Fake accounts

The label covers several different things, and the response depends on which one applies. Automated accounts (bots) post at machine speed. Sock puppets are run by people posing as several independent individuals. Impersonation accounts copy a real person or organization; see brand impersonation. Spam and scam accounts are created in bulk to sell, phish or defraud. Some fake accounts are aged and maintained for months so they look real when they are activated.

Researchers look at several groups of signals: account metadata such as creation date, handle patterns and profile images; behavior such as posting frequency, timing and the ratio of original posts to reposts; network structure such as who follows whom and which accounts always amplify one another; and content such as repeated text or links. No single signal is definitive, and legitimate users can look odd on any one of them. Combining signals and reporting confidence is more reliable than a single flag. See bot detection for a fuller treatment.

Platforms publish periodic reports about removing fake accounts under their own policies. Those reports describe what each platform found and how it defines the term, and they are not directly comparable between platforms.

Fake accounts change how to read a conversation. A wave of criticism may involve few real people, and a wave of praise may be purchased. They also create direct risks: scam accounts that use a company's logo to defraud customers, and hostile accounts that harass staff. A team should be able to say what share of a conversation comes from accounts that look inauthentic, and how confident the assessment is, before it responds.

A useful report to a platform lists the account handles, the behavior that violates policy, sample links and the dates observed. For an internal audience, analysts describe how many accounts were reviewed, which signals were used, how many met the threshold, and what could not be assessed. A stated sample size and method keep a finding from being read as a claim about all accounts in a conversation.

Researchers outside a platform see only public data. They cannot see device information, sign-up details or private messages, all of which the platform can use. For this reason, public findings about fake accounts are best stated as patterns consistent with inauthentic behavior, with a confidence level and the method used. Claims that a specific person or organization operated the accounts need evidence beyond behavior on the platform.

Accounts with no profile picture or few followers are not necessarily fake; many real users lurk. A large follower count is not proof of authenticity either, since followers can be bought. And removing fake accounts does not resolve the underlying narrative. A false claim may continue to spread among real users after the accounts that started it are gone.