Guide Β· Media Literacy
Not every account arguing with you is a person. Some are automated, some are run in bulk to manufacture the look of a crowd, and in 2026 the well-made ones write like humans. Here's how to read an account for what it actually is β the profile signals worth checking, the follower math that gives bots away, and the one test that still works when the writing doesn't give it up.
It helps to name the different things hiding behind the word "fake." A bot is automated β software posting or replying on a schedule. A sock puppet is a fake persona a real person operates to seem like someone they're not. And a coordinated network is many such accounts moving together to make a fringe view look popular β what Meta calls coordinated inauthentic behavior in its regular takedown reports. They overlap, and the tells below catch all three.
The honest framing first: no single sign is proof. A generic username, a new account, a burst of activity β each has an innocent explanation on its own. What's hard to fake is the whole picture at once. So the method isn't to find one smoking gun; it's to stack a few cheap checks until a real person clearly emerges, or clearly doesn't.
Before you engage with what an account is saying, spend ten seconds on who it claims to be. The bio, the avatar, the handle, and the join date together tell you more than any single post. A blank bio, a default or stock photo, a name-plus-numbers handle, and a creation date measured in days are individually forgivable and collectively a strong signal.
Then scroll the history. Real accounts accumulate the texture of a life β inside jokes, replies to the same handful of friends, posts anchored to real dates and places. An account built to push a message tends to feel flat and recent by comparison. These are the patterns to weigh:
Auto-generated accounts still lean on the default suggestions the platform hands out at signup β a common first name welded to a long string of digits, or a jumble of letters no person would choose. On its own it proves nothing; plenty of real people keep the name they got assigned in 2011. But paired with an empty bio and a blank or stock avatar, it's the cheapest first flag there is.
Open the account and scroll. A real person leaves a trail β old photos, replies to friends, posts that reference things happening in their actual life. A puppet account made to amplify one message tends to have a thin, recent history that starts abruptly and points in a single direction. Check the join date: an account created three weeks ago that posts like a seasoned commentator is worth a second look.
Bots built to inflate reach follow aggressively and get little back β a following count in the thousands against a handful of followers, or the mirror image, a big follower count paired with engagement that never rises above a trickle. When the ratios are lopsided in a way a normal social life wouldn't produce, treat the account as a number, not a neighbor.
Automation shows up in the clock. An account that posts around the clock with no gap for sleep, fires off dozens of replies a minute, or reliably wakes at 3 a.m. local time is either scheduled software or run from a very different time zone than it claims. Human posting is bursty and irregular; machine posting is relentless and even.
A single-topic account that only ever posts about one candidate, one product, or one grievance β and always with the same three hashtags β is the classic shape of an amplification account. The strongest version of this tell is copy-paste: several accounts posting word-for-word identical text within minutes of each other. That's not coincidence; it's the fingerprint of a coordinated push.
Numbers lie less than profiles do. Follower-to-following ratios that no real social life would produce β thousands followed, a dozen following back β are a familiar bot shape. So is the opposite: a large follower count sitting on top of posts that draw almost no replies, likes, or shares. When engagement runs far below what an audience that size should generate, a good chunk of that audience is probably purchased or automated.
You don't need a tool for the rough version β eyeballing the ratios catches the obvious cases. For a closer look at accounts on X, Indiana University's Botometer scores likely automation, though be clear about its limits: after platform API access was cut, it now runs in an archival mode on pre-2023 data, and its own makers warn it struggles against the newest AI-driven accounts. Read any bot score the way you'd read a detector score β one weak vote, not a verdict.
Open the account's timeline and look at the timestamps, not just the content. Humans post in bursts with long silences β sleep, work, life. Automation posts evenly and endlessly: hundreds of items a day, replies fired seconds apart, activity that never pauses for a night. A cadence that's too regular, too fast, or wide awake at 4 a.m. in the time zone the account claims is one of the clearer signs you're looking at a script rather than a person. Cross-checking the local time of an account's most active hours against where it says it lives often does more than any single post ever could.
The most revealing tell isn't inside any one account β it's the resemblance between many. When a claim suddenly floods your feed, take one distinctive sentence from a post and search it as an exact phrase. If a dozen accounts are posting word-for-word identical text within the same short window, you're not watching a groundswell; you're watching a campaign. Identical wording, the same image, the same hashtag stack, all landing in a tight burst, is the signature pattern researchers use to flag coordinated inauthentic behavior.
Look, too, for the accounts that only ever exist to boost each other β replying, quoting, and amplifying within a closed loop while never engaging with the wider platform. A cluster of single-topic accounts that talk mostly to one another, and all appeared around the same time, is a network doing a job, not a community forming on its own.
When the writing is too fluent to judge β and with AI-generated replies, it often is β stop reading and start probing. Ask a specific, personal question that only a genuine participant could answer: a detail about the place they claim to be, a follow-up that requires actual context. Automated and mass-run accounts tend to dodge, change the subject, or repeat their talking point rather than answer. Machines are fluent; they're rarely accountable to a specific fact.
And check the face. Drop the profile photo into reverse image search β if it's a stolen picture of a real stranger, a stock-library headshot, or a synthetic face from a generator, that's often the fastest confirmation of all. Our reverse image search guide walks through which engine finds the original copy quickest.
A new account isn't automatically a fake
People join platforms every day, and everyone's account was three weeks old once. Newness plus a generic handle plus single-topic posting plus copy-paste text is a pattern. Newness alone is just newness. Weigh the signals together β no single one carries a verdict, which is the whole point of running several.
The comments read like a real person wrote them
They might have been written by a model. As the Global Investigative Journalism Network notes, large language models let today's bots produce fluent, context-aware replies that sail past the old 'stilted grammar' test. Fluency is no longer evidence of a human. What machines still handle badly is a specific, personal follow-up β so shift your attention from how the writing reads to whether the account can respond to something only a real participant would know.
It has a verified badge
A checkmark used to signal that a platform had confirmed an identity. On several networks it now signals a paid subscription and nothing more. A badge is a weak positive at best; it doesn't clear an account, and coordinated networks have bought them in bulk. Read the account, not the sticker.
The profile photo looks like a normal person
Run it through reverse image search anyway. Fake accounts routinely use stolen photos of real strangers or faces generated by sites like This Person Does Not Exist. If the same picture belongs to a stock library, a dentist in another country, or turns up nowhere at all, the friendly face is a mask. Our reverse-image-search guide covers which engine to reach for.
You can't see inside β the account is locked
A private or restricted account you can't inspect can't be judged on its posting history. That's not proof of anything, in either direction. Fall back on what you can observe: how it showed up in your mentions, whether identical wording appears across other accounts pushing the same thing, and whether anything it claims can be corroborated elsewhere.
No one tell settles it β not the handle, not the bot score, not how the comments read. What holds up is the stack: read the profile, run the follower math, watch the rhythm, search for copy-paste twins, and test the account with something only a real person could answer. A genuine account clears all five without effort. A fake one has to survive every layer, and it usually trips on at least one. And when an account can't be resolved, the safe move is the same as with any unverified thing online: don't treat it as a person, and don't let it decide what you believe.
Bot networks exist to spread specific claims. If the same message is flooding your feed, someone may have already checked whether it's true. FAXTR searches 100+ fact-checking organizations across 11 languages in one box β free, no login β so you can see if that claim already has a published verdict.
Go to the verifier β