← Verikast
Verikast Measurement Note 01/ 20 November 2026/ 3,822 posts measured

The Phantom Audience

We measured what the sixteen largest accounts on Farcaster actually get when they post. Between them they hold 2.1 million followers. The median post earns two likes.

0.03 likes per 1,000 followers — median across the 16 largest accounts we could sample (0.025 before rounding). A healthy social account returns 5–20.

If you pay a creator because they have 100,000 followers, you are buying a number that has come loose from the thing it is supposed to represent. This note is the measurement behind that claim, with the method and the limits written down so you can argue with it.

What we measured

Method

Four samples, all taken on 20 November 2026, all from public APIs with no special access — 3,822 posts in total:

The platform-wide sweep and the creator sweep agree: the median Farcaster post earns one like in both.

For every account we took the median like count, not the mean. One viral post should not be allowed to speak for an account's normal day, and on these platforms the mean is almost entirely one or two outliers.

Likes per 1,000 followers — measured vs. expected
expected range for a healthy account: 5 – 20 0 5 10 15 20 likes per 1,000 followers all 16 measured accounts
Every account we measured sits on the axis origin. Not low — indistinguishable from zero at the scale a healthy account occupies. The largest account in the sample, at 482,986 followers, returns a median of 18 likes.

The sixteen accounts

Farcaster · above 20,000 followers · 25 most recent casts each

Accounts anonymised. The finding is about the platform's follower graph, not about any individual — none of these people bought anything, and saying otherwise would be an accusation the data does not support.
AccountFollowersMedian likesPer 1,000 followers
A482,986180.04
B433,61420.00
C146,997150.10
D132,82210.01
E108,73320.02
F103,82920.02
G103,62540.04
H97,85530.03
I89,37710.01
J71,50510.01
K70,65910.01
L66,07730.05
M58,27910.02
N57,92320.03
O43,75030.07
P34,97030.09
Median93,61620.025

Farcaster is not the only option, and that matters

Same method, both platforms, 40 creators each

We ran the identical measurement on Bluesky. The accounts there are considerably smaller by follower count, which makes the result more interesting, not less.

Farcaster · 1,000 casts
≥10 likes17.7%
≥208.9%
≥306.0%
≥502.6%
≥1000.8%
≥2000.2%

Median post: 1 like

Bluesky · 822 posts
≥10 likes40.0%
≥2030.7%
≥3024.7%
≥5017.3%
≥1009.9%
≥2003.9%

Median post: 5 likes

Bluesky creators clear 100 likes at twelve times the rate of their Farcaster counterparts, while holding a fraction of the follower counts.

Did we check our own instrument

Verification

Yes, because a measurement this lopsided is more likely to be a broken tool than a broken platform. Two checks:

1. Are the like counts real?

We took the reported like count for individual posts and independently fetched the actual list of accounts that liked them. Reported 32, fetched 31. Reported 51, fetched 45. The counts are real; the API is not under-reporting engagement into existence.

2. Can our detector tell anything apart, or does it call everything clean?

Our first sweep of 32 established accounts returned 100/100 for all 32, which looked like a rubber stamp. It was a sampling error on our side: we had selected only the highest-follower accounts from curated channels, which is precisely the population expected to be clean. Re-run against accounts advertising giveaways and follow-for-follow:

Positive control — Bluesky. Score is driven by the share of engaging accounts that are empty-bio, low-follower profiles.
Account typeBot-profile engagersScoreLabel
Giveaway account65.5%39Heavily inflated
Giveaway account21.1%93Authentic
Whitelist account18.4%96Authentic
Established creators (n=32)0–1.4%100Authentic
The instrument separates. It spans 39 to 100 and the separation tracks a cause we can point at, rather than a score we cannot explain.

What this does and does not mean

Limits

Read this before quoting the number

A low likes-per-follower ratio is not proof that anyone bought followers. Farcaster's follow graph carries years of mass-follow behaviour and dormant accounts from its invite era, and a dormant follower is indistinguishable from a purchased one at this level of measurement. What the ratio establishes is narrower and still decisive: the follower number does not predict reach.

Three further limits we would rather state than be caught on:

The practical version

If you pay creators

Follower count is the number in every media kit and the one nobody re-checks. On the platform we measured, it has almost no relationship to how many people see the post you are paying for. The fix is not complicated and does not need our permission:

That third check is the one that needs a tool, and it is the one we built. The first two take a minute and will catch most of it.

Reproduce this

Every figure here came from public endpoints with no paid tier: Neynar's channel and user-cast feeds for Farcaster, the public AT Protocol app.bsky.* endpoints for Bluesky. We are happy to be corrected by anyone who runs it and gets something different.