No single signal is proof of inauthenticity
Coordinated inauthentic behavior is designed to look organic. Single-metric approaches — bot detection based on account age alone, or velocity alerts without network analysis — produce high false-positive rates that erode trust in the output.
Rolli IQ uses a multi-signal ensemble approach. The authenticity score is a weighted composite across four signal families: velocity & timing, account behavior, network structure, and content & language. Only when signals converge across multiple families does the score trend toward coordinated.
The score is explicitly probabilistic. A score of 24 does not mean we have proven coordination — it means our models assign 76% confidence that the amplification is not organic. Rolli IQ outputs are intelligence aids for human decision-makers, not automated verdicts.
Signal families
Each narrative is evaluated across all four families. Convergence across families drives a lower authenticity score.
Velocity & Timing
- —Post burst rate: how many accounts post within narrow time windows
- —Cross-platform synchrony: whether spikes appear simultaneously across platforms
- —Engagement-to-follower ratio anomalies (activity disproportionate to account size)
- —Timestamp clustering that indicates automated scheduling or bot coordination
Account Behavior
- —Account age vs. activity level (new accounts with high post volume)
- —Profile completion patterns associated with inauthentic actors
- —Historical activity distribution (dormant → sudden high activity)
- —Content repetition patterns across accounts (near-identical language or phrasing)
Network Structure
- —Coordinated cluster detection: accounts that amplify each other abnormally
- —Hub-and-spoke amplification patterns (central coordinator → satellite accounts)
- —Cross-platform account correlation (same actor appearing across platforms)
- —Follow/follower network analysis for artificial account inflation
Content & Language
- —Semantic similarity scoring across accounts in a narrative cluster
- —Template detection: near-identical posts with minor surface variation
- —Language anomalies inconsistent with organic discourse (translated content, unusual phrasing)
- —Hashtag coordination: simultaneous adoption of identical tags across accounts
Approximate signal weights in IQ score
Why 94.2% accuracy matters for your team's credibility
The score ranges from 0 (highly coordinated) to 100 (highly organic).
See the score on a real narrative — in 8 minutes.
Free trial, no credit card, no demo call required.
What Rolli IQ is — and isn't
The pipeline
Five deterministic stages — every narrative runs the same pipeline, every time.
Ingest
Continuous collection across 8 platforms. Posts, replies, reshares, and engagement events are normalized into a unified event schema. Deduplication applied at ingest.
Enrich
Every signal is enriched with topic extraction, entity tagging, emotion classification (6 emotions × platform × country), and narrative velocity metrics.
Score
The multi-signal authenticity model runs across velocity, account, network, and content families. A composite authenticity_score (0–1) is assigned to each narrative cluster.
Structure
Scored signals are organized into narrative clusters, timeline views, and key voice rankings — surfaced in the Rolli IQ dashboard or delivered via REST / MCP stream.
Brief
Rolli IQ Agents synthesize scored data into plain-English intelligence briefs with recommendations — optimized for stakeholder communication and executive decision support.
Where your narratives live, we watch
Rolli IQ monitors every major platform where coordinated narratives spread — social, video, news, and forums — giving you a single, unified view of the information environment your brand operates in.
More platforms
Continuously expanding
Additional platforms are actively evaluated. We add new sources when data quality and coverage density meet our signal reliability threshold.
News source network: In addition to social platforms, Rolli IQ monitors a curated network of 120,000+ news and media sources. This layer provides narrative origin context — identifying when social activity is amplifying news-driven narratives versus when social activity is preceding news coverage, which is a key indicator of coordinated injection.
How the authenticity model is calibrated
Authenticity scoring is calibrated against a ground-truth dataset of 8.4 million known authentic and inauthentic accounts, assembled from platform enforcement disclosures (Meta, Twitter/X, Google), academic CIB datasets, and Rolli's own documented case history. Updated quarterly to reflect evolving tactics.
The four signal families — velocity & timing, account behavior, network structure, and content & language — each contribute weighted inputs dynamically adjusted based on platform context and narrative type.
The multi-signal ensemble approach means no single signal can produce a false positive in isolation — convergence across families is required to drive a low authenticity score, eliminating noise-driven alerts.
Validation
Peer-reviewable results your leadership can defend
Rolli IQ's detection methodology has been evaluated against independent academic datasets and reviewed by researchers at leading institutions studying the information environment.
