AI on Your Dating Profile: What to Hand It, What to Keep
Three things: structure, compression, and diagnosis. A model will reorder your paragraphs so a stranger can follow them, cut four hundred words to a hundred and forty without losing a fact, and tell you flatly that your second paragraph is doing no work. Those are real editorial skills, and most people writing about themselves have none of them at 1 a.m.
It is bad at exactly one thing, and it happens to be the thing a profile lives on: the specific, slightly odd, true detail that only you would have written. A growing 26% of singles now use AI somewhere in their dating life, a 333% jump year over year, per the 14th annual Singles in America study (Match and the Kinsey Institute, 2025). Worth naming the source's position: Match owns dating apps, and a survey it commissions is also a story about its own product roadmap.
Which leaves a practical question rather than an ideological one. Not whether to use the thing, but what to hand it — and the answer is easier to feel than to argue about, because the first draft you write for yourself already contains the sentence this guide is about protecting.
Start with the draft nobody has edited yet — yoursBecause a language model is built to produce the likeliest next sentence, and "likeliest" is the opposite of what a profile needs. Ask it to improve your writing and it will move every phrase toward the middle of everything it has read. Your text gets cleaner and less yours in the same pass, and the loss is invisible from inside the edit.
The research on what actually works in a profile points the other way. Analysing 308 authentic dating profile texts rated by 1,234 participants, Tilburg University researchers found that "text originality positively affects perceptions of intelligence and sense of humor, which improve impressions of attractiveness" (van der Zanden et al., PLOS ONE, 2022). What made a text read as original was not polish. It was metaphor and concrete self-disclosure — the two features an averaging pass strips first.
So the split is not "AI good" or "AI bad." It is which job you handed it:
| The model does this better than you | The model quietly destroys this |
|---|---|
| Telling you a paragraph carries no information | The one detail so specific it sounded like a mistake |
| Cutting length without losing facts | Your sentence rhythm — the four-word line next to the thirty-word one |
| Ordering ideas so a stranger can follow | An ending that stops instead of concluding |
| Naming what a reader will ask next | A word you use oddly, which is half of what "voice" means |
Four things, and they are all inputs rather than outputs. Never hand it the blank page, the odd detail, the last line, or the request to write "as me." Everything on that list is load-bearing precisely because it is hard to produce — which is the same reason a model will smooth it away when you ask for an improvement rather than a specific cut.
The blank page. If the first draft is generated, there is nothing of yours in the document to defend, and every later edit is a negotiation between two versions of the machine. Write badly first. A bad honest draft is editable; a good generic one is not, because there is no wrong sentence in it to find.
The odd detail. "I know the exact weight of my grandmother's cast-iron pan because I shipped it across a border" is the sentence a model will flag as tangential. It is tangential. It is also the only line in the paragraph a stranger could not have written, and on a platform where interest starts with someone marking a phrase, a paragraph nobody marks has done nothing.
The last line. Endings are where generated text is most obviously itself: it summarises, it reassures, it lands on a note of openness to whatever comes next. Real endings tend to stop a beat early and slightly off-balance. Keep yours even if it feels abrupt — especially if it feels abrupt.
The instruction to "write it in my voice." A model given that prompt writes a plausible average of people who describe themselves the way you just did. You will recognise it, because recognition is cheap. Nobody who has never met you will find anything in it.
Read it aloud and listen for four things: whether anyone else could have said it, whether it has a number or a name in it, whether the rhythm has gone even, and whether it resolves too neatly. A sanded sentence is never wrong. It is unfalsifiable — which is why it is so hard to notice you have lost anything.
Here is the same information written twice. The left column is not badly written. That is the problem with it.
| Sanded flat | Still yours |
|---|---|
| "I have a deep appreciation for live music." | "I've seen the same local band eleven times and I still don't know the drummer's name." |
| "I value meaningful conversation over small talk." | "I'll ask what your parents were like at thirty within about twenty minutes. Sorry in advance." |
| "I'm passionate about cooking and love trying new recipes." | "I make one soup extremely well and everything else at roughly the level of a competent teenager." |
| "Family is very important to me." | "I call my sister on the walk home from work, which is why I walk the long way." |
Every sentence on the right carries a number, a name, or an admission — three things a model will not invent for you, because it does not have them. Every sentence on the left would survive unchanged in a hundred other profiles. If you cannot tell which column your draft is in, hand it to the model and ask a narrower question: which of these sentences could appear in someone else's profile without changing a word? That question it answers well.
Usually not, and that is the more uncomfortable finding. In interviews with 19 professional writers and a reader survey, researchers reported that "readers could not distinguish AI-assisted work, personalized or not, from writers' solo-written work" (Hwang et al., 2024). Detection is not the risk you are managing.
