Insights · Technology breakthroughs · 4 minute read

X, Twitter, and the Community Notes Experiment

The goal of an on-platform context feature is to give readers relevant evidence while it still matters. X’s Community Notes makes that ambition testable through a public contribution and rating system.

By Vitaver Staffing ·

A public way to add context

In October 2022, Elon Musk’s acquisition brought Twitter under new ownership. The subsequent X redesign raised many questions about the company’s direction. Community Notes offers a specific piece of the platform to examine. Twitter had launched it as Birdwatch in January 2021, before the acquisition. The project continued under Musk’s ownership as Community Notes. Its public ranking changelog documents work in late 2022 to support global expansion. Its history makes the case useful to product leaders. An acquired capability can become a prominent product decision, and its designers, early contributors and later operators all have a part in the result.

The core mechanism is public. Contributors propose context for posts. X describes a system that displays a note after enough contributors with different rating patterns judge it helpful. Those patterns come from how contributors rate notes, so they describe behavior in the system. X also publishes data and scoring code, giving researchers a basis to examine how notes are selected. Published scoring rules and public feedback make it possible to ask sharper questions about the product.

Decide what evidence counts

A good note can point directly to a date, document, original image or missing condition that changes the meaning of a claim. It arrives while people are still making decisions based on that claim. A rating model cannot make a calendar move faster or ensure that people with the relevant expertise evaluate every post. For a fast-moving story, an accurate note a day later can clarify the record while missing much of the initial audience.

Start an evaluation with a sample of posts for which context would materially change how a reader understands the claim. Agree on that test before checking which posts got a note. Include posts with no note and posts with several proposals. Measure time until the first useful note is visible, how often it cites a source a reader can check, and what happens when the source or claim changes. For a small or specialized subject, find out whether enough independent reviewers can participate. Report sample limits and the range of outcomes, including cases with conflicting evidence.

Design for the reader and the reviewer

A product owner building a similar feature can start with one type of claim where citations are practical. Give contributors a way to state the missing fact and attach the original evidence. Assign an owner to monitor abuse and make the process for correction, updating and withdrawal visible. Have an independent evaluator review sample cases at a prearranged date and publish which categories the feature serves well and which it misses. Preserve the ability to examine failures when reviewers disagree or a note is late.

Readers have a simpler test. If a note gives them a source, open it and check whether it supports the exact correction. If the claim depends on missing information, record what remains uncertain before acting on it. The product’s strongest idea is that context can appear where a decision is being made, with clues the reader can verify. Its value depends on the evidence and the time at which the reader gets it.

← All Insights