💻 computer science

A field experiment of social influence and behavioral contagion with bots on Reddit

A field experiment on Reddit reveals that while symbolic awards from both human and bot accounts fail to increase user activity or downstream impact—and may even reduce them when attributed to a random lottery—they do successfully encourage direct communication, suggesting online users possess some resilience to simple behavioral manipulation by artificial agents.

Original authors: Hiroki Oda, Kinga Makovi, Taha Yasseri, Milena Tsvetkova

Published 2026-07-02

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Hiroki Oda, Kinga Makovi, Taha Yasseri, Milena Tsvetkova

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine Reddit as a massive, bustling town square where people share stories, ask for advice, and chat. In this town, there are two types of residents: real humans and "bots" (computer programs pretending to be people). The researchers wanted to see what happens when these residents give out "digital gold stars" (awards) to others. Do these stars make people work harder, talk more, or feel better? And does it matter if the star comes from a human or a robot?

Here is the story of their experiment, broken down simply:

The Setup: The "Gold Star" Experiment

The researchers created two types of accounts on Reddit: one that looked like a normal human and one that clearly admitted, "I am a bot, beep boop." They then gave out digital awards to random users with four different reasons (or "rationales") attached to the gift:

The Logic Star: "I liked your clear, smart reasoning."
The Random Star: "You won a lottery! I'm just giving these out randomly."
The Heart Star: "I loved your emotional tone and empathy."
The Moral Star: "I admired your integrity and moral compass."

They wanted to see if receiving a star made the recipient:

Post more often (Volume).
Get more attention (Impact).
Write nicer or smarter things (Content).
Talk to the giver directly.

The Results: The Stars Didn't Spark a Fire

Surprisingly, the "gold stars" didn't work the way the researchers hoped.

1. The "Work Harder" Myth Didn't Happen
You might think, "If I get a prize, I'll want to do more!" But on Reddit, getting a star didn't make people post more or write longer comments. In fact, it didn't change their activity levels at all.

The Bot Backfire: There was one weird twist. When a bot gave a star saying, "You won a random lottery," the recipient actually posted less and wrote shorter comments afterward. It was as if the random gift from a robot made them feel a bit weird or discouraged, like they were being played with.

2. The "Quality" Myth Didn't Happen
Did the stars make people write nicer or more logical posts? Generally, no. The content of their writing stayed the same.

The One Exception: The only time the content changed was when a bot gave a random lottery award. In that case, the recipients' posts became slightly more negative or less cheerful.

3. The Only Real Win: Starting a Conversation
The one thing that did work was getting people to say "Thank you."

Humans vs. Robots: People were much more likely to send a private "thank you" message to a human who gave them a star than to a robot.
The Message Matters: If a human gave a star with a genuine compliment (like praising someone's logic or morals), the recipient was very likely to reply. But if a human said, "I just picked you randomly," the recipient was less interested in talking back.
The Robot Limit: Robots rarely got a reply at all, regardless of what they said.

The Big Picture: What This Means

The study suggests that online users are surprisingly tough to manipulate with simple tricks. Giving out a digital "gold star" isn't enough to make people work harder or change their behavior, especially if the star comes from a robot.

The "Robot" Problem: When a robot tries to act like a human by giving random rewards, it can actually backfire and make people less active.
The Human Connection: The only thing that really worked was using the award to start a real, private conversation between two humans.
The Takeaway: People on the internet are becoming more skeptical. They aren't easily swayed by simple, automated gestures. If a robot tries to pretend to be human to influence behavior, it might fail or even annoy people. The researchers conclude that we need to be very clear about who is a human and who is a bot, because people react differently to them, and transparency is key to keeping the online world trustworthy.

In short: Giving out digital awards didn't make people work harder or write better. It mostly just made them say "thanks" to real humans. When robots tried to give random awards, it actually made people pull back. The experiment shows that simple automated nudges might not be as powerful as we thought.

Technical Summary: A Field Experiment of Social Influence and Behavioral Contagion with Bots on Reddit

Problem Statement
As modern society evolves into a hybrid system of humans and intelligent machines, a critical question remains: how do artificial agents influence human behavior compared to human agents, and does this influence cascade through social networks? While prior research suggests humans may treat machines similarly to other humans (the CASA paradigm), other studies indicate distinct differences in perception, particularly regarding moral agency, trust, and emotional response. This study investigates whether social influence and behavioral contagion differ based on the source of the influence (human vs. bot) and the rationale behind the action (logical, random, emotional, or moral). Specifically, the authors examine if users are susceptible to positive behavioral stimuli from artificial agents and whether these stimuli trigger downstream effects in online communities.

