Can AI Tell If Someone Likes You After a Date?

Not reliably: AI may detect weak patterns in a specific dataset, but it cannot determine an individual person’s feelings. A 2025 study found modest links between model predictions and dating outcomes in speed-date transcripts. It did not establish a reliable way to classify whether one particular person is interested in you. ^[1]

What did the 2025 study actually test?

The researchers analyzed 964 speed-date transcripts matched to post-date self-report data. The transcripts came from an original dataset of 1,039 dates involving 187 undergraduate students at a large Midwestern university in 2007. Each speed date lasted four minutes. ^[1]

The study’s measures capture different things. A participant’s report describes their own stated interest; it does not verify what the other person felt. Contact-information exchange is an observable action, but it does not prove mutual attraction or show whether a relationship will develop.

The main analysis used ChatGPT (gpt-4-0613), with results for Claude 3 reported in supplementary analyses. The analysis concerned model predictions in relation to the study outcomes. The analysis concerns language in recorded speed dates—not dating-app message histories, live conversations, or a commercial dating-assistant product. ^[1]

How strong were the predictions?

The study reported weak associations between language-model predictions and dating outcomes. These findings are not a percentage chance that a date will succeed or a calibrated probability for a particular person.

A correlation shows how strongly two sets of results vary together. These correlations indicate weak associations at the group level: predictions contained some signal across the study’s dates, but much of the variation was not captured by the predictions. They do not establish a dependable way to classify an individual date as interested or uninterested, or to estimate the odds that a specific person will want another date.

For predictions about the study’s match outcome—whether participants exchanged contact information—ChatGPT performed on par with human judges who had access to the same information. The researchers also found that the model’s predictions added information beyond speed daters’ own predictions. ^[1] In this analysis, “on par” means the model’s performance was comparable to those judges on that outcome. It does not establish practical equivalence across other dating situations, nor does it show that the predictions are accurate enough to guide a decision about one person.

ChatGPT’s predictions were correlated with human raters’ predictions at r = 0.21–0.35. ^[1] The authors describe the models’ accuracy as modest and note that not all cues were valid predictors of matching. They conclude that humans and language models rely on shared but imperfect heuristics when judging romantic attraction. ^[1] A cue that sounds meaningful to a chatbot may still have several explanations.

How can you use AI to reflect on a date?

Use a simple framework—Cue, Uncertainty, Next step—to keep observations separate from interpretations. This is a reflection tool, not a method tested in the study or a validated way to predict someone’s interest.

• Cue: Record what the person actually said or did, without adding a conclusion.
• Uncertainty: List more than one plausible explanation for the cue. Avoid treating a chatbot’s preferred explanation as a fact.
• Next step: Choose a respectful action that lets the other person answer freely. Don’t use AI to pressure them or to override what they say.

For example, suppose your date said, “We should do this again,” while saying goodbye, but has not messaged you since. The observed facts are that they made the comment and have not followed up. “They are definitely interested” and “They were only being polite” are interpretations; neither is established by those facts. Other explanations could include uncertainty, busyness, or a different communication style.

You could ask an AI tool to list those possibilities while clearly labeling them as guesses, rather than asking it to decide which one is true. Then take a low-pressure next step yourself: “I enjoyed meeting you. Would you be interested in getting together again sometime? No worries if not.” This gives them room to answer. The example illustrates a way to reflect and communicate; it does not predict whether they will say yes. Ask once, then take their answer—or lack of response—as the guide. If they say no, accept it without trying to persuade them.

If you want AI to help you think through another situation, try: “Separate what I observed from what I inferred. List plausible alternative explanations and suggest one respectful way to clarify interest without assuming how the other person feels.” The aim is to make your own thinking clearer, not to have the tool deliver a verdict.

What can’t this research tell you about your own date?

The dataset came from undergraduate students at one Midwestern university in 2007, so the findings may not apply to people with different ages, backgrounds, or dating contexts. ^[1] The study analyzed transcripts, not tone of voice, body language, or the full context surrounding a particular date. ^[1]

It did not validate tools that assess private dating-app conversations or establish that AI can reliably interpret a person’s feelings. ^[1] Treat the results as evidence that models can find some patterns in a defined dataset—not as a reason to discount what someone tells you directly or to keep pursuing them when they decline.

Frequently asked questions

Is it safe to paste someone else’s private messages into an AI tool?

Private messages can remain sensitive even after you remove names and other identifying details. Instead of uploading the exchange, summarize it in your own words and leave out details that could identify the person. Before sharing sensitive content, check the tool’s data-retention and model-training settings. If you cannot confirm how it handles the content, don’t share it. The speed-dating study did not evaluate AI privacy practices or test analyses of private app-message histories. ^[1]

What should I do if an AI’s interpretation conflicts with what my date told me?

Give more weight to what the person told you directly than to a chatbot’s interpretation. An AI interpretation is not a basis for overruling someone’s own statement. If their meaning is unclear, ask a respectful follow-up question.

Can AI tell whether someone is flirting from emojis alone?

The study does not establish that AI can identify flirting from emojis alone. It analyzed speed-date transcripts, not emoji-only conversations, so it cannot validate a definitive emoji-based reading of someone’s interest. ^[1] Treat an emoji as ambiguous unless the person’s words and actions clarify what they mean.

Reference

^[1] “Large language models can detect verbal indicators of romantic attraction.” Scientific Reports, volume 15, 2025.