The scoring standard

What counts as a good CreatorScore?

Every creator gets one score from 1 to 100 across their whole cross-platform footprint. This page is the rubric behind it: what we measure, how much each signal weighs, the caps that override everything else, what each band should mean to a brand — and what a creator can actually do to move it.

1 — highest risk100 — lowest risk
1–59
Poor
60–69
Fair
70–79
Good
80–89
Excellent
90–100
Exceptional

Most creators brands actually partner with land between 70 and 89. A score below 60 nearly always means a cap is active — a specific, named failure that no amount of good performance can average away.

What the number means — and what to do about it

A score is only useful if it changes a decision. Here is the band, what it actually indicates, and the move it should trigger on your side.

90–100
Exceptional
Sign
What it indicates

Clean across all seven agents with real audience engagement. No open flags, verified disclosure history, authentic followers.

Rare. Typically creators with a long, consistent, well-disclosed track record.

Your move

Standard terms. Low-touch monitoring is enough — you still want alerts, but you are not managing a risk.

80–89
Excellent
Sign
What it indicates

Strong on every dimension with at most minor, well-understood dents — a little profanity, a modest engagement dip, one thin data source.

Common among established creators who take partnerships seriously.

Your move

Read the open flags once, then proceed on normal terms. This is the band most brands are looking for.

70–79
Good
Sign with terms
What it indicates

Safe to work with, with specific things worth reading. Usually a handful of flags that are real but contextual — a spicy transcript moment, casual profanity, an older post.

The largest band. Most creators you partner with live here.

Your move

Open the flags and decide against your own category. A comedy creator at 74 and a finance creator at 74 are not the same risk. Add the relevant contract clause and move.

60–69
Fair
Monitor closely
What it indicates

Something real is dragging the score down — a pattern across content, weak audience authenticity, sustained negative sentiment, or an active cap in the 60s.

Uncommon. Treat it as a prompt to read, not a reason to ghost.

Your move

Not a rejection — a different operating mode. See the operating manual below for exactly what "monitor closely" means in practice.

1–59
Poor
Hold — read the cap first
What it indicates

Almost always a cap. A score of 55 does not mean "slightly worse than 60" — it means a specific knockout fired and is holding the number down.

Rare, and never silent — the reason is always named on the profile.

Your move

Open the knockout and read its evidence before you do anything. If it is real, decline. If it looks like a false positive, the flag names the exact posts, and both you and the creator can contest it.

The band and the label come from the scoring pipeline. The recommended action is our guidance, not a verdict the product issues — CreatorScore never auto-rejects a creator on your behalf. We surface the evidence; the decision is yours, and it should depend on your category, your audience, and your tolerance.

What we score against — and what each part weighs

Seven agents each score the creator 0–100 on one question. The weighted average of those seven is the CreatorScore. No hidden hand-tuning, no secret eighth factor.

One CreatorScore = 100% of the weight
Content Risk 20%Authenticity 20%Brand Safety 15%Audience Quality 15%Sentiment & Voice 10%Community Trust 10%ROI Prediction 10%
20% of score

Content Risk

What the creator publishes — across text, audio, and pixels.

Hate signals, deceptive/scam content, NSFW, misinformation, on-screen visuals, spoken transcript moments, flag severity, profanity.

20% of score

Authenticity

Whether the audience is real.

Comment-level bot analysis (80%) blended with follower authenticity (20%): bot probability, engagement-pod detection, growth-curve anomalies, like-to-comment ratios.

15% of score

Brand Safety

Whether attaching a brand to them is safe.

FTC disclosure compliance (26%), brand-partnership patterns (23%), controversy breadth (15%), corroborated web reputation (11%), X feuds (10%), video transcripts (10%), following-graph risk (5%).

15% of score

Audience Quality

Whether the audience is worth reaching.

Engagement rate (30%), community health (20%), engagement depth (15%), velocity (10%), niche fit (10%), plus demographics, loyalty, and live performance (5% each).

10% of score

Sentiment & Voice

Whether their voice is coherent, and how the audience receives it.

Voice consistency (30%), audience sentiment (25%), tone safety (20%), engagement style (15%), content adaptability (10%). Shown as "Voice Stability" in the dashboard.

