Extracting a clean first name is more difficult than it appears. Display names contain brand words ("Amy's Cake Shop"), short stylized handles ("xX_dani_Xx"), middle names, last names, and non-Latin scripts. The pipeline pulls candidates from three sources in priority order, scores each one, normalizes non-Latin scripts, and falls back to the username only when nothing else resolves.
Coverage
Share of profiles where we resolved a confident first name.
Coverage by Platform
The Three Sources
Three independent sources are checked. The email path is most accurate when an email exists. The display-name path covers most cases. Username is the last-resort fallback when nothing else resolves.
Sources in Priority Order
When the profile lists an email, we use a third-party name inference service to identify the first name from the email. This is the highest-accuracy path, but only available when an email is present.
The display name is tokenized into individual words. Each token is checked against a multilingual name database covering hundreds of thousands of names, combined with an in-house curated names list. The candidate with the highest probability score wins, after filtering for length and excluding business / location words.
If neither email nor display-name produces a confident first name, we surface the username with an @ prefix so it's visually obvious to the customer that this is a handle, not a real name. Example: travelmoments2023 → @travelmoments2023.
Non-Latin Scripts Get Normalized First
Before any database lookup, names written in non-Latin scripts are transliterated into Latin characters. This lets a Russian name and an English name match the same probability table.
Examples of Transliteration
First Name Extraction for Two Illustrative Profiles
When the email prefix and the display name agree on the same first name, confidence is highest. Both sources independently point to the same candidate, which makes false positives extremely unlikely. Brand-like display names (like "BorisLifeLog" or "NikolaTravels") trigger additional checks to confirm that the first part is actually a given name and not a brand. The country filter and name database probability score work together to make that determination.
Cross-Platform Signal
When we match a creator's profiles across platforms, first name data from one platform can fill gaps on another. If a creator's channel name on one platform clearly contains a first name, but their Handle is just a brand word, the matched profile carries the name over.
How a First Name Fills Across Platforms
TikTok handle did not parse to a name, now filled from the matched YouTube profile
Our profile matching system links accounts that belong to the same creator. When a first name is extracted on one platform but the other only has a brand-style handle that does not parse cleanly, the matched value fills the gap. If two platforms extract different names (for example, a nickname on one and a legal name on the other), the higher-confidence extraction wins.
Validation 
Every accuracy number on this page comes out of a two-track validation process. We compare at least two independent AI sources on tens of thousands of records, and we manually validate at least 1,000 records per subset, with subsets defined to cover every meaningful use case.
Source-Level Accuracy
An entry counts as wrong when the extracted first name is not a real first name (a brand word, a last name, or an unrelated token). Cases where a real first name was extracted but the profile is ambiguous (e.g. "Amy's Cake Shop", where Amy is a real first name belonging to the owner) are excluded from the error count, since the extraction is technically correct.
Validated on 20K profiles per platform