interest subreddits. Practical subreddits covered life tips, personal finance, frugal living, career advice, and travel hacks. Knowledge communities included explain like I’m five, today I learned, science, books, uplifting news, and you should know. Politics, mental health, self-help, special-interest groups, and media-heavy content were left out. The posts and comments had to be recent and low-attention. They needed at least 300 characters and no downvotes. The writer had to be relatively new. Their account was older than 30 days. It had between 10 and 499 prior comments. It also had between 0 and 99 prior posts. A keyword list first screened for sensitive material. Then GPT-4 checked whether the text touched mental health, trauma, personal loss, serious illness, or other delicate issues. The model saw only the text, not the username.

The public badge was visible to everyone on Reddit. The reason message and sender identity were not. Recipients could see who had awarded them and why. Other users could only see that an award existed. Because Reddit blocked third-party apps from giving awards automatically, the setup used a browser-control tool called Selenium. It filled in fields and clicked buttons like a person would. Selections happened every 30 minutes. After each award, performance was recorded at the moment of sending and then every two hours for 48 hours. Recipient activity and karma were tracked every 12 hours for 168 hours. A week later, full user histories were pulled to compare behavior before and after.

Direct replies in the first twelve hours after an award

Within the first twelve hours, a small but clear pattern appeared. Recipients were more likely to answer privately when the award came from an account that looked human. They did this less often after a bot sent the same kind of award. The gap was large enough to be trusted. That means it probably reflected real behavior, not chance.

The bot replies were so uncommon that no reason could be compared across them. With the human account, the wording mattered. Messages praising clear reasoning drew private answers. Messages praising moral integrity also drew private answers. A message saying the award came from a random lottery drew fewer replies. Emotional praise fell between those patterns, but its difference was not strong enough to count as reliable.

This points to one simple idea. People respond when they think another person is speaking to them directly. A public badge can be ignored. A private note with a reason can start a conversation. The effect is stronger when the reason feels like real recognition. It is weaker when the award looks like a coin toss. The award itself was small. The message carried the weight. The first reply often depends less on the reward than on being seen.

Changes in posting, commenting, and tone over the following days

The follow-up window ran for seven days after an award was given. Activity was checked every 12 hours. After one week, full histories were pulled so that before and after patterns could be compared. This mattered because the chosen users had just posted or commented. A person who has been active often slows down next. That burst to lull pattern can make any drop look like harm when it is not.

Across all award types, there was no reliable rise in later posting or commenting. Total words did not climb either. In plain terms, a public badge and private message did not turn recipients into more active writers over the following days. The expected effect of a small reward was missing.

One clear exception pointed downward. Recipients who got a random lottery award from the bot account made fewer later contributions than users with no award. They also wrote less text. Their output fell below several other award groups too. This pattern suggests that an unexplained, automated giveaway can feel empty or even irritating. It may reduce motivation rather than spark it.

Writing style did not shift in line with the stated reason for the award. Logic praise did not reliably produce more logical wording. Empathy praise did not reliably add emotional language. Moral praise did not reliably add moral vocabulary. The only reliable tone change was negative. Bot random lottery recipients used a less positive tone in their next 12 hours of posts and comments than users with no award. That brief dip is small, but it shows that the message can color how people write soon after.

Audience votes and replies to later contributions by recipients

When a recipient later posted or commented, the main question was whether other Reddit users treated that person as more credible. The answer was mostly no. Most award types did not change how much attention the recipient’s future contributions received. Votes stayed close to what would be expected without an award. Replies were about the same length and number. This held for awards framed around logic, empathy, random luck, or moral praise. It also held when the award came from a bot account rather than a human styled account.

The one exception was moral praise sent through the human styled account. Recipients who got that message later received more upvotes on their posts and comments. They also drew longer replies than recipients who had received logic praise or random luck awards from the same kind of account. The difference is meaningful because it points to a social effect, not just a private feeling. Other users were willing to spend time responding. They seemed to treat the recipient as someone whose later words deserved attention.

The result should be read with care. It was one positive signal among many null results. A stricter statistical check that accounts for testing several award types at once made it disappear. That does not prove there is no effect. It means the evidence is thin. The moral praise finding suggests a possible channel: when an award says someone acted with integrity, bystanders may later give that person more weight. But the data do not show a reliable rule that awards boost future audience response.

What the Reddit award test leaves behind

The field test points to a narrow but real opening. A visible award can start a private conversation when it feels like recognition from another person. The human styled account drew more replies than the bot labeled one. Praise that named reasoning or moral integrity did better than a random lottery message. That suggests symbolic rewards work best when they carry meaning, not just value.

The bigger picture is less cheerful. Most awards failed to move behavior in a reliable way. Recipients did not post more, write longer, or adopt the praised style. One combination even backfired. Bot sent random lottery awards were followed by fewer contributions and shorter writing than no award at all. The tone of those recipients also turned less positive for a short time. This is an obstacle any platform should take seriously. Low effort recognition can feel like noise, and bad timing or poor framing can make it worse.

There are still opportunities. Human moral praise produced the clearest indirect benefit. Later posts by those recipients earned more votes and longer replies than several other groups. The effect was small and fragile under stricter testing, but it points to a possible path. Recognition may matter most when it is specific, credible, and tied to a value people respect.

The obstacles are structural as well as psychological. Reddit activity comes in bursts, so weak effects can disappear into normal variation. Platform changes around awards and notifications may have weakened their meaning. Automation also faces friction. API limits forced browser based workarounds, and two human styled accounts were later shadowbanned. Ethical limits matter too. The test used mild deception and lacked informed consent, which narrows what can be claimed.

The lesson is not that rewards are dead. It is that they are becoming harder to make meaningful. If users grow cynical about automated praise, likes, upvotes, and badges may lose power unless platforms make the source of recognition clearer. A bot that says it is a bot may preserve trust even when its gift does little else. The question is whether future systems can turn small symbolic gestures into durable signals without turning every interaction into a performance.