10% of score

Community Trust

Whether their recommendations hold up.

FTC disclosure compliance on verified partnerships (45%), brand alignment (30%), and creator conduct toward their own community (25%).

10% of score

ROI Prediction

Whether a campaign is likely to work.

Engagement (40%), growth trajectory (25%), shares and saves (20%), community health (15%) — a forward-looking signal, not a history lesson.

Inside Content Risk
Hate signals18.4%
Deceptive / scam content18.4%
NSFW13.8%
Flag severity9.2%
Web controversy9.2%
Visual analysis9.2%
Misinformation9.2%
Transcript controversy8%
Profanity4.6%

When a signal genuinely doesn't apply — no video to transcribe, no web coverage to search — its weight is redistributed across the signals we did measure. We never hand a creator a default score for data we don't have.

The score is always the creator's full cross-platform footprint, never a single account. Per-platform numbers exist as a breakdown underneath — but a creator is a person, not a TikTok handle, and that is what a brand signs. See the full pipeline.

Knockouts — the caps that override everything

Some failures can't be averaged away. A creator with world-class engagement and a bot farm for an audience is not a 78 — the bot farm caps them at 20. Here is every cap in production, what fires it, and the evidence bar it has to clear first.

Cap
What fires it
1
Child exploitation material
Any CSAM indicator detected in content or visual analysis.
20
Bot audience
Bot commenter ratio above 60%, measured across at least 10 analyzed comments.
25
Promotion of illegal activity
High-confidence, high-severity flags for actively promoting illegal activity.
25
Grooming / predatory signals
Three or more distinct grooming-pattern signals detected.
30
Engagement pods
Coordinated reciprocal-engagement probability above 80%.
35
Produced hate speech (pattern)
Three or more posts where the classifier reads ≥95% hate AND an LLM review confirms the creator produced it (not covered or condemned it).
35
Sustained explicit content
At least 3 vision-confirmed explicit posts AND explicit content in ≥20% of analyzed posts, outside protected contexts (maternity, fitness, medical, art, cosplay).
35
FTC disclosure failure
Disclosure on under 10% of posts, across at least 3 verified brand partnerships. Caption guesses never fire this — only confirmed partnership records do.
40
Critical real-world controversy
Web search surfaces a critical, corroborated controversy — felony charges, assault, active proceedings.
40
Unqualified medical claims
Health claims presented without professional credentials, above the misinformation-risk threshold.
50
Produced hate speech (isolated)
One or two LLM-confirmed posts producing hate speech. Fewer posts, softer cap — but still a cap.
55
Deceptive content, corroborated
Scam / MLM / get-rich-quick patterns in ≥20% of posts AND a web record corroborating the allegations.
55
Manosphere / red-pill content
That content in ≥20% of posts. Commentary and news creators who cover the topic are exempt — unless conduct analysis shows they are also bullying.
60
Deceptive content, high rate
Scam / MLM / get-rich-quick patterns in ≥30% of posts, even with no external corroboration.
60
Bullying / rage-bait pattern
Conduct quality below 40 plus at least two corroborating signals: punch-down posts, protected-trait mockery, harassment call-outs, or documented harm.
65
Documented harassment
Third-party external evidence only — a cease-and-desist, a lawsuit, or a doxxing incident. An AI reading of the creator’s own tone can never fire this.

“Cap” is the highest score a creator can hold while the condition is true. When two caps fire, the lowest one wins. Clear the condition and the cap lifts on the next rescore — a knockout is a state, not a permanent mark.

A single post never caps a score

Every content-based knockout requires a pattern — multiple confirmed posts, or a prevalence threshold across the whole history. One bad frame, one heated caption, one bikini photo is a flag, not a cap.

A classifier alone is never enough

Hate caps require the classifier AND an LLM review confirming the creator produced it. Disclosure caps require verified partnership records, not caption keywords. NSFW caps require vision confirmation, never a raw thumbnail score.

Covering a topic is not committing it

Commentary, news, and true-crime creators discuss hate, crime, and scandal for a living. Knockouts are gated so that reporting on a thing does not score like doing it — a distinction keyword scanners get wrong constantly.

