The AI That Calls Bullshit
Most AI tools summarize content. GriftFilter interrogates it. Three specialized AI agents work together to separate genuine insights from manipulative marketing.
The Problem With Online Advice
YouTube is flooded with self-proclaimed experts promising life-changing results. Some deliver genuine value. Many don't. The problem? It's nearly impossible to tell the difference until you've already wasted hours watching, or worse, spent money on their course.
What you usually get:
- 10 minutes of "value" buried in 45 minutes of hype
- Vague advice like "provide value" with no specifics
- Cherry-picked success stories, survivorship bias ignored
- Hard prerequisites conveniently skipped over
What you actually need:
- Honest assessment: is this worth your time?
- The useful parts extracted and made actionable
- What they didn't tell you (the hard parts)
- Marketing tactics identified so you're not manipulated
Our Three-Agent System
GriftFilter doesn't just use AI to summarize content. We've built a pipeline of three specialized AI agents, each with a distinct role. This adversarial approach catches what single-AI systems miss.
The Ruthless Critic
Claude Opus 4.5A battle-hardened skeptic who has seen every marketing trick in the book. Explicitly designed to be harsh, direct, and uncharitable to commercial claims.
What it does:
- • Identifies misleading claims with evidence
- • Spots the business model and profit motives
- • Finds valid nuggets anyway (credit where due)
- • Assigns deception & substance scores
Why it matters:
- • No benefit of the doubt given to marketing
- • Every criticism backed by transcript quotes
- • Assumes commercial intent by default
- • Still recognizes genuinely good content
The Extractor
Claude Sonnet 4.5A pragmatic translator who takes only the Critic's validated insights and transforms them into something you can actually use.
What it produces:
- • Step-by-step action playbooks
- • Copy-paste templates and checklists
- • Time and cost estimates
- • Failure modes to watch for
What it strips out:
- • Self-promotion and upsells
- • Vague motivational fluff
- • Unnecessary storytelling
- • "Subscribe for more" filler
The Verifier
Claude Sonnet 4.5The fairness guard. Not here to soften criticism—here to ensure we're right. A wrong harsh take is worse than a right mild one.
What it checks:
- • Did the Critic misrepresent anything?
- • Are quotes accurate and in context?
- • Were any fair defenses overlooked?
- • Are scores properly calibrated?
Why this matters:
- • Catches unfair dunks before you see them
- • Can adjust scores up or down
- • Produces the "fair take" for public display
- • Our credibility depends on accuracy
Why This Actually Works
Adversarial by Design
Single AI systems are sycophantic—they tend to agree with whatever they're analyzing. Our agents argue with each other. The Critic assumes the worst. The Verifier pushes back if that's unfair.
Structured, Not Summarized
We don't produce fluffy summaries. Every agent outputs structured data: scores, arrays of claims, actionable playbooks. This makes the analysis searchable, comparable, and actually useful.
Evidence-Based Claims
Every criticism must cite the transcript. No vague accusations. You can verify every claim we make by checking the exact quotes and timestamps we provide.
Calibrated Scoring
A 90+ deception score means something specific: demonstrably false claims, predatory tactics. A 0-9 score is reserved for genuinely educational content. We don't inflate scores for drama.
Understanding the Scores
Deception Score (0-100)
Substance Score (0-100)
How much actionable, useful content is actually here? Would someone benefit from this even without buying anything the creator sells?
High deception + high substance = valuable content packaged with manipulation. High deception + low substance = avoid entirely. Low deception + high substance = the rare gem worth your time.
Ready to Stop Getting Played?
Paste any YouTube URL and get an honest, evidence-based analysis in minutes. Free to use.
Analyze a Video