How to tell if you're talking to a bot
Chats, DMs, support windows and even phone calls: more and more of the "people" on the other end are AI. Here is how to tell, and what to do about it.
Nine signs
- It answers instantly, at any hour, with long, well-formed paragraphs. People pause, type, delete and send in bits.
- It's always agreeable. Bots rarely disagree, sulk or say "no, that's wrong".
- It never quite answers the question but says something reasonable-sounding near it.
- No real "now". Ask what the weather is doing or what they had for lunch and you get something generic ("it's a lovely day!").
- Perfect spelling and punctuation, every time, in a casual chat.
- Too-tidy structure: bullet points, bold headings and a neat summary, in a DM.
- It forgets. Details you mentioned five messages ago vanish, or it contradicts itself.
- Favourite chatbot words: "delve", "absolutely!", "great question", "I hope this helps". See signs of AI writing.
- It steers. Every reply nudges toward a link, an app, a purchase or a different platform.
None of these proves anything on its own. Some people write tidy paragraphs at 3am. It's the pile-up that counts.
By situation
Each place bots turn up has its own tells. Pick yours:
- Dating apps: instant charm, pushing to WhatsApp. Is your match a bot?
- A "wrong number" text that keeps chatting. Wrong number texts
- Replies and DMs on social media. Bot accounts on social media
- Support chat going in circles. How to reach a human
- A panicked call in a family member's voice. Voice clone scams
- A recruiter with a dream job by text. Fake recruiters
- An email that wants you to act now. AI phishing emails
- An online game opponent. Is my opponent a bot?
- A friend whose texts suddenly got very polished. Friend texting with ChatGPT?
- Reviews that sound like ads. Fake and AI-written reviews
- A photo or video that seems too much. AI images and deepfake videos
Three quick tests
1. Ask about right now
"What can you hear right now?" "What's the last thing you ate?" People give lazy, specific, slightly boring answers. Bots give plausible ones.
2. Break the pattern
Ask something with a twist: "Reply with only the third word of your last message." "Answer in exactly four words." Many bots follow perfectly; many people say "what? lol".
3. Ask it straight
"Are you a human or an AI?" Plenty of customer-service bots must say they're automated when asked. A bot that's been told to pretend may dodge with a joke, which is itself a clue.
Practise on a real bot: play today's round →
On the phone: voice clones
AI can now copy a voice from a short clip. If "a relative" calls in a panic asking for money, hang up and call them back on the number you already have. Agree a family code word that only you know; a cloned voice won't know it.
What to do if it's a bot
- Don't share codes, passwords, bank details or ID, whoever it is.
- On support chats, type "agent" or "human" to be passed to a person.
- On social media and dating apps, report the account. See dating app bots.
Questions
Can a bot pass for a human?
Often, in short text chats. In a 2025 UC San Diego study, people picked GPT-4.5 as the human 73% of the time when it was told to play a persona.
Is it rude to ask "are you a bot"?
Not really. A person will laugh it off; a business bot will usually say it is automated.
Do AI detectors work on chat messages?
Not reliably, especially on short messages. Use the conversation itself: ask about right now, break the pattern, and watch for steering.