An OnlyFans statistic is only interpretable when you know what was counted. Profiles, people, links, messages and test cases are different units. Our own research illustrates the problem: 461 directory records, 120 selected biographies and 40 search cases cannot be pooled into a total of 621 creators.
Three studies, three different questions
These are dated OnlyDetect observations, not estimates of the size or demographics of OnlyFans.
Scroll across the table on smaller screens.
| Study | What one row represents | Selection | Question it can answer |
|---|---|---|---|
| Directory snapshot | One eligible directory profile | Active, indexable, not suppressed, validated within 180 days | What fields and prices were stored for these 461 records? |
| Official-catalog audit | One selected platform biography | Ten initial cases followed by catalog return order to 60 athletes and 60 comedians | How many of these 120 pages exposed account links? |
| Search experiment | One search request | Four variants for each of ten selected targets | Did that request return the intended target within the tested result limit? |
The directory can contain accounts belonging to the same person. The biography sample was deliberately balanced by two professions. The search experiment reused each target four times. None of those designs measures the percentage of all fans who prefer a particular kind of creator.
A denominator changes the meaning
Our catalog audit extracted account links from 115 of 120 selected biographies. That is a link-coverage result for those pages. Comparing the same 115 handles with normalized public names answers another question: 42 matched that string rule and 73 differed. Those differences are not 73 incorrect accounts.
The search experiment produced 40 requests, but only ten target profiles were involved. Treating all 40 as independent people would exaggerate the breadth of the sample. It would also hide that the deliberately incompatible offer-type-filter cases tested a different condition from the exact-handle cases.
This is why we publish the search method and outcome table separately from our name-and-handle comparison. Sharing a platform keyword does not make two measurements interchangeable.
Five checks before repeating a number
- Identify the unit. Is the number about registered accounts, active creators, paying fans, respondents, posts or transactions?
- Find the denominator. A percentage without the group it describes is incomplete.
- Read the selection rule. A directory, voluntary survey or forum thread has a different selection process from a platform-wide administrative record.
- Keep the observation date. A current article can still cite an older snapshot.
- Check what is missing. An empty subscriber field does not mean zero subscribers, and an inaccessible page does not mean an account was deleted.
What our comparison adds
The contribution here is a crosswalk between three actual study designs. It shows exactly where an apparently simple total would double-count targets or mix units. We have not estimated platform demographics, adjusted for selection bias or inferred anyone's private identity.
For a subscription decision, statistics can provide context. The current official account, its stated inclusions and the displayed purchase terms remain the relevant evidence for that specific offer. Start with the public-profile search guide, then keep any broad statistic separate from the account-level facts.