The myth that everyone on the other side is a bot
It is one of the first conclusions people reach on random video chat, and it is usually wrong. A few flat conversations in a row, a couple of people who said hi and nothing else, one person who did not react to a direct question — and the explanation assembles itself: none of them were real. It is a satisfying theory because it explains a bad evening in one move, and asks nothing further of you.
The trouble is that it is expensive. Deciding in advance that nobody is real changes how you open, which makes conversations worse, which confirms the theory. Meanwhile the genuinely useful skill — telling a nervous stranger apart from a scripted account, and knowing what to do about the second — never develops.
This is a testing procedure. Ten steps, most taking a few seconds, that separate "a real person having an awkward minute" from "something running a script". At the end there is what to do in the second case, which is short and does not involve arguing with anyone.
1Name the specific thing that triggered the suspicion
Before testing anything, identify what actually made you doubt it. Was the reply too fast, too slow, too generic, unrelated to what you said, or identical to something you heard ten minutes ago from someone else? Each of those points at a different explanation, and lumping them together as a feeling makes the rest of the process impossible. Pick the one concrete thing.
What this tells you: Almost every one of those symptoms has a boring human cause available, but they are different causes — lag, translation, nerves, distraction, a stock opener. You cannot rule any of them out while the observation is still a vague sense that something was off.
2Rule out your own audio before testing them
If your suspicion is that they ignored what you said, first confirm you were audible. Ask a question that can be answered with a nod or a wave rather than words, and watch the face. If they respond to visible gestures but never to anything you say out loud, the problem is your microphone or their speakers, not their existence.
What this tells you: A person who cannot hear you behaves exactly like a script that is not listening: they carry on with their own opening, they answer questions you did not ask, and they eventually give up. It is a very common false positive, and it takes about five seconds to eliminate.
3Give it ten seconds before judging anything
Hold the verdict for the first ten seconds of a match. Say hello, wave, and let them arrive. Many people spend that time working out whether their camera is on, whether they are muted, and what to say. Judging inside the first three seconds means you are almost always judging someone at their most flustered.
What this tells you: The opening seconds have the highest concentration of behaviour that looks artificial and the lowest information content, because nobody has settled yet. Waiting costs you ten seconds and clears a lot of the false alarms without any testing at all.
4Ask for a physical action in the moment
This is the strongest single test and it takes four seconds. Ask them to hold up three fingers, wave with their left hand, or touch their nose. Pick something arbitrary and specific — the arbitrariness is the point, because it cannot be anticipated. Then watch for the action to arrive after your request, in the order you asked for it.
What this tells you: A live person can perform an unpredictable physical instruction on demand; pre-recorded or scripted material cannot, because whatever it plays was chosen before you spoke. Being contingent on your request is the thing you are testing, not speed — a delayed but correct response is still a correct response. Vary the instruction if you use this often, since the value is entirely in it being something nobody could have prepared for.
5Ask about something visible behind them
Comment on something specific in their frame and ask about it: a poster, a guitar, the colour of the wall, a cat crossing the background. Then ask a follow-up that requires them to look at it — what is it, where did it come from, is that yours. You are asking them to interact with their own surroundings in response to your curiosity.
What this tells you: Anything scripted has no relationship to what is actually behind the person on your screen, so this produces either a non-answer or a generic pivot back to the script. A real person will turn their head, which is a response no amount of clever text generation produces. It also has a side benefit the other tests lack: asking about someone's guitar is a genuinely good opening, so the test costs you nothing even when it passes.
6Ask about right now where they are
Ask what the weather is doing outside their window, what time it is where they are, or what the noise in the background was. Keep it to the present moment rather than about them as a person, and keep it small enough that answering is easy. Then compare the answer to what you can see and hear on your own screen.
What this tells you: Present-moment questions cannot be answered from a stock of prepared material because the answer changes hourly, and they cross-check against evidence you already have — the light in their room, the sounds behind them. A mismatch between the answer and the picture is far more informative than the answer alone.
7Break the rhythm on purpose
If the exchange feels like it is running on rails, interrupt it. Change the subject mid-sentence, ask something completely unrelated to what you were discussing, or answer a question with a question. A real conversation absorbs the disruption — the other person laughs, says "wait, what?", or follows you across. A scripted one carries on toward wherever it was heading.
What this tells you: Scripts are built around a path, and the reliable tell is not the quality of individual lines but the refusal to be diverted. Watching whether a conversation can be knocked off course is a better test than analysing any single response for realism.
8Watch the timing of reactions, not their content
Pay attention to when a laugh, a nod or a smile lands rather than whether it looks convincing. A real reaction arrives on the punchline, on the pause, on the moment you fumble a word — locked to your timing, sometimes overlapping you. A reaction that lands two beats late, or in a spot where nothing happened, is out of sync with the conversation you are actually having.
What this tells you: Timing is much harder to fake than expression and much harder for you to talk yourself out of noticing. It also catches the case a gesture test misses: a real person paying no attention at all, who is technically present but reacting to something else entirely.
