Powered by XRAE

Media intelligence for the people who need it most.

The only platform that detects both persuasion techniques and addiction patterns in media content — across podcasts, YouTube, and news. Continuously. At scale.

The Engine

XRAE isn't a prompt.
It's a purpose-built detection model.

Anyone can write a prompt that says “find persuasion techniques.” That's not what this is. XRAE is a fine-tuned large language model — trained through thousands of expert-annotated examples, distilled from the most capable AI systems available, and validated against peer-reviewed behavioral science. It doesn't guess. It detects.

Trained, not prompted

XRAE was built through teacher-student distillation — the most capable frontier models annotated thousands of media segments, then that knowledge was distilled into a specialized detection model. This isn't prompt engineering. It's model engineering. The result is a system that has internalized the patterns, not one that follows instructions about them.

Local. Private. Ours.

XRAE runs on dedicated hardware. No cloud API calls. No third-party data sharing. No per-token costs. No rate limits. Your content never leaves the infrastructure. This means we can analyze at a scale and speed that API-dependent services simply cannot match — and the model can never be taken away, deprecated, or repriced by a vendor.

Every code has a citation

The 32-code taxonomy wasn't invented — it was compiled from decades of published research. Each addiction pattern maps to specific peer-reviewed work: Skinner's variable reinforcement schedules, Zeigarnik's cognitive load theory, Przybylski's FOMO research (250+ studies), Horton & Wohl's parasocial interaction theory (70 years of validation). When XRAE flags a pattern, it's invoking science — not opinion.

Multi-pass consensus

Every piece of content is analyzed multiple times at varying temperature settings. Only detections that survive consensus voting make it to your dashboard. This eliminates noise and false positives — when XRAE flags something, it means the pattern was strong enough to be detected consistently across independent passes. Zero false positives on clean, informational content.

There will be others who try to do this. They'll wrap a prompt around a general-purpose API and call it “media analysis.” Some already have. But there is a fundamental difference between asking a model to look for something and training a model to see it.

A prompted model is renting someone else's intelligence. When the API changes, your analysis changes. When the price goes up, your margins go down. When the model gets deprecated, you start over. You don't own anything.

And XRAE doesn't stand still. Every week, new training data flows back into the pipeline. Detection failures get corrected. Edge cases get labeled. The model retrains, re-evaluates, and improves — a compound learning loop that gets sharper with every episode it analyzes. The longer XRAE runs, the better it gets. That's not something you can replicate by updating a prompt.

XRAE is ours. The training data is ours. The taxonomy is ours. The hardware is ours. The model weights are ours. The improvement curve is ours. No dependency on any vendor's API, pricing, or product roadmap. That is what makes Prizm possible — and what makes it defensible.

32
Detection codes
0
Cloud dependencies
0
False positives on clean content
70+
Years of cited research

Content is the new liability.

After KGM v. Meta/Google, the question isn't whether content causes harm — a jury already decided that. The question is whether you can see it coming. Prizm gives you the data.

Litigation Teams

Build evidence for content liability cases with forensic-grade analysis of addiction patterns. Every finding traced to peer-reviewed psychology.

Media Buyers

Know what you're buying. See the manipulation profile of any show before you place an ad. Protect your brand from association with addictive content.

Researchers

Access the largest continuously-updated dataset of media influence techniques. 32-code taxonomy grounded in 70+ years of behavioral science.

32 techniques. 6 families. Every episode.

Two dimensions of detection that no other system offers: the techniques that influence what you think, and the patterns that keep you coming back for more.

Emotional

Fear manipulation and emotional exploitation

EM01–EM02

Faulty Logic

False reasoning, authority appeals, selective evidence

LF01–LF04

Loaded Language

Charged wording, minimization, euphemism

LG01–LG03

Trust Manipulation

Source positioning, identity construction, commitment pressure

SM01, IC01–IC03, CC01–CC03

Framing

Selective framing, narrative imprinting, social pressure

FR01–FR06, SP01–SP02

Addiction Patterns

Content mechanisms that drive compulsive consumption

AD01–AD08

The 8 patterns that make content addictive

This is what separates XRAE from everything else. These aren't persuasion techniques — they're the structural mechanisms that make content psychologically compulsive. Each one is mapped to published research.

AD01
Open Loops
Incomplete narratives that compel return visits
Zeigarnik Effect (1927)
AD02
Rage Bait
Outrage engineered as an engagement product
Brady et al., PNAS (2017)
AD03
FOMO Induction
Anxiety about missing out drives compulsive checking
Przybylski et al. (2013) — 250+ studies
AD04
Parasocial Dependency
Pseudo-intimacy that creates emotional attachment
Horton & Wohl (1956) — 70 years of research
AD05
Serial Dependency
Multi-part content creates sunk-cost attachment
Flayelle et al. (2020)
AD06
Variable Reward
Slot-machine pacing with unpredictable payoffs
Skinner (1957) — fMRI neuroimaging proof
AD07
Urgency Manufacturing
Artificial time pressure on non-time-sensitive content
Cialdini & Kahneman — Nobel laureate research
AD08
Identity Lock-in
Consumption becomes an identity marker
Swann et al. (2009) — Cambridge Handbook

Continuous analysis. No manual reviews.

XRAE ingests content from 135+ feeds, runs every episode and article through the detection model, and delivers results to your dashboard — automatically.

Ingest

RSS feeds, podcast audio, YouTube, and news articles flow into the pipeline continuously.

Analyze

XRAE runs multi-pass consensus detection across all 32 codes. Only consistent findings survive.

Deliver

Findings, trends, and exportable data appear in your dashboard within hours of publication.

135+
Active feeds
56+
Podcasts
61+
News outlets
24/7
Monitoring

Broad coverage. Deep analysis.

Prizm monitors content across the political spectrum — left, center, and right. Same model, same rubric, same standards for everyone.

Podcasts

56+ podcasts across left, center, and right. Full episode transcripts analyzed for all 32 techniques. Trending analysis shows patterns over time.

News & Articles

61+ news outlets monitored continuously. See how different outlets frame the same story. Compare influence profiles across the media landscape.

One plan. Full access.

No tiers. No feature gates. Every subscriber gets everything.

$199/month
  • Full access to podcast and news analysis
  • 32-technique detection incl. 8 addiction patterns
  • 135+ feeds monitored continuously
  • Historical data and trend analysis
  • Per-episode transcript-level findings
  • Export capabilities (CSV/JSON)
  • New feeds and features added regularly
Start Your Subscription

Cancel anytime. No long-term contracts.