Tony Wang4 min readEvery Public X Dataset You've Seen Is Biased Toward Fame — Including Ours
Our X dataset is seeded from notable people: only 19.6% of accounts have under 100 followers, vs. ~50% for active accounts in published research.
Most public X/Twitter datasets — including our own — are built by starting from some list of accounts worth indexing and expanding outward. That's a reasonable engineering choice. It also means the resulting dataset looks nothing like the platform's actual population, and the honest thing to do is show exactly how much, using our own data as the example rather than picking on someone else's.
What our seeded index looks like
What broader research on X/Twitter says instead
2023 academic sample, median ≈99 followers
Both figures are internally consistent with how they were built. Our index is seeded from people who are independently notable enough to have a Wikidata entry, or technical enough to link an X handle on GitHub — a genuinely useful population for, say, finding a public figure's or a developer's account, and a genuinely unrepresentative one for asking "what's a typical X account look like."
The part worth generalizing
This isn't a complaint about our own dataset — it's already documented honestly in the dataset's own source comments, which is exactly why it was easy to write this post. The generalizable point: any public profile dataset for a social platform that's built via search, discovery, or a "notable people" seed — TikTok creator indexes, LinkedIn company/profile scrapes, Instagram business directories, X/Twitter datasets like ours — inherits a fame skew by construction, and it's rarely stated as plainly as "here's the actual gap between our numbers and a broader population estimate."
If a number like "the median account has N followers" is doing real work in a decision — sizing an influencer campaign, benchmarking your own account, judging how niche a topic is by who talks about it — the question to ask first is what seeded the dataset that number came from, not just how big the dataset is.
Query our X/Twitter profile dataset — skew disclosed up front
796,611 public profiles, seeded from notable-person and developer sources and documented as such. Follower bands, verification, bio and account-age facets as structured JSON.
Frequently asked questions
Is a public X/Twitter profile dataset representative of the platform?
Usually not, and ours is a clear example. Our 796,611-account dataset is seeded from Wikidata-notable people and GitHub-linked handles, so only 2.4% of accounts have zero followers and 19.6% have under 100 — versus published academic research on active Twitter accounts, which puts roughly half of active accounts under 100 followers (median near 99).
Why does this happen?
Any dataset built by discovery, search, or a 'notable people' seed list inherits a fame skew by construction — it's easiest to index accounts that are already independently findable. This isn't unique to X/Twitter datasets; the same pattern shows up in TikTok creator indexes and other social-profile datasets built the same way.
How was this compared?
By citing published third-party research on active-account follower distributions alongside our own dataset's documented figures — a cross-dataset comparison, not a fresh random sample we drew ourselves.