How Legit works
Two probabilities per profile. Never a verdict. Every input is listed, every score carries a confidence level, and nothing anyone pays for can move a number.
The two scores
Being fake and being ghost-written are different problems. A binary "scam or legit" would hide the second one entirely, so we measure both, separately.
Account authenticity
Does this account belong to the person shown? Only facts we can observe ourselves count — accounts that link to each other in both directions, a claim made from her own bio, the account's public history — and votes never move it. Until those facts are enough to say something, no figure is published.
Chatter probability in DMs
How likely is it that someone else answers her messages? Built only from structured reports by subscribers: whether replies remember earlier conversations, whether attention fades once you have paid, scripted or round-the-clock replies, paid extras that never arrived. Each answer moves the estimate by a fixed, published amount.
Trust index
One readable number for comparison: 60 % authenticity, 40 % inverted chatter probability. A summary of the two scores above, nothing more.
Bayesian, not a star average
A five-star average is trivially gamed and says nothing about who left the stars. We start from evidence and let contributions move the needle only when they earn it.
Public facts
When a profile is created we read its public page once: account age, posts, and the accounts its bio links to — and whether they link back. These are shown as facts. Before anyone reports, the chatter estimate sits at a fixed prior and is not published.
Weighted reports
A swipe plus three structured questions. Each contribution carries the weight of its author, not one vote each.
Manipulation checks
Velocity spikes and correlated accounts are quarantined. Forty votes from day-old accounts move nothing and become a signal themselves.
Honest confidence
Thin evidence ships as low confidence and says so, instead of dressing a guess up as a percentage.
It rests on facts we observe ourselves on her public page — audience, size of the library, the accounts she links to and whether they link back, the platform's own mark — never on votes. The full rubric is below; below three facts read, no figure is published.
Whether someone else answers the DMs is behavioural. Nothing on a public page reveals it, so it stays unscored until subscribers report. We show "not enough evidence" rather than a number built from proxies.
When a platform exposes too little, both gauges stay empty. The profile still exists, still lists what we could and could not read, and carries no figure at all. An invented number is worse than no number.
Platforms are not equal here. Fansly publishes a public profile we can read when an account is submitted. MYM's public page shows a few facts, such as its own "Certifié" mark and the number of posts. OnlyFans shows a visitor nothing outside its app, so an OnlyFans account carries no public fact at all. We would rather say that plainly than paper over it.
The chatter rubric, in full
This is the entire model behind the chatter probability — every question, every answer, and exactly how much each one moves the number. It is loaded live from the scorer itself, not retyped here, so what you read is what runs. If a coefficient changes, this page changes with it.
It produces the chatter probability only. Authenticity rests on public facts observed on the platform, never on opinions — mixing them would let a coordinated group move a number that is meant to be a measurement — and has its own rubric, below.
Adding up evidence would be right if each contribution were an independent observation of a fixed truth. They are not. They are correlated opinions from people who chose to show up. A mean pulled towards the prior converges slowly and refuses to become certain, which is the correct posture when the subject is someone's livelihood.
Whether someone recalls your previous conversation, and whether the attention survives the first billing cycle, are the closest things to a direct read on whether a paid operator is answering. Tone weighs far less: plenty of people write briskly, and plenty of chatters are warm.
It cannot reach certainty. Sixty detailed hostile reports still land below 95%, and a profile with two reports publishes nothing at all. If you contest a score, we can show you the exact lines that produced it and how many people triggered each one.
The authenticity rubric, in full
Does this account belong to the person it shows? Every rule, and exactly how much it moves the estimate. Loaded live from the scorer, like the chatter rubric: what you read is what runs.
Copied photos are easy; making the creator's real account link back to a copy is not. Two of her accounts naming each other, each from a page she controls, is the closest thing to identity a public page can show.
A large audience says an account is a real business, not whose face it is — an account can be sold. That is why size alone never lifts confidence above low, and why the figure never reaches 0 or 100.
