New Research on the Societal Impact of Large Language Models: When AI Writes Dating Messages
In Edmond Rostand’s 1897 play, Cyrano de Bergerac feeds eloquent love letters to the handsome but tongue-tied Christian, who passes them off as his own. The woman they both court falls for words written by one man and delivered by another. More than a century later, the same arrangement has returned to online dating. The poet is now an algorithm.
A new study by our co-founder Prof. Lennart Ante in Telematics and Informatics looks at what happens when people use large language models such as ChatGPT or Claude to write or polish their dating messages. The practice already has a nickname: “chatfishing.” A recent survey found that 26 percent of US singles have used AI to help with their dating lives, a jump of more than 300 percent in a single year. Six in ten dating app users believe they have read a message written by AI.
The study draws on in-depth interviews with 45 dating app users. Twenty-three of them used AI to craft their own messages. Twenty-two had been on the receiving end. Their accounts reveal three patterns.
1. The people who use AI feel more authentic, not less
This was the surprise. Most AI users rejected the label of cheater. They described AI as a way to sound more like their real selves. One interviewee, who called herself “great in person but terrible at texting,” said her brain short-circuits in front of a blank text box, and that the AI works as a translator for thoughts she already has. A non-native English speaker used it so she would be judged on her personality rather than her phrasing.
The study calls this prosthetic authenticity. These users see the AI as a corrective lens that brings their true self into focus, rather than a mask that hides it. A second group was blunter. For them, early dating is a numbers game, and AI is simply an efficient way to reach the real conversation, which begins on the date itself.
2. The people who receive AI messages feel betrayed
From the other side, the experience looks completely different. Recipients used words like “violated,” “duped,” and “catfished.” The hurt came from a specific place. The deception was about the act of communication itself. One man explained that he had fallen for the way a woman expressed herself, then learned the expression was scripted.
Recipients also became amateur detectives. They learned to spot the signs: replies that are too polished, compliments that feel generic, answers that arrive unnaturally fast. Some tested their matches by dropping in a typo or some slang to see whether the other person would mirror it. One called the result exhausting, like needing to be a forensic linguist just to have a conversation. The cost is a creeping cynicism that makes it harder to trust anyone, including people who simply happen to write well.
3. The hardest part is meeting in person
The third pattern appears when the chat moves offline. AI users described real dread before a first date. They had built a witty, confident persona, and now they had to perform it without the algorithm. One compared it to cramming for an exam about a fake version of himself. The polish that won the date often set a bar the person could not clear in the room, and the connection collapsed under the weight of the mismatch.
What it means
Prof. Ante calls this whole arc the Cyrano Effect. It describes how AI splits the author of a message from the person sending it, and what that split does to trust on both sides.
For dating platforms, the question is no longer whether people use AI, because they already do. The harder question is what to do about disclosure, and how to support people with genuine social anxiety or language barriers without rewarding deception.
For the rest of us, the study points to a new kind of digital literacy. Eloquence is now cheap. The messy, specific, slightly awkward message may be the more honest one. And the old question from Rostand’s play has not gone away. When we fall for the words, who, or what, are we really falling for?
Ante, L. (2026). The Cyrano effect: LLM-assisted impression management and authenticity in online dating. Telematics and Informatics, 108, 102422. Open access. https://doi.org/10.1016/j.tele.2026.102422