Signs your dating app match might be an AI: it avoids video and phone calls, replies instantly at any hour, sounds warm but never mentions anything specific or inconvenient about its real life, and its photos look almost too polished. You can test it by asking about a very recent local event, requesting a spontaneous selfie with something specific, or dropping in an absurd off-script remark — real people stumble; AI usually doesn't.
You matched, you've been chatting for a few days, and the conversation feels good — warm, interested, a little exciting. But something is slightly off. The replies come too fast, the messages are too smooth, and somehow you've talked for a week without learning a single specific thing about this person's actual life. You might be talking to an AI.
This isn't a rare edge case anymore. Research into romance-scam operations suggests AI is already doing most of the conversational heavy lifting — a 2025 academic study (arxiv.org/abs/2512.16280, later presented at USENIX Security 2026) found that 87% of the conversational labor in romance-baiting scams consists of systematized tasks readily susceptible to LLM automation. And it isn't only scammers: some dating apps have been accused of running their own bots to keep users engaged and less likely to cancel subscriptions. A 2025 survey found 60% of dating app users believe they've had at least one AI conversation.
Two Kinds of AI You Might Encounter
Before getting into the signals, it helps to know what you might be dealing with — because the two main scenarios are different.
Platform bots are run by the dating app itself. Their goal is to keep you swiping, messaging, and staying subscribed. They're not necessarily trying to trick you into sending money; the app already profits from your continued engagement. These are the AI encounters where a conversation can feel great for a while, then eventually you notice nothing is leading anywhere.
Chatfishing is when someone creates a fake profile to deceive you personally — using AI to handle the conversation while building enough trust to eventually ask for money, personal information, or something else of value. A human operator may or may not be actively involved; the AI can run much of the conversation on its own. Norton reported blocking 17 million dating scam attacks in the fourth quarter of 2025 alone, a 19% increase year-over-year.
The signals below apply to both. The difference matters mainly when you're deciding what to do about it.
Signal 1: It Won't Get on a Video or Phone Call
This is the clearest single indicator. A real person who is genuinely interested in you will eventually agree to a phone or video call — not necessarily the first time you ask, but at some point, the question of actually talking comes up and gets answered.
An AI bot cannot do a live video call without a human behind it. A chatfisher using AI as a writing tool can do a call, but doesn't want to, because the illusion ends immediately.
Watch for the pattern of excuses. One declined call is nothing. A consistent string of camera problems, bad WiFi, odd working hours, or always "just about to go somewhere" — across weeks of otherwise enthusiastic chatting — is meaningful.
Ask early. You don't need to wait weeks before suggesting a quick call. A real person who is interested but genuinely can't right now will say "soon" and follow through. Watch whether they do.
"Hey, would you want to do a quick five-minute video call this week? Just to actually see each other — no pressure."
Signal 2: Perfect Replies, Any Hour, Instantly
Real people have lives. They reply when they're free — sometimes immediately, sometimes hours later. They type fast sometimes and slow other times. They occasionally misread a question, drift off-topic, or send a message that clearly got written in a hurry. They're tired, distracted, in the middle of something.
AI doesn't have any of that. It's available at 3 a.m. on a Tuesday. It never takes longer than a few seconds. Its messages arrive clean and coherent every single time.
Consistent instant replies at unusual hours are a tell. One late-night chat isn't unusual. A pattern of perfectly smooth, fast responses whenever you message — including 2 a.m., on public holidays, in the middle of workdays — starts to stand out.
The writing never varies. No typos, no autocorrect errors, no trailing sentences that went somewhere and then didn't. Some people do write carefully, but if the style is identical at midnight and 9 a.m. and during their supposed lunch break, that's worth noting.
Signal 3: Warm, Interested, and Weirdly Vague
AI is genuinely good at seeming emotionally present. It reflects your feelings back at you, asks follow-up questions, says the supportive and interested thing. In a dating context, where you're hoping the other person genuinely likes you, this lands well — sometimes very well.
What AI struggles with is specificity anchored in a real life. A real person who's told you they work at a bakery will eventually mention something specific — a burnt batch, a difficult customer, a coworker's name. Their life leaks into conversation in small, real ways.