Every cap on a profile names itself and shows its evidence — the specific posts, the quote, the article. A creator can read exactly why they were capped and contest it; a brand can read the same thing and decide whether it matters for their category. We don't apply a penalty we can't show you the proof for.

For brands

What “monitor closely” actually means

Every vetting tool says it. Almost none of them define it. Here it is as an operating mode — five things you actually do differently, not a feeling you have about a creator.

01

Read every open flag before you sign

Not the count — the flags themselves. Each one names the post, the quote, or the article behind it. Half the work of "monitor closely" is discovering the flag is irrelevant to your category, or that it absolutely is not.

02

Turn on daily monitoring for that creator

New content is synced and scanned every day, and you get alerted when a new flag fires or the score moves. You want to hear about a problem from your dashboard, not from a journalist.

03

Tighten the contract, not the relationship

Disclosure requirements in writing, a morality clause with named triggering conduct, content pre-approval for the first flight, and the right to pause without terminating.

04

Pre-agree what pauses the campaign

Decide the trigger before you need it: a new high-severity flag, a score drop past a threshold you set, a live controversy. Pausing paid amplification is cheap and reversible — indecision is what costs money.

05

Re-check before every renewal

The score is refreshed monthly and monitored daily. A creator who was a 74 in March can be a 58 in July. Never renew on a number you pulled a quarter ago.

When a flag fires mid-campaign

Decide this in advance. The moment a flag appears is the worst possible moment to invent a process.

1A new flag fires mid-campaign

Open the evidence first. A surprising share of severe-looking flags are context failures — satire read as sincerity, a quoted slur in a video condemning it. Verify before you act.

2The evidence holds up

Pause paid amplification (hours, not days), ask the creator directly, then decide against the tolerance band you wrote down before the campaign started.

3The flag is wrong

Mark it. A false positive dismissed on your account stops distorting that creator’s profile for you, and a globally-wrong flag gets corrected for everyone.

The cost nobody counts is the creator you rejected who would have been your best partner of the year. Over-rejecting is not caution — it is a different, quieter kind of expensive. Flags are visibility, not verdicts.

For creators

How to actually raise your score

Your score exists whether or not you have ever heard of us — it is built from public content, and brands can pull it. The good news: every lever is knowable, and none of them are secret.

This weekCommunity Trust · Brand Safety

Disclose every sponsored post properly

#ad or #sponsored in the first line of the caption — before the "see more" cut — plus the platform's native paid-partnership label, plus a verbal disclosure in the first seconds of video. Disclosure rate is measured against verified partnerships, so every correctly-labelled new post starts pulling the number up immediately.

Highest-leverage fix available. Also the one that clears a cap at 35.
This weekContent Risk

Audit your own back catalogue

Run your handle through the free score check and read the flagged posts. Old content counts — a 2018 tweet, an early video, a caption you forgot. Delete what is indefensible, and be ready to give context on what is defensible but spicy.

Removing a flagged post clears its contribution on the next rescore.
30–60 daysAuthenticity

Stop the inflow of fake engagement

No bought followers, no follow-for-follow, no engagement pods. There is no fast fix here — bot ratios fall as platforms purge fake accounts and real growth dilutes what remains. It is slow, and it is the only thing that works.

The bot cap at 20 and the pod cap at 30 are the two harshest caps we apply.
30–90 daysAudience Quality · Sentiment

Cultivate the comment section you want scored

Reply substantively. Moderate the toxicity. Your community is part of what a brand is buying, and comment quality and audience sentiment are measured directly — a great creator with a hostile comment section scores like a risk.

Slow-moving but compounding across two agents worth 25% combined.
ImmediatelyAny

Contest a flag that is wrong

Every flag names the post behind it. If a satire read as sincerity, or a quoted slur in a video condemning it got scored as hate speech, that is our error and you can dispute it. False positives get corrected — globally, for every brand looking at you.

The fastest score correction there is, when the flag is genuinely wrong.

What doesn't work

  • Buying followers to look bigger — it is the single most detectable thing you can do, and it caps you at 20.
  • Deleting only your recent posts — archives and our full-history scan still see the rest.
  • Starting a clean second account — a CreatorScore is the whole footprint, and we resolve the accounts you did not mention.
  • Waiting it out — the score refreshes monthly, so nothing improves until the underlying content or audience does.

