CreatorScore▸ How it works
One handle in. One score out. Everything in between, read.
Give us any account a creator owns. We find the rest, check each one is really theirs, read every post, spoken word, frame and comment, and weigh it all through seven agents into one score from 1 to 100, with the proof behind every finding.
Find every account, and prove each one is theirs.
From one handle we look for the creator’s other accounts, then fetch each profile and check it against what the creator published: a link in their bio, a bio that points back, a matching name and handle. Accounts that turn out to belong to someone else are dropped and listed as checked, not theirs.
Read every post, not just the caption.
Videos are transcribed, so what was said counts as much as what was written. Frames are checked by vision models, text on screen is read, and brand mentions are pulled from all three. The risk is usually in a layer a person scrolling never reaches.
Read the audience, one comment at a time.
Follower counts are easy to inflate. Comments are harder. Each one is scored for bot signals, coordinated engagement pods are found by who comments together, and sentiment comes from the audience’s own words.
Weigh it through seven agents.
Content Risk, Authenticity, Brand Safety, Audience Quality, Sentiment & Voice, Community Trust and ROI Prediction each score 0 to 100 on fixed weights. Severe, confirmed problems cap the score however good the rest is, and every finding links to the post behind it.
Every step leaves a trail. The score you see links back to the accounts we checked, the posts we read and the comments we counted, so anyone can follow it from the number to the proof.
▸ Questions
How the score is built, answered.
Do you analyze the actual video, or just the caption?
The actual video. Videos are transcribed, so what was said counts as much as the caption, and frames and thumbnails are checked by vision models for risk that never appears in text. Text on screen is read too.
How do you detect fake followers and bots?
At the comment level, not the follower level. Follower counts are easy to inflate and hard to verify. Each comment is scored for bot signals, engagement pods are found from groups of accounts that comment together, and the share of bot-like comments feeds the Authenticity agent.
What are the 7 agents?
Content Risk (20%), Authenticity (20%), Brand Safety (15%), Audience Quality (15%), Sentiment & Voice (10%, shown as Voice Stability in the dashboard), Community Trust (10%) and ROI Prediction (10%). Each scores 0 to 100 and the weighted average is the CreatorScore.
Is a CreatorScore one platform or all of them?
All of them. Give us one handle and we find the other accounts the creator owns, check each one is theirs, and return one score for the whole footprint, with the per-platform breakdown underneath.
How far back does the score look?
The scoring window looks back 180 days, widened to 720 days for creators who post less often. The window is stated on the score. When a brand needs the entire history, a Social Background Check reads all of it.
Can the score be gamed by buying engagement?
No. Caps apply when severe problems are confirmed: bot-like commenters above 60% cap the score at 20, engagement pods above 80% cap it at 30, and so on. A high engagement rate can't hide a real problem.
Is it fair to smaller creators?
Yes. Every creator is compared with creators their own size on the same platform, so a focused creator with 5,000 followers can score as well as one with 5 million.
How current is the score?
On a recurring plan the whole footprint is rescored every 30 days. With monitoring, new posts are read every day and a critical or high finding that holds up on review is emailed the same day. Live streams aren't monitored.
Can I see why a flag was raised?
Always. Every finding links to the post, comment or video frame behind it, with the reason. A creator can see exactly what to fix, and a brand can see exactly why a flag exists.