X's Attention Economy Is More Unequal Than Any Country's Wealth — Gini Coefficient 0.94

We hold 820,548 public X profiles in a search index, and between them they claim 47.5 billion followers — more than five times the world's population, because most of those followers are the same relatively small set of accounts being counted over and over by everyone who follows them. Sort that pile of followers by who holds it, and the picture is not evenly spread. It is barely spread at all.

A Gini coefficient higher than any country's wealth gap

The Gini coefficient is the standard way economists measure how concentrated something is, from 0 (perfectly even) to 1 (one entity holds everything). Applied to followers across our full corpus, it comes out to 0.94.

For comparison, UBS's 2025 Global Wealth Report puts national wealth Gini coefficients — already a far more concentrated measure than income — at 0.82 for Brazil and Russia, the most unequal major economies it tracks, 0.74 for the United States, and 0.38 for Slovakia, the most egalitarian country in its sample. X's follower distribution among the accounts we index is more concentrated than any of those national wealth gaps. Attention on X isn't just unevenly distributed the way money is — it's unevenly distributed by more.

The top of the pyramid

Here are the fifteen highest-follower accounts in the corpus:

@elonmusk alone holds 240.9 million followers — more than the bottom 293,000 accounts in the entire corpus combined hold between them. Every account on this list is verified, and every one crossed 100 million followers or is closing in on it, in a corpus where the median account has 1,414.

Merging two unequal datasets made a more unequal one

We recently added a new discovery channel, Common Crawl, which finds X accounts by reading what the open web already links to rather than starting from a notability list. It added 324,525 more accounts — ordinary ones, mostly: a quarter as likely to carry X's blue check, nearly half as likely to have a bio, a sixth as likely to clear 100,000 followers. Adding that many "more normal" accounts to the index should, intuitively, dilute the inequality of the whole. It did the opposite.

Curated tier alone (n=496,023)0.93

Common Crawl tier alone (n=324,525)0.93

Combined corpus (n=820,548)0.94

Each tier is less unequal measured on its own than the merged whole. The curated tier — the original Wikidata/GitHub/founder/journalist seed, grown organically since — has a Gini of 0.93. The Common Crawl tier, on its own, also comes out to 0.93, even though it's a completely different population (its own largest account, @NetflixBrasil, tops out at 37.5 million followers, a sixth of @elonmusk's total). Combine them and the Gini rises to 0.94.

This isn't a rounding artifact — it's a textbook between-group inequality effect. The two tiers have wildly different average follower counts (the curated tier alone holds 44.2B of the corpus's 47.5B followers, on less than two-thirds the accounts), so merging them adds a large gap between the populations on top of the concentration within each one. Two moderately unequal groups, stacked on top of each other at very different scales, produce a more unequal combined group than either alone — the same reason global wealth inequality exceeds the wealth inequality of any single mid-sized country, just illustrated on our own dataset in real time.

How concentrated, exactly

Half the corpus has fewer than 1,414 followers. To crack the top 1%, an account needs just over 1.4 million — a number that sounds enormous until you remember the corpus's own top account has 170x that. The distance from the median to @elonmusk isn't linear, or even close to it: it's a straight line on a log scale, the classic signature of a power-law distribution, which is exactly what you'd expect from a network where following is free and attention is the scarce resource.

What this shows, and what it doesn't

What holds up: the corpus's follower distribution is more concentrated than the wealth distribution of any country UBS's 2025 report tracks. The concentration is extreme at every cut — top 1%, top 10%, bottom 50% — and it's driven by a genuine mathematical property of combining differently-scaled populations, not a data artifact.

What limits it:

• This is a seeded, notability-skewed corpus, not a random sample of X's 600M+ total user base. It was built from independent sources — people notable enough for a Wikidata entry, public GitHub developers, startup founders and journalists — plus, more recently, whatever the open web happens to link to. Both discovery methods over-represent accounts that already have some public footprint. A truly random sample of every X account, most of which have a handful of followers and never appear in a search index or a web page's outbound links, would likely show an even more extreme Gini at the bottom (more accounts near zero) but a less extreme one at the top (proportionally fewer mega-accounts) — the net direction isn't obvious without pulling one, which we haven't done.
• The Gini figures here are estimated from grouped data, not computed on every individual record. Our public dataset API caps pagination at 10,000 records per query but returns exact totals for any filter, so we reconstructed the full distribution from 33 follower-count buckets (each with an exact account count) plus exact values for the top 2,000 accounts by followers, which dominate the total-follower sum. We validated the method by reproducing our own earlier finding: the same technique applied only to the curated tier returns a Gini of 0.932, matching the 0.930 we published from a full census of that same population's earlier, smaller state.
• Follower counts are a single-snapshot read, and the corpus keeps growing. Between our last two studies the index grew by hundreds of thousands of profiles in days. The numbers here are accurate as of this crawl; they will already be somewhat out of date by the time you read this, though the shape of the finding is stable across every re-measurement we've done.

None of that changes the headline: on this dataset, at this moment, attention on X is concentrated past the point where any national wealth gap on record can match it — and combining two differently-shaped populations made that concentration worse, not better.

Sources

Methodology

We queried the X users dataset live as of this crawl. The public search API caps pagination at 10,000 records per query but returns an exact total count for any combination of min_followers/max_followers filters, uncapped. We used that to build a 33-bucket, log-spaced histogram of the full follower-count distribution (bucket boundaries roughly doubling from 0 to 250 million), confirming each pass that the buckets summed exactly to the live corpus total with zero drift. We then pulled the top 2,000 accounts by exact follower count via sort=followers_desc — these dominate the total-follower sum and the top-percentile calculations, so we used their real values rather than a bucket approximation; every account below that cutoff is estimated at its bucket's geometric mean, the standard approach for computing inequality measures from grouped/binned data. The Gini coefficient is the discrete trapezoidal-rule estimate over the resulting Lorenz curve.

We validated the method by re-running it scoped to source_tier != common-crawl (via subtraction, the same technique the Common Crawl study used) and comparing the result — 0.932 — against our earlier full-census figure of 0.930 for that same, then-smaller population. The 0.4% gap is consistent with the curated tier's organic growth between the two measurements, not a methodology error.

Want to run your own cut? The X users dataset is queryable directly, the how to scrape Twitter/X guide covers the calls, and pricing has the free tier. See also the rest of this series: the most prolific poster on X is a machine, X's real follow limit is 7,500, a random developer is 5x more likely than a notable person to have a blue check, what X bios reveal, how the Common Crawl tier changed our X corpus, and where X accounts say they live.