One thing we will never do: penalize you for data we don't have. A quiet account, a platform we can't read, a thin comment section — those produce an honest “not enough data,” never a low number dressed up as a verdict.

Benchmark questions

The questions brands and creators ask most about what the number means.

  • What is a good CreatorScore?

    +
    A CreatorScore of 80 or above is Excellent and is what most brands look for. 70–79 is Good and is where the majority of creators brands actually partner with land — safe to work with, with specific flags worth reading. 60–69 is Fair and calls for close monitoring. Below 60 is Poor and almost always means a knockout cap is active — a specific, named failure that no amount of strong performance can average away. 90 and above is Exceptional and is rare.
  • What is a knockout factor?

    +
    A knockout is a hard cap on the score, applied when a severe issue is detected, regardless of how strong the creator is everywhere else. A bot commenter ratio above 60% caps the score at 20. Engagement pods above 80% cap at 30. A confirmed pattern of produced hate speech, sustained explicit content, or FTC disclosure under 10% across verified partnerships each cap at 35. Documented external harassment caps at 65. Caps are a state, not a permanent mark: clear the condition and the cap lifts on the next rescore.
  • How are the seven agent weights distributed?

    +
    Content Risk 20%, Authenticity 20%, Brand Safety 15%, Audience Quality 15%, Sentiment & Voice 10%, Community Trust 10%, and ROI Prediction 10%. Each agent scores the creator 0–100 on its own question, and the weighted average of the seven is the CreatorScore. When a signal genuinely does not apply to a creator — no video to transcribe, no X account, no web coverage to search — its weight is redistributed across the signals we did measure, so a creator is never penalized for data we do not have.
  • What does it mean to "monitor closely" a creator?

    +
    It is an operating mode, not a feeling. Concretely: read every open flag and its evidence before signing; turn on daily content monitoring and alerts for that creator; tighten the contract with disclosure requirements, a morality clause, content pre-approval for the first flight, and pause rights; pre-agree the trigger that pauses the campaign (a new high-severity flag, a score drop past your threshold); and re-check the score before every renewal rather than relying on a number you pulled a quarter ago.
  • How can a creator improve their score?

    +
    The highest-leverage fix is FTC disclosure: put #ad or #sponsored in the first line of the caption, use the platform native paid-partnership label, and disclose verbally in the first seconds of video. Then audit your back catalogue and remove or contextualize old flagged content. Stopping the inflow of bought followers and engagement-pod activity improves authenticity over 30–60 days as platform purges and organic growth dilute the fake accounts. Improving comment quality and community health moves two more agents. And if a flag is simply wrong, you can contest it — every flag names the post behind it.
  • Does a low score mean a creator is bad?

    +
    No. A score is a risk measurement against a general brand-safety standard, not a judgment of a person. Risk is also relational: a creator whose content is disqualifying for one brand can be exactly right for another. That is why CreatorScore never auto-rejects anyone on a brand's behalf — every flag names the specific post, quote, or article behind it so the brand can decide what matters for their category. Flags are visibility, not verdicts.
  • Why does a single post never cap a score?

    +
    Because one post is not a pattern, and false positives are expensive in both directions. Every content-based knockout requires corroboration: hate caps need a classifier hit AND an LLM review confirming the creator produced the content rather than covering or condemning it; explicit-content caps need multiple vision-confirmed posts and a prevalence threshold across the full history, with protected contexts such as maternity, fitness, medical, and art excluded; disclosure caps need verified brand-partnership records, never caption keywords.
  • How often is a CreatorScore updated?

    +
    Monitored creators are re-scored monthly, with daily content syncing and risk scanning in between, so a new high-risk post or a sudden change in audience authenticity raises an alert rather than waiting for the next refresh. This matters for brands because a creator who scored 74 in March can be a 58 in July — a score is a snapshot, and monitoring is what keeps it true.

Want the mechanics behind the numbers? Read the scoring methodology, see the accuracy benchmark, or check any creator's score free.

Vet your next creator
before you sign the deal.

One score across every platform, every risk named and traced to the post that caused it. Make the call in minutes — and back it up with evidence.

Vet a creator
No setup fees
Results in 15 minutes
Cancel anytime