Platform

How Refute detects coordinated manipulation

Two complementary layers: behavioural clustering of account networks and linguistic signal analysis of content. Together they produce a single, actionable authenticity score in under five seconds.

Book a Demo
What gets scored

From a URL to a verdict

Submit any story URL or account handle. Refute returns an authenticity score, a confidence band, a list of flagged signals, and an account network map showing who is amplifying the content.

Story URL scoring

Paste any article URL and Refute maps the account network distributing it, checks for coordinated amplification, and scores the content signals in the story itself.

Account handle scoring

Submit an account handle and Refute analyses posting cadence, coordinated timing patterns, network overlap with known inauthentic clusters, and historical behaviour.

Authenticity score 0-100

Each score comes with a confidence band and a breakdown of which signals contributed most. Low scores include an evidence summary ready to share with your compliance or editorial board.

Network map

Visualise the cluster of accounts amplifying a story, flagged nodes shown in red, confirmed authentic accounts in teal. Export as SVG for presentations or compliance filings.

Detection pipeline

The three layers of analysis

Every Refute score is the product of three independent analysis layers, each examining a different dimension of authenticity.

Account behaviour

Posting cadence anomalies, coordinated timing windows where dozens of accounts post within seconds of each other, and shared device signals that indicate centralised operation. These patterns distinguish organic engagement from manufactured amplification.

Content signals

Linguistic fingerprinting detects template similarity across posts from different accounts, unusual phrasing clusters, and recycled narrative structures. Content from coordinated operations consistently shows lower linguistic variation than organic commentary.

Network structure

Graph clustering of accounts amplifying a story reveals inauthentic networks: tightly-connected subgraphs with low follower diversity, hub-spoke amplification patterns, and temporal synchronisation that no organic group exhibits.

newsdailywire.net
Consumer group calls for product recall over safety concerns
34
Coordinated behaviour detected
Network overlap: 78% Amplification spike: detected First seen: Jun 18, 2026
Integrations

Fits your existing workflow

Scored results go wherever your team already works. No workflow changes required.

REST API
Scored results in your own tools via a documented JSON API. Rate limits scale with your plan.
Slack alerts
Threshold-based notifications direct to your brand safety or editorial Slack channel. Configurable per score band.
CSV export
Bulk scoring output for research teams and compliance reporting. All signal data included in export.
Webhooks
Real-time pipeline integration. Fire scored events to any endpoint your monitoring stack listens on.
View API Docs Book a Demo