Influencer brand safety
Influencer brand safety, proven before you sign.
One unsafe partnership can cost millions. CreatorScore scans every post, transcript, frame, and comment a creator has published — across 12 platforms and 200+ signals — and turns it into one score your team can act on before the contract, then watches it 24/7 after.
Reading 412 posts across 3 platforms…
What we screen for
Every post, transcript, and frame — checked
A clean caption over a risky voiceover is exactly what slips past manual review. CreatorScore reads the whole footprint — text, audio, on-screen text, and imagery — with NLP, computer vision, OCR, and speech-to-text.
Hate speech & extremism
Slurs, extremist ideology, and discriminatory language across 35+ patterns — in captions, transcripts, and on-screen text.
NSFW & explicit content
Computer vision on thumbnails, video frames, and images for nudity and sexually explicit material.
Profanity & vulgar language
Niche-aware — tells casual comedy language apart from genuinely hostile communication.
Misinformation & false claims
Health misinformation, conspiracy, and misleading claims that draw regulatory scrutiny or backlash.
Controversial topics
Divisive political and social content that can alienate segments of your audience.
Violence & graphic content
Visual and textual analysis for violent imagery and glorification of harm — including live streams.
@maya.fit
Fitness · wellness · daily workouts 💪
How scoring works
The Content Risk Agent, in the open
Content risk is the most heavily weighted part of every CreatorScore — because one incident can do lasting damage. Nine signals combine into a 0–100 risk score, and every one is visible and auditable.
Beyond the creator's own content
Open-web reputation is corroborated across multiple independent sources — never a single unverified thread.
Partnership & disclosure history
FTC disclosure rates and past collaboration outcomes factor in — measured from confirmed partnerships, not caption guesses.
Traceable to the post
Every penalty links back to the exact post, comment, transcript line, or frame that caused it, with SHAP explainability.
Content Risk Agent
9 weighted signals
- Hate signals18.4%
- Deceptive / scam content18.4%
- NSFW detection13.8%
- Flag severity9.2%
- Web controversy9.2%
- Visual risk analysis9.2%
- Misinformation9.2%
- Transcript controversy8%
- Profanity4.6%
When a signal doesn't apply, its weight redistributes across measured signals — creators are never penalized for missing data.
Non-negotiable thresholds
Some risks cap the score — no matter what else is good
A few risks are severe enough that no amount of positive signal should override them. Each cap is precise, evidence-gated, and designed so a single ambiguous frame never sinks a creator.
Bot commenter ratio > 60%
Capped at 20/100More than half the audience is artificial. Any spend reaches bots, not consumers — the most severe cap short of an auto-fail.
Engagement pods > 80%
Capped at 30/100Overwhelming evidence of coordinated fake engagement. Metrics are inflated and don't reflect genuine interest.
Produced hate speech (confirmed)
Capped at 35/1003+ posts read ≥95% hate AND an LLM confirms the creator produced it. Commentary that covers or condemns hate is never counted — a classifier hit alone can't fire this.
Sustained explicit content
Capped at 35/1003+ vision-confirmed explicit posts across ≥20% of the history. Maternity, fitness, medical, art, and cosplay are excluded, so one ambiguous frame never caps a creator.
FTC disclosure < 10%
Capped at 35/100Measured across 3+ verified brand partnerships. Caption keyword guesses never fire this — only confirmed partnership records do.
Documented harassment
Capped at 65/100Fires only on external evidence — a cease-and-desist, a lawsuit, a doxxing incident. An AI reading of tone alone can never cap a score.
Knockouts apply after all seven agents score — non-negotiable thresholds that override the weighted average when triggered.
- Hate signal detectednow
New Reel — slur in on-screen text (OCR)
@ryder.games
Real-time monitoring
Brand safety isn't a one-time check
A creator who scored 88 at signing can score 57 a month later. New posts, stories, and live streams are analyzed as they appear — you hear about a problem from CreatorScore, not from your CMO.
Instant alerts
Email, webhook, and dashboard notifications the moment a monitored creator crosses your thresholds.
Continuous scanning
24/7 across every connected platform — no waiting for a weekly review.
Score trending
Track how a creator's risk profile shifts over time, with a full historical audit trail.
Manual review vs CreatorScore
Why teams stop screening by hand
How AI-powered brand safety screening compares to a manual review process, across the metrics that matter.
| Metric | Manual review | CreatorScore |
|---|---|---|
| Content analyzed per creator | 10–20 recent posts | All posts, comments, transcripts |
| Time per creator | 2–5 hours | Under 15 minutes |
| Visual content screening | Manual spot-check | AI frame-by-frame analysis |
| Video transcript analysis | Rarely done — too slow | Automatic transcription + NLP |
| Consistency across reviews | Varies by reviewer | One fixed, reproducible model |
| Ongoing monitoring | Periodic manual checks | Continuous 24/7 scanning |
| Historical content review | Limited by time | Full-history analysis |
| Cost per creator | $50–200+ in staff time | From $0.50 / creator |
FAQ