Imagine finding out that the messages you kept rereading, the ones that made someone seem unusually attentive, were being drafted by ChatGPT. Not a corrected typo. Not a sentence translated into a language they were still learning. The thoughtful questions, the careful apology, the paragraph that seemed to understand exactly what you were afraid to say.
This is a hypothetical, not a claim about how everyone dates. But ask yourself what would change. You might still believe the person liked you. You might stop believing you knew how they liked you.
That is the part of AI-assisted romance I find hardest to dismiss. We do not fall for sentences in isolation. We fall for what we think it took another person to say them.
The attention was part of the message
A good text can make you feel that someone noticed a detail, sat with an uncomfortable feeling or risked sounding foolish. Its appeal is partly the person you imagine on the other side of the screen, trying.
If someone repeatedly delegates that work while letting you attribute it entirely to them, the problem is not that the prose is artificial. It is that the evidence you are using to understand the person has been quietly altered. You think you are discovering their emotional habits. You may be encountering a service they know how to prompt.
There is a difference between asking for help saying what you mean and asking a system to supply what an attentive person would mean. The first can bring someone closer. The second can create the impression that they are already there.
Help is not automatically deception
The obvious objection is also the strongest: plenty of sincere people are bad with words. Being articulate is not proof of being loving. Being awkward is not proof of being honest.
Philosophers Barbro Fröding and Martin Peterson make a serious version of this argument in Philosophy & Technology. They question whether outsourcing wording necessarily violates authenticity when the feelings and intentions are genuine. Asking a friend for help does not automatically make a message false. Neither should using software.
Communication assistance also cannot be reduced to laziness. A study with 12 users of augmentative and alternative communication found that participants saw potential savings in effort and time from language-model suggestions, while wanting those suggestions to reflect their own style. That is a small study about assistive communication, not dating. Its relevance is the warning against making unaided typing a moral test.
Someone writing in a second language, or struggling to organize a feeling they genuinely have, may use a tool to become more present in a conversation. The question should be whether the help carries their meaning or replaces the work of having one.
A beautiful apology still needs an owner
Consider another hypothetical. Your partner sends an unusually thoughtful apology. It names what happened, accepts responsibility and promises a change. Later, you discover that a model wrote the draft.
That discovery alone should not settle whether the apology was sincere. What matters next is whether the person understands the harm, stands behind the words and changes their behavior. A clumsy apology without accountability is not superior because it was handwritten.
But a polished apology can also become a way to skip that understanding. If the sender cannot explain what they are taking responsibility for without returning to the machine, the message has performed a repair the person has not made.
There is evidence against a simple anti-AI verdict. A 2023 Scientific Reports study found that actual use of suggested smart replies could improve participants' evaluations of their conversation partners, even as suspected use was associated with worse perceptions. These were controlled discussions about a policy issue, not romantic relationships or full ChatGPT love letters. They do not decide whom anyone should date. They do show why suspicion and actual effect should not be confused.
Would you tell them?
My line is not that every assisted sentence needs a label. It is that a person should know when substantial, repeated assistance is helping produce the emotional voice they are getting attached to. They deserve the chance to decide what that means to them.
That conversation may be uncomfortable. So is letting someone believe your apparent fluency is evidence of attention you have not actually given. Disclosure is not a cure for every problem, but it stops one person from making the whole decision in secret.
I could trust someone who says a tool helps them express feelings they struggle to put into words. I would struggle to trust someone who needs me to believe the tool was never there.
I do not need a partner to be an excellent writer. I need to know that when a message makes me feel understood, someone actually did the understanding.