Methodology
The researchers conducted a large-scale field experiment on Reddit, a platform with an endemic bot population and a culture of peer monitoring.

Experimental Design: The team created two accounts: one appearing as a typical human user and one as an apparent bot (adhering to Reddit's "Bottiquette" by including "bot" in the username). These accounts awarded a symbolic "Diamonds are Forever" award (worth $1.00) to posts or comments from relatively inexperienced users.
Treatments: The experiment utilized a 2 (Source: Human vs. Bot) × 4 (Rationale) factorial design, plus a control group. The four rationales were:
1. Logical (LOG): Praising clear reasoning and logic.
2. Random (RAN): Stating the award resulted from a random lottery draw.
3. Emotional (EMO): Praising emotional tone and empathy.
4. Moral (MOR): Praising moral intuition and integrity.
- Control: Contributions were selected based on the same criteria but received no intervention.
Sample Selection: The study targeted 12 general-interest subreddits, filtering for non-sensitive topics, specific engagement levels (1–5 comments, 2–6 upvotes for posts), and user inexperience (account age >30 days, 10–499 comments, 0–99 posts). The final dataset comprised 2,442 observations.
Data Collection & Analysis: Nine days post-treatment, the researchers retrieved the complete activity history of treated users. They measured:
- Direct Effects: Volume (number of contributions, text length), content (positive sentiment, lexical prominence of specific rationales), and direct communication (private thank-you messages).
- Indirect Effects (Impact): Votes, number of replies, and total text length of replies to the treated user's subsequent contributions.
- Statistical Approach: Due to the long-tailed distribution of Reddit activity, the authors analyzed within-individual changes ( $\Delta M = M_{after} - M_{before}$ ) using non-parametric Mann-Whitney U tests. They focused on contributions made within 12 hours of the award.

Key Results
The study yielded several counter-intuitive findings regarding the efficacy of symbolic awards in the current AI era:

No General Increase in Activity: Contrary to expectations that awards would boost productivity, the study found no statistically significant evidence that awards increased the volume of user activity (number of posts/comments) or total text length compared to the control group.
Backfire Effect of Bot-Lottery Awards: Awards from bots using a random/lottery rationale ( $B_{RAN}$ ) significantly reduced user activity. Recipients of these awards made fewer contributions and wrote shorter texts compared to the control group and recipients of awards with other rationales.
Direct Communication: Awards successfully initiated direct contact. Users were significantly more likely to send private thank-you messages to human accounts than to bots. Among human accounts, the random rationale elicited fewer replies than logical or moral rationales.
Content Changes: The only significant content shift was a decrease in positive tone for users receiving random lottery awards from bots.
Indirect Impact (Spillover): Awards generally did not increase the impact (votes/replies) of subsequent contributions. The sole exception was awards from human accounts with a moral rationale ( $H_{MOR}$ ), which garnered significantly more votes and longer replies compared to the control and other human treatments. However, this effect was sensitive to multiple comparison corrections.
Resilience to Manipulation: The results suggest that online users exhibit resilience to simple behavioral manipulation via symbolic awards, particularly when the source is a bot or the rationale is perceived as arbitrary.

Significance and Claims
The paper claims that the findings highlight a potential resilience of online users to simple behavioral manipulation by platform algorithms and artificial agents. The authors argue that while influence can cascade, the strength of this influence depends heavily on the perceived source and rationale.

Human vs. Machine Perception: The results support the hypothesis that humans judge machines differently; bots are less effective when employing emotional or moral rationales (outside their perceived competence) and can even be counterproductive when acting randomly.
Ethical Governance: The study underscores the necessity of transparently labeling automated agents. Since users react differently to perceived humans versus bots, and because bots can backfire or fail to engage, transparency is essential for ethical platform governance and maintaining trust.
Contextual Shift: The authors suggest that the lack of positive effects may stem from a shift in user cynicism due to aggressive platform monetization and the proliferation of AI, making users less responsive to low-effort peer interactions like symbolic awards. They caution that while simple schemes may fail, more sophisticated schemes simulating human conversation could still be effective, necessitating continued vigilance.

The paper concludes that while symbolic awards can kindle direct human-to-human communication, they do not currently serve as a reliable lever for increasing productivity or positive contagion in the Reddit ecosystem, and their use by bots carries the risk of reducing engagement.

The Setup: The "Gold Star" Experiment

The Results: The Stars Didn't Spark a Fire

The Big Picture: What This Means

Technical Summary: A Field Experiment of Social Influence and Behavioral Contagion with Bots on Reddit

More like this