Analysis: 44,058 randomly selected public Twitter accounts that have tweeted in the past 90 days and could resemble mDAUs shows 19.42% are likely spam or fake
SparkToroRand Fishkin
Context & Ripple Effects
SparkToro's random-sample audit of 44,058 active public accounts lands in the middle of a running fight over how big Twitter's real audience is. The company had already conceded it overstated mDAUs by up to 1.9M per quarter because single users held multiple accounts, and it publicly insists false and spam accounts run well under 5% — a figure it defended with daily removals of more than 1M spam accounts.
An independent 19.42% estimate, roughly four times Twitter's ceiling, matters because ad revenue is billed against those user counts. Earlier academic work already showed bots punching far above their weight in what spreads on the platform — one analysis found they were only 6% of accounts but drove 34% of shares of low-credibility articles.
First-order effects
Advertisers buying against Twitter's reported mDAU figures face a direct measurement conflict: if SparkToro's sample holds, the billable audience base is materially smaller than the official number implies.
Twitter's sub-5% self-assessment loses its monopoly on the question — buyers now have a third-party methodology to cite when disputing the metric.
Second-order effects
Independent audits like this force platforms toward verifiable disclosure — expect pressure for external access to data or third-party attestation rather than self-reported percentages.
If fake-account share becomes a priced-in risk, effective CPMs on Twitter inventory come under downward pressure relative to platforms that can demonstrate cleaner audiences.
Third-order effects
The pattern points toward identity verification becoming an advertising prerequisite: if audience counts can't be trusted, ad markets will demand some form of proof-of-personhood behind the accounts being billed.
Platform-reported engagement metrics generally lose standing as a currency, pushing the industry toward audited or independently measurable standards.
The trend: Social platforms are moving from self-reported user counts toward externally audited, identity-verified audience metrics as the basis for ad pricing.
On Friday, @ElonMusk said his deal to acquire Twitter was on hold pending an analysis of “spam/fake” accounts. We've now completed an analysis of 44,058 public Twitter accounts who've tweeted in the last 90 days. 19.42% are likely spam or fake. Details: https://sparktoro.com/...
There's just one issue that sticks in my craw. If Twitter successfully IDs 95% of the millions of spam accounts in analyses like ours, why leave them active on the platform? He says they remove 500k/day. Why are all these obvious spam accounts still here for us to catch?
Staggeringly high since this is of only accounts that have tweeted in the past 90 days. I suspect if you consider also accounts that haven't tweeted, that number skyrockets. And if you filter by certainly industries, it becomes mostly spam/bots. https://twitter.com/...
@paraga Parag, u don't need to convince us, we are well aware that Twitter is full of bots! You keep trying to justify but that's the truth Here's an analysis of 44,058 public Twitter accounts who've tweeted in the last 90 days. 19.42% are likely spam or fake https://sparktoro.co…
In the past three weeks, the general informed take on Musk/Twitter has morphed from “Why did they just take $54.20 without a fight?” To “Oh that's why they took $54.20 without a fight.”
@elonmusk ... I encourage you to read the full report: https://sparktoro.com/... I tried not to make it too dry, even in the dry bits. Huge thanks to @caseyhen, @marcmims, and @amandanat (who edited).
Wtf nearly 1/5 of all Twitter users are bots according to this analysis. As ai is developed even further it will be almost impossible to discern a bot from a real person. Crazy https://twitter.com/...
This is fascinating. The only charitable interpretation I can come up with for twitter's number is that the “mDAU” figure is post-spam-analysis (though they obv don't kick that spam off the platform) https://twitter.com/...
This study is very very good. And not for the debate opened by Muck, but also (and especially) for us SEOs, who share so much here. https://twitter.com/...
Over the weekend, we provided @sparktoro with a random sample of ~44,000 Twitter accounts for a fascinating analysis. Loved teaming up with @caseyhen and @randfish for this. They did the heavy lifting. https://twitter.com/...
65.6M (or 70%) of @elonmusk 93,4M followers are unlikely to be authentic active users who see his tweets 6.7M (or 23%) of @elonmusk 26,8M active followers are likely fake or spam 19% of active accounts on Twitter are likely spam or fake accounts https://sparktoro.com/...