Who weighs what
The anti-gaming design is the product. Weight is earned, never bought, and we never ask for an account before you have contributed something.
One creator, many handles
Profiles track creators, not handles. A rename, a second platform, a fresh account after a ban — all of it merges into one reputation instead of fragmenting into three.
Three anchors
The platform's internal ID, which survives renames. A perceptual hash of the avatar, which survives recrops. Linked socials, which tie platforms together.
Consolidated scores
When the same creator runs OnlyFans and Fansly accounts, the profile lists both and the scores cover the person. Reputation follows her, not the paywall she posts behind.
Merging is never a user action
Tying two accounts together merges two reputations, and a wrong merge is invisible to whoever reads the result. So contributors can only suggest a link. Accounts are joined automatically when an identity anchor matches, or by a moderator — never on one person's say-so.
Old URLs keep working
Previous handles join the profile's history and their URLs redirect to the current one, so a rename never orphans what people already know.
"Known as": public names, weighted
Handles are unreadable and change often, so profiles also show the name people actually use. That name is crowdsourced and weighted, never counted: a name pushed by ten fresh accounts loses to a name reported by two long-standing contributors.
Names she has published herself, on an account she controls — Instagram bio, X, a link-in-bio, her own site.
Legal names, locations, employers, former identities she has moved away from. Removed on sight, and it costs the contributor their entire weight.
Any creator can contest a name by claiming her profile. No payment, no ticket, no delay.
The ranking only goes up
We publish the accounts with the strongest signals, and only those. A profile enters the ranking with at least 25 weighted contributions, confidence above low, authenticity at 70 % or more and chatter probability at 45 % or less.
A ranking of the worst accounts. A low probability is a warning to read the signals yourself; turned into a leaderboard it becomes an accusation, which is exactly what this project refuses to make.
Profiles below the bar are simply absent from the ranking. They keep their page, their listed signals and their right of reply.
The badge, and what we actually sell
Legit has to pay for itself without ever selling a score. So we sell distribution and insight to creators, and nothing else.
Free, forever
Reporting an error, contesting a signal, asking for a "known as" entry to be removed. Anything that could change what the page says about someone is free, for anyone, with no account.
The creator plan
Ownership verification and the live badge for your link-in-bio, today. Profile analytics and impersonation alerts are planned, not built. Free while we are in beta, and we will tell you before that changes.
What paying never buys
A better score. The badge serves the current number: if it drops, the badge drops with it. It also always carries the handle and the number of reviews behind the score, so it cannot be lifted onto another account.
One practical note, because it shaped the design: Linktree and most link-in-bio tools accept a link and a thumbnail image, not HTML or iframes. So the badge is served as an image as well as an embed — the image is what actually works on a link-in-bio, the iframe is for sites you control.
Creators: what exists and what does not
There is no "claim your profile" ritual. It would be a door onto nothing: a profile is public data and public reports, and neither becomes yours by pressing a button. So there are exactly two things a creator can do here.
Report an error — free, for anyone
Anyone can flag a factually wrong signal, a public name that should not be there, or an account wrongly tied to a profile. Creators included, at no cost, with no account required. A contested signal is marked as disputed on the page while it is reviewed. This is not a favour: publishing a score about someone without a free way to challenge it would be indefensible.
The creator plan — paid
Ownership verification and the live badge; analytics and impersonation alerts are planned. Free while we are in beta. Verification happens as part of joining, which is also what puts the "creator verified" marker on the profile.
Privacy by design
Pseudonymous accounts
Magic link, no password, no real name, no public profile. The email is hashed and deleting the account leaves your contributions anonymous.
Nothing leaves
Hashed IP addresses, EU hosting, and no sharing of what you check or report. That is not a policy promise, it is the architecture.
Creators are people too
We only use strictly public data, we host thumbnails rather than hotlinked images, we blur what is flagged, and we run a documented erasure process.