How much of X is actually bots? What the research actually says

Ask this question and you'll get wildly different answers depending on who you ask — 5%, "9 to 15%," "a quarter," "most of it, honestly." That spread isn't people making things up. It's a real methodology problem: counting accounts and counting activity give you completely different numbers, and most of the scary-sounding figures you've seen quoted are actually the second kind wearing the first kind's clothes.

The account count versus the activity count

X's own disclosures have put the bot share at around 5% of monetizable daily active users — a figure that became famous in 2022 when it was at the center of a very public dispute over the platform's acquisition. Independent researchers looking at the same platform, using their own detection models rather than the platform's self-reporting, have consistently landed higher: work associated with groups like the Stanford Internet Observatory and Carnegie Mellon has put automated-behavior accounts at somewhere around 9 to 15% of active accounts, with that share climbing sharply — sometimes to 15-44% — inside specific, high-heat conversations like politics or major entertainment events.

Then there's the much bigger number people actually remember: "bots make up most of the traffic." That's usually a real stat, just measuring something else. Pew Research Center's widely-cited 2018 analysis found that around two-thirds of tweeted links to popular websites were shared by suspected automated accounts — but that's a measure of link-sharing activity, not a share of accounts. A relatively small number of extremely active bot accounts can dominate the traffic of a platform where the account-level bot share is genuinely in the single or low double digits (Pew Research Center, "Twitter Bots: An Analysis of the Links Automated Accounts Share").

Why the estimates disagree so much

Academic surveys of the bot-detection field are pretty blunt about this: different studies use different definitions of "bot," different sampling windows, and detection models trained on different labelled datasets — and all of that changes the number you get out the other end, sometimes by an order of magnitude, without anyone involved being dishonest (arXiv, "Twitter Spam and False Accounts Prevalence, Detection and Characterization: A Survey").

Why this actually matters, beyond trivia

It matters because "how many bots are there" quietly turns into "how much can I trust what I'm seeing." If a genuinely small share of accounts is responsible for a genuinely large share of the loudest activity, the practical takeaway isn't "don't trust anything," it's that volume and popularity are exactly the things automation is cheapest to fake — see the retweet amplifier and the reply farm for the two most direct examples of that gap between "looks popular" and "is popular."

What to actually do with this

Treat any single bot-percentage figure — ours included, if we ever quote one — as a snapshot of one methodology at one moment, not a settled fact. What's far more useful than a headline percentage is being able to look at one specific account and actually judge it (see our own no-tools checklist), because whatever the platform-wide number is this month, the accounts you're actually interacting with are the ones that matter to you.

Sources:

• Twitter Bots: An Analysis of the Links Automated Accounts Share — Pew Research Center

• Twitter Spam and False Accounts Prevalence, Detection and Characterization: A Survey — arXiv