9Switch the question from "is this a bot" to "is there an ask"
At some point stop testing and watch instead for what is being requested. The patterns worth noticing are a request for money in any form, an invitation to continue on another app or through a link, a page asking you to sign in or verify something, or pressure to hand over contact details early. Any of those matters far more than whether a person is real, and none of them requires you to settle the question first.
What this tells you: Automation is not the risk; the ask is. A scripted account with nothing attached wastes thirty seconds of your evening, while a real person running a scam can cost considerably more, and the tests earlier in this list would clear them without hesitation. The ask is the signal that actually predicts harm. It is also the one that survives being wrong about everything else: you never have to settle what is behind the camera in order to decline what is being requested in front of it.
10Do not argue — leave, block, report
If you conclude an account is scripted or is working an angle, do not announce it and do not debate it. Hit Next or end the call. If there was an ask attached, open their profile and use block, then report with the reason from the list the app provides — the submission screen notes that reports are reviewed within 24 hours and that false reports can restrict your own account, so choose the reason that genuinely fits.
What this tells you: Arguing keeps you in the conversation, which is the one thing that serves the other side and never serves you. Reporting is also the only step that goes anywhere beyond your own screen, so a skip protects you for thirty seconds while a report is what removes the account from everybody else's evening.
Why real people look scripted so often
Almost every behaviour that reads as robotic has a mundane human explanation, and they cluster in the first thirty seconds because that is when people are most uncomfortable. Nerves produce short, flat answers. A translator produces long pauses followed by oddly formal sentences. A weak connection produces delayed reactions that land in the wrong place. Half-attention — someone cooking, someone with a show running on another screen — produces exactly the vagueness people file under "not real". And someone who has opened twenty conversations that evening will have a well-worn opening line, because saying something fresh twenty times is genuinely hard.
None of that is exotic. It is what a normal person looks like at the start of a conversation with a stranger, viewed through a small window with no context attached.
What these tests cannot tell you
The limits are worth stating plainly. The gesture check in step four is the strongest test on the list, and it still only establishes one thing: a person was there at that moment and was paying enough attention to follow an instruction. It says nothing about whether they are who they appear to be, whether the conversation is going anywhere, or whether the friendly opening is leading toward a request later.
Timing tests have the opposite weakness. A long-distance connection, a translation step or a moment of buffering will all delay a genuine reaction into the wrong place, and someone half-watching a show will produce reactions that never land right at all despite being entirely real. Reading late responses as evidence of automation will make you wrong about ordinary distracted people constantly.
Treat all ten as a quick way to stop wondering, not as a verdict. Passing them is weak evidence that the conversation is worth continuing; failing them is a reason to move on, which was already available to you for free at any point. The one thing worth acting on decisively is in step nine, and it does not depend on any of the earlier results.
Where people get stuck
- "They never spoke, only typed." Extremely common and entirely human. People type when a housemate is asleep, when they are somewhere public, when their microphone is not working, or because they are not ready to be heard yet. Test with the gesture check rather than treating silence as proof.
- "Their English was perfect but the answers were generic." That combination usually means a translator or a phrasebook, not automation — machine translation produces fluent, slightly bland sentences with the specifics stripped out. Ask something concrete and physical and watch the answer get more specific.
- "They ignored a direct question." Check that the question was heard before deciding it was ignored. A microphone problem on either side produces exactly this pattern, and it is one of the most common invisible causes — the usual suspects are covered in the problems with laptop built-in microphones.
- "I asked them to wave and they did, but slowly." Latency. On a long-distance connection a round trip takes a noticeable moment, and someone who is also processing a language will take longer still. You are testing whether the response is contingent on your request, not how fast it arrives.
- "Several in a row felt fake." Runs happen, and a string of similar experiences does not necessarily mean anything has changed. If the pattern persists across an entire session, look at the variables you control — the room you picked, the time you are on, and your own opening.
- "I am still sure it was a bot." You may well be right, and it does not change what you do. Block, report if there was an ask attached, move on. The conclusion is only worth reaching if it changes an action, and there is exactly one action — which is why it is not worth spending a second match trying to prove.
The part of the myth that is not a myth
Scripted and scam accounts exist on every platform where strangers meet, and pretending otherwise would be useless. The important correction is not that they do not exist — it is that "is this a bot?" is the wrong question. What matters is whether there is an ask attached: for money, for a link to somewhere else, for a login somewhere, for your details. A charming real person making that ask is a bigger problem than a clumsy script making no ask at all.
So run the tests for your own curiosity, but make your decisions on the ask. When one arrives, the answer is the same regardless of what is behind the camera — no, and then leave. That sequence is set out step by step in what to do when someone asks you to do something.
The better use of the same attention
The energy people spend on bot detection is far better spent on reading whether a real conversation is working, which is a much more useful signal and much more often present. If you want the other half of this — the signals that somebody is actually engaged rather than merely present — that is reading whether someone wants to keep talking. And if flat openings are the recurring problem, starting inside a room where people arrived for the same reason changes the pool you are testing in the first place: the open ones are on video chat rooms.