Look for depth without detail. If your match is warm and engaged but after many conversations you still don't know what their neighborhood looks like, what their apartment is like, or one specific story from their actual week — that vagueness is a signal.
Notice if it never interrupts. Real people interject unexpectedly, circle back to something you said earlier in an odd moment, forget what they already told you, or occasionally say something that doesn't quite follow. AI maintains perfect conversational continuity and never forgets what the "character" has established. That consistency is, paradoxically, a red flag.
Signal 4: Photos That Look Almost Too Good
AI-generated profile photos have improved dramatically. The giveaways of a few years ago — odd fingers, blurred ears, wavy backgrounds — are rarer now. But some patterns remain.
Look for an oddly consistent aesthetic. Real people's photo collections are messy — some from a great day, some not, different lighting, different company. A profile where every photo looks like it came from the same professional shoot, with consistent soft lighting and the same flattering camera angle in every single one, might be AI-generated or curated from someone else's photos.
Try a reverse image search. On a desktop browser, right-click any profile photo and choose "Search image" (Chrome or Edge) or save it and upload to Google Images or TinEye. If the same face appears elsewhere under different names, that's conclusive.
Tests That Can Help Expose an AI
None of these are foolproof, but they create situations AI handles poorly.
Ask about a very recent local event. Something from the last 48 hours in a specific location — a local sports result, a neighborhood news story, anything from that city's local feed. AI has a training data cutoff and doesn't have live local news. A real person will either know it or say they haven't heard. An AI will deflect, give a generic answer, or make something plausible-sounding up.
"I saw something happened near downtown their city yesterday — did you catch that?"
The spontaneous specific selfie. Ask for a photo right now with something specific and unplanned — holding up three fingers, standing next to whatever is on their left, or with their hand on their chin. Make it odd and unscripted. A real person can do this in 30 seconds. A bot cannot. A chatfisher with a stolen photo library cannot pull it off quickly and naturally.
"Random request — can you send me a selfie right now holding up two fingers? I just want to see your face today."
Absurdity and humor. Drop in something genuinely odd mid-conversation: a non-sequitur, a terrible pun, a completely random question about whether they'd rather fight one horse-sized duck or a hundred duck-sized horses. Real people react unpredictably. AI responds smoothly and thoughtfully, treating the odd remark as a reasonable conversational contribution. The absence of genuine confusion or amusement is a tell.
Recall a specific detail unexpectedly. If they mentioned something specific early on, bring it up again in an unusual way a week later. Real people misremember things, get details slightly wrong, or say "wait, when did I tell you that?" AI maintains perfect recall of every detail the character has established.
Why Detection Is Genuinely Hard Now
A 2025 study found that GPT-4.5 passed as a human 73% of the time in structured Turing test conditions. Scientific American has described chatfishing as "the modern Turing test" — a scenario where telling human from machine in conversation is no longer easy or obvious. In a dating context, where warmth and attention feel meaningful and you want the connection to be real, the difficulty is even higher.
This is also why the gut feeling that something is off is worth taking seriously, even when you can't name exactly what it is. The signals described here often register before you can articulate them.
What to Do When You're Suspicious
Stop sharing personal information right now. Don't give out your phone number, home address, workplace, or anything financial until you've resolved the doubt. This costs you nothing if the person is real; it protects you if they're not.
Try the tests together. A video call request, a spontaneous specific selfie, and a local-event question — all three at once are hard to fake simultaneously. Real people clear them easily. AI-driven conversations can't clear all three.
Report it. If you're on a paid app and believe you've been matched with a platform bot, report it to the app's support team. In the US, you can also file a complaint at reportfraud.ftc.gov.
Trust the mismatch. If the conversation feels emotionally warm but nothing in it is grounded in real, specific detail — no inconvenience, no specifics, no evidence of an actual life behind the messages — that mismatch is the signal itself.
One scope note: if things have moved into money requests, emergency stories, or anything financial, that goes beyond detecting AI into romance scam territory. The guide on AI romance scams on dating apps covers that situation specifically.
What to Try Next
If you want to apply the same kind of detection to writing in general, How to Tell If Something Was Written by AI walks through the signals across any kind of text. And if you've been on a video call and wanted to know whether the face on screen was real, How to Spot a Deepfake Video Call covers exactly that test.