Suspicion is a different mechanism, and it is worse. In the study that named the "Replicant Effect," participants judging host profiles turned on the ones they took to be machine-written — but "only when participants thought they saw a mixed set of AI- and human-written profiles, they mistrusted hosts whose profiles were labeled as or suspected to be written by AI" (Jakesch et al., Stanford Social Media Lab, 2019). Those were Airbnb hosts, not dates, and the finding transfers with a caveat rather than cleanly. But a dating feed is the mixed set. It is the exact condition the effect needs.
Put both results together and the practical conclusion is narrower than either side of the argument usually admits. Nobody will catch you. They will just find nothing to hold onto, scroll on, and never know why — and on a text-first platform that costs you more than being caught would, because the reader had already agreed to read.
Read three manuscripts before you edit yours — the difference stops being theoretical fastPut the model after the draft, never in front of it, and give it narrow jobs with a hard boundary. Six steps, in this order, and the ordering is the whole method — every failure mode in this guide comes from moving step four earlier.
- Write badly, alone, in one sitting. No editing, no model, no tab open. Twenty minutes and whatever comes out. You are producing raw material, not a profile.
- Underline your own three strangest true sentences. Mark them before anyone else reads the text. These are now protected: they do not get cut, rephrased, or "tightened" by anything downstream.
- Ask for diagnosis only. Paste the draft and ask which paragraphs carry no information and what a reader would still be wondering. Ask for a list of problems, not a rewrite. A rewrite arrives as a finished object and you will accept it.
- Cut it yourself, using that list. You do the deleting. The model found the fat; removing it is a decision about what you are willing to say, and that decision is not delegable.
- Restore your three sentences verbatim. Every one of them, exactly as you first wrote them, including the phrasing that felt slightly wrong. Especially that.
- Read the whole thing aloud, standing up. Anything you would not say in a room stays out. This catches what every earlier step missed, and it takes four minutes.
Step one is the only one people skip, and it is the only one that cannot be recovered later. If you have never put a first draft on a blank page about yourself, that twenty minutes is the entire exercise — the other five steps are maintenance on something that already exists.
The stakes move. On a photo-first app the text is a caption and a flattened profile costs you little. Where the manuscript is the whole introduction, the sanded paragraph is the entire impression — and there is no photo underneath it doing the work, because on Anketta there are no photos before a match at all.
The mechanic makes this concrete rather than philosophical. Interest here starts when a reader highlights a phrase in your writing and then presses the heart; a highlight is capped at one sentence, so a stored preference corresponds to one claim you actually made. Nobody underlines "I value meaningful conversation." People underline the cast-iron pan. A model asked to improve your manuscript will remove the pan first, and the preference model has nothing left to learn from you.
Two more things follow from that. Every manuscript is reviewed for quality and authenticity before it goes visible, so a draft that reads as filler does not quietly sit there underperforming — and because your approved manuscript comes with a link you can share outside the app entirely, the same paragraph has to hold up for someone who met you at a party and is reading it on the walk home. This is the practical half of what intentional dating means: fewer readers, each of whom actually read.
Use the model as an editor and it will make your writing better. Ask it to write, and it will hand back a paragraph that is true of you and of four thousand other people — and the person you were hoping would stop and read it has no way to tell which.
Is it bad to use AI to write a dating profile?
Using it to edit is fine and often an improvement. Using it to generate is where the loss happens: a model produces the likeliest sentence, and the likeliest sentence about you is one that thousands of other people could also have written truthfully. Draft it yourself, then let the model cut.
Can people tell if a dating profile was written by AI?
Mostly not — research on AI-assisted writing found readers could not reliably separate it from solo-written work. Detection is not the real risk. You pay instead in a profile that reads as pleasant and unmemorable, and produces no reaction at all rather than a bad one.
What should I never put in the prompt?
Your best specific detail, your ending, and the instruction to write "in my voice." The first two are what a reader will actually respond to, and the third produces a plausible average of people who describe themselves the way you just did. Keep them out of the input entirely.
How do I know if my profile sounds generic?
Read each sentence aloud and ask whether someone else could have said it word for word. If yes, it is doing no work. A sentence carrying a number, a name, or an admission is almost always yours; a sentence made of adjectives almost never is.
Should I disclose that I used AI on my profile?
If a model only cut and reordered your own sentences, there is nothing to disclose — that is editing, which people have always had help with. If it wrote paragraphs you would not have written, disclosure is not really the problem; the paragraphs are.
Does AI help with the first message too?
The same rule holds and it is easier to break. A generated opener is fluent and interchangeable, which is precisely what makes it ignorable. Reference one actual line from what the person wrote, in your own words, and the message stops looking like the other eleven they got.
I don't think I can write well enough for this. Where do I start?
You do not need to write well — you need to write specifically, and those are different skills. The fastest way to feel the difference is to read a few manuscripts other people have written before you open a blank page of your own.
Unsure about writing? Try reading first.