AI dating answers can make a favourable impression even when a reader believes it came from a machine. In a 2026 matchmaking experiment involving 307 participants, researchers found that labelling a set of responses as human-written or AI-generated did not have a statistically significant effect on reported romantic interest or feelings of interpersonal closeness. What mattered more was whether the words sounded convincing, thoughtful and human-like.
That is a provocative finding for online dating, where people often agonise over what to write about themselves. But it needs an immediate qualification. Participants evaluated text in a simulated matchmaking task. They did not meet the person behind a profile, begin a relationship or discover that an actual romantic partner had deceived them. The study is about first impressions of written responses, not proof that people prefer AI partners or that AI-written profiles produce successful dates.
What AI dating answers did to first impressions
The researchers, Jessica Szczuka, Lisa Mühl, Paula Ebner and Simon Dubé, presented participants with ten answers associated with a potential partner’s dating profile. The answers came from a task designed to encourage personal disclosure and a sense of closeness. All were generated by a large language model, but participants did not know that every response had the same underlying origin. The experimental labels led people to believe that some text came from a human and some from AI.
After reading the responses, participants rated their interest and perceived closeness. The source label alone did not significantly change those outcomes. Instead, respondents’ impressions depended on qualities they perceived in the writing, including whether it seemed human-like. This is not surprising in one sense: people often judge a potential match on the words available to them, especially before meeting. What is notable is that knowing or believing AI was involved did not reliably extinguish the appeal of the text.
The difference between the actual author and the perceived author is central. A participant might find a response attractive while knowing nothing reliable about the individual a dating service claims it represents. A polished answer can create warmth without giving a reader independent evidence that the profile owner thinks, feels or behaves as the words suggest. That uncertainty exists with human-written dating profiles too. AI simply makes producing convincing language easier.
Good writing can signal qualities that are not there
Dating profiles are miniature acts of self-presentation. A person chooses details that make them seem interesting, funny or sincere. Long before generative AI, friends helped rewrite biographies and dating websites encouraged users to improve their descriptions. A chatbot changes the speed and scale of that process. It can offer versions that sound playful, confident or emotionally perceptive in seconds, even when the person using it would struggle to express those qualities unaided.
That creates a gap between communication and evidence. Someone may honestly ask a tool to tidy a clumsy sentence about a real hobby. Another user may ask it to invent an unusually sensitive answer that bears little relation to their behaviour. The resulting text can be equally fluent. A reader must decide what they are entitled to infer from it. Attraction to the wording is not confirmation that the apparent personality belongs to the person behind the account.
Research published in the same journal on how people judge the human-likeness of text-based chatbots helps explain why style matters. It examined different dimensions of human-like impressions, from communication ability to warmth and personality. That is a separate study, with its own methods and sample. Together, the papers show why ‘sounds like a person’ is more complicated than a single test of whether an answer appears natural.
The source label did less than readers might expect
Many people believe they would reject romantic language once told it came from an algorithm. The experiment suggests that a simple label may not produce such a predictable reaction when the text itself is appealing. But a non-significant result is not proof that labels never matter. It means the researchers did not find convincing evidence of a label effect under the tested conditions. Different wording, stronger disclosure, a different audience or an actual dating encounter could produce other outcomes.
There is also an important distinction between an AI-generated message and a conversation with an AI companion. In the matchmaking experiment, participants evaluated answers as though they belonged to a potential partner. They were not asked to form a continuing bond with an artificial character. Research on AI companionship examines different relationships, where users may know that the conversational partner is a machine from the beginning. LiveAIWire has separately examined AI companions and loneliness, where ongoing interactions create different questions from a brief dating-profile judgement. Confusing these situations would turn a modest experimental result into an exaggerated claim about love.
LiveAIWire has explored why AI dating tools can blur the signals people use to judge attraction. The new study concentrates on the written first impression rather than visual appearance or live interaction. It is a reminder that online dating often begins with a thin slice of somebody’s presentation. A few paragraphs can influence expectations long before two people discover whether they communicate well in person.
Authenticity is a practical problem, not only a philosophical one
There is a reasonable distinction between using AI as an editor and allowing it to impersonate a personality. If someone writes a sincere account of their own interests and asks for spelling help, the core facts remain theirs. If the system invents memories, values or an emotional history, the other person is judging a performance. The difficulty is that readers cannot reliably tell which process produced a smooth reply.
That matters because trust in dating is built through consistency. Someone’s account of their experiences should broadly match how they act and speak later. A profile crafted to maximise attention may attract initial interest while increasing disappointment when the interaction moves to a video call or a meeting. The study does not establish how often that happens, but its findings make the possibility worth considering. Language that creates a sense of closeness can be cheap to generate; reliable behaviour cannot.
The concern becomes more serious when technology is used deliberately to deceive. LiveAIWire has reported on AI-enabled fake dating profiles and romance scams. A study about perceived romantic interest is not evidence about crime rates, yet it helps explain why persuasive text alone should not be treated as identity verification. People should still check whether a match is genuine and be cautious about requests for money, private data or pressure to move conversations away from established services.
A further issue is what a dating platform should disclose. Services may decide to identify heavily generated profiles, offer optional editing assistance or provide warnings about impersonation. Each approach has trade-offs. Overly broad labels could stigmatise honest people who need writing help, while no disclosure at all can make deliberate fabrication easier to hide. The matchmaking study does not test platform regulation. It instead highlights the uncertainty such policies are trying to address: readers may respond emotionally to language before establishing how it was created.
Why a thoughtful answer still needs a real person behind it
For people using dating apps in good faith, an AI assistant may be most valuable as a mirror rather than a ghostwriter. It can identify awkward wording, suggest clearer ways to describe a hobby or point out that a profile says little about what someone values. The final answer should remain accurate to the person who will have to stand behind it. A line that is attractive but untrue may win attention at the cost of a more meaningful connection.
The experiment also offers a caution to readers tempted to judge every polished message as suspicious. Human beings can write beautifully, and people with less confidence in writing can use tools without intending to mislead. Accusing someone of using AI on the basis of fluent prose would be unreliable. Better questions concern the quality of the actual conversation, whether details are consistent and whether both people are honest about their expectations.
Another study of relationships with humanised AI companions found that interviewees could experience emotionally meaningful interactions while still recognising technical limitations. That separate evidence cannot be added to the matchmaking experiment as though the participants were the same. It does, however, reinforce the broader point that people respond socially to language even when its source is unusual.
The lasting lesson from the 307-person study is not that a chatbot has cracked romance. It is that written warmth can influence people without confirming the authenticity of its source. In dating, a compelling introduction is only the beginning. The part that matters most still has to survive the move from words on a screen to a person who can answer for them.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.