The Drifter bot experiment (Chen et al., Nature Communications, 2021) deployed 15 neutral social bots on Twitter to probe political bias in social media ecosystems. The bots were programmed with identical, politically neutral behavior models — the only difference was the initial news source they followed (Left, Center-Left, Center, Center-Right, or Right).
Key findings#
No platform bias#
No strong or consistent evidence of political bias in Twitter’s news feed algorithm. The content users saw closely matched what their friends produced.
Strong ecosystem bias#
Despite platform neutrality, early choices about who to follow created dramatically different experiences:
| Outcome | Conservative | Liberal |
|---|---|---|
| Echo chamber density | Denser, more homogeneous | Less dense, more moderate |
| Follower growth | Higher | Moderate |
| Bot exposure | Follow more automated accounts | Moderate |
| Low-credibility content | ~15% of feed links | Minimal |
| Political drift | Stayed on the right | Shifted toward center |
Mechanism#
Neutral algorithms don’t guarantee neutral outcomes. Right-leaning accounts attracted more followers, ended up in denser communities with more automated accounts, and were exposed to significantly more low-credibility content — even though the bots themselves had no political bias.
Significance#
- Demonstrates that ecosystem bias can emerge without platform bias due to user interactions and network effects
- Challenges claims that social media platforms systematically suppress conservative speech
- Shows that early social connections have outsized influence on long-term information exposure
- Provides a methodology (neutral drifters) for studying platform bias generally
Source#
Chen, W., Pacheco, D., Yang, KC. et al. Neutral bots probe political bias on social media. Nat Commun 12, 5580 (2021). https://doi.org/10.1038/s41467-021-25738-6