In Rolli IQ internal validation, Rolli's detection model achieved 94.2% precision on coordinated inauthentic behavior detection — meaning 94.2% of clusters flagged as coordinated were confirmed as coordinated by ground-truth labels.
The same evaluation achieved 91.7% recall — meaning Rolli's model identified 91.7% of all confirmed coordinated clusters in the dataset, with 8.3% missed (false negatives). This tradeoff is calibrated to minimize false positives at the cost of some false negatives.
External review
Rolli's methodology is aligned with frameworks used by leading research institutions studying coordinated inauthentic behavior. We welcome scrutiny from the research community — contact research@rolli.ai to discuss methodology or request documentation.
Transparency
Honest documentation of what Rolli IQ cannot do is as important as documenting what it can. We publish our limitations publicly because responsible deployment requires users to understand the boundaries of any analytical tool.
Public content only
Rolli IQ monitors publicly accessible social media content only. We do not access private messages, closed groups, direct messages, or any non-public content. This is a deliberate design constraint — not a technical limitation. The intelligence we provide is fully reproducible from public sources.
Language coverage
Our models are optimized for English-language content. Spanish and French are in beta, with accuracy levels below our English-language benchmarks. Arabic, Mandarin, Russian, and German are on the development roadmap. For non-English campaigns, we recommend treating scores as directional indicators and applying human review.
Probabilistic, not deterministic
Authenticity scoring is probabilistic — it estimates the probability that observed behavior is organic, not coordinated. Scores in the 30–70 range represent genuine ambiguity where behavioral evidence is mixed. Human expert review is strongly recommended before making high-stakes decisions based on scores in this range.
Novel coordination tactics
The calibration dataset and model are updated quarterly, but sophisticated actors continuously evolve their tactics. Novel coordination techniques that differ significantly from documented patterns may initially receive higher authenticity scores until the pattern is characterized and the model is updated. Rolli's human analyst team monitors for emerging tactics as a first-line detection layer.
Platform access constraints
Signal coverage depends on what data is made available by each platform through official APIs and public-facing access. Changes to platform API policies can affect signal availability. In the event of a platform access change that materially affects coverage quality, we notify customers and update our public methodology documentation.
Detection methodology
How Rolli Detects Coordinated Inauthentic Behavior
A rigorous, multi-stage detection process runs on every narrative Rolli IQ monitors — from first signal to delivered intelligence brief. Learn more about our influence operation detection approach.
Signal Ingestion
ContinuousActivity across monitored platforms is ingested in near real-time. Every incoming signal — posts, engagement events, account interactions — is normalized into a unified behavioral schema before entering the detection pipeline.
Account Profiling
ContinuousEach account accumulates a multi-dimensional behavioral fingerprint updated continuously as new activity arrives. Accounts whose profiles diverge significantly from organic baseline patterns are surfaced for downstream scoring.
Authenticity Scoring
Continuously updatedEvery monitored account carries a live authenticity score derived from its behavioral profile. Scores are banded into analyst tiers — from clearly organic activity to high-probability inauthentic amplification — with a human review layer before any escalation.
Narrative Clustering
With each signal batchPosts are continuously grouped into narrative clusters based on semantic and structural similarity. Rolli identifies when nominally independent accounts are amplifying the same message with suspiciously aligned vocabulary, framing, and timing.
Velocity Measurement
Baseline-normalizedEach narrative cluster is tracked against its own historical baseline. Abnormal acceleration — the sharp velocity spikes that distinguish injected amplification from organic viral growth — triggers deeper investigation regardless of absolute volume.
Coordination Detection
ContinuousWhen multiple accounts amplify the same narrative within statistically tight time windows, Rolli computes a coordination confidence score. The model is calibrated to distinguish genuine audience resonance from manufactured amplification networks.
Origin Attribution
Post-detectionOnce a coordination pattern is confirmed, Rolli traces the earliest amplification nodes to establish seeding pathways. This temporal map gives analysts a documented evidence chain and an investigative starting point.
Brief Generation
On-demand or scheduledA structured intelligence brief captures the full detection picture: narrative summary, coordination evidence, confidence assessment, and recommended next steps. Formatted for direct stakeholder delivery or integration into existing workflows.