A simple test for where the line is
Ask whether the other person is still forming an accurate impression of you. AI that improves how clearly you express something true, fixing grammar, cutting a rambling bio, suggesting you mention the hobby you forgot, keeps the impression accurate. AI that invents the substance changes who they think they are talking to: a wit that is not yours, opinions you do not hold, a personality generated to match their profile. Nobody has to disclose spellcheck. Everyone should worry about outsourcing the part of the conversation that is supposed to reveal them.
Profiles: usually fine, with limits
An AI-assisted bio is close to universally accepted now, and app makers largely encourage it, several ship prompt suggestions and photo-picking tools in-app. It stays honest as long as the facts are yours and the voice is recognisably how you speak, because the person will meet that voice on the first date. Two things break it: inventing biography, jobs, travel, languages, interests you do not have, and AI-generated or heavily AI-retouched photos, which is the one area where users draw a hard, near-unanimous line. A face that does not match the one that walks into the bar is catfishing regardless of what generated it.
Messages: the genuinely contested ground
Opinion splits sharply here. Using AI to get past a blank first message, or to find a reply when you are stuck, reads to most people as nerves rather than deception. Running the entire conversation through a model, every reply generated, tone tuned to what the model predicts will land, is a different act, because the whole point of chat is deciding whether you want to talk to this person. The practical rule: use it for the first line and the stuck moments, write the rest yourself, and never let it answer questions about your feelings, your history or your intentions.
Where AI becomes a safety problem
The same tools have industrialised romance fraud. Generated faces defeat old reverse-image checks, translation makes convincing contact possible in any language, and a scammer can now run hundreds of conversations at once with consistent personas. Protect yourself with things AI still struggles to fake in real time: a spontaneous video call, a specific unplanned request, references to something only a real local would know. Verified badges help. So does the oldest rule, no money, ever, to someone you have not met.
“Using AI to sound like yourself is fine. Using it to sound like someone else is the problem.”
What to disclose, and how
You do not owe anyone a footnote on every edited sentence. What is worth saying, lightly and early, is if AI wrote something they are reacting to as though it revealed you, 'honestly, I had help polishing that bio' costs nothing and defuses the discovery later. Never let AI answer for you in an emotionally significant exchange, and never use it to imitate a specific person. If you would be uncomfortable telling them on the first date how a message was written, that is the answer.
How the apps are responding
Platforms are moving in two directions at once: shipping their own AI features, photo selection, prompt help, conversation starters, while tightening rules against AI-generated imagery and automated messaging. Expect verification to become more prominent, detection of generated photos to improve, and bans to focus on synthetic faces and bulk automation rather than on someone using a model to write a better bio. The honest reading is that assisted writing is now normal, and synthetic identity is what gets you removed.