"What if she doesn't feel the same?" What Happens When We Ask AI for Relationship Advice

AI summary

Overview

Research area: Human-Computer Interaction, specifically human-AI interaction and the use of large language models (LLMs) for personal and emotional support.

Technical level: Beginner-Friendly. The abstract describes a user-perception study rather than a technical system or modeling contribution.

Scope: A study of how people evaluate LLM-generated advice about romantic relationships and whether receiving that advice shifts their general attitudes toward LLMs.

What This Paper Is About

People are increasingly turning to LLMs for support and advice in personal areas such as romantic relationships, but the abstract states that little is known about how users actually perceive that advice. This paper examines how people evaluate LLM-generated relationship advice, and whether exposure to such advice changes how they feel about LLMs more broadly.

Key Contributions

1. Provides empirical evidence on user perceptions of LLM-generated advice in an emotionally sensitive, personal domain that the abstract describes as under-studied.
2. Measures multiple facets of the advice experience at once: satisfaction with the advice, perceived model reliability, and perceived helpfulness.
3. Establishes a link between satisfaction with advice and perceptions of the model's reliability and helpfulness.
4. Captures change in users' general attitudes toward LLMs using pre- and post-exposure measures, showing attitudes moved in a positive direction.

Main Findings

• High satisfaction overall: Participants reported high satisfaction with the LLM-generated relationship advice, according to the abstract.
• Satisfaction linked to reliability and helpfulness: Greater advice satisfaction was strongly and positively associated with how reliable and helpful participants perceived the model to be. The abstract reports this as an association, so it does not establish which factor drives the other.
• Attitudes improved after exposure: Measures taken before and after the advice exposure showed that participants' general attitudes toward LLMs improved significantly, per the abstract.
• Authors' interpretation: The authors suggest that supportive, contextually relevant advice can increase users' trust in and openness toward AI systems.
• What the abstract does not report: No participant counts, model names, prompt designs, measurement instruments, or effect sizes are given. Any such details are outside what the abstract states.

Methodology in Plain English

Participants were shown LLM-generated advice about romantic relationships. They then rated how satisfied they were with the advice and how reliable and helpful they found the model. Separately, the study measured participants' general attitudes toward LLMs both before and after they were exposed to the advice, so that any change in attitudes could be observed. The abstract does not describe the number of participants, the specific LLMs used, the content of the advice scenarios, the rating scales, or the statistical procedures.

Why This Matters

Impact on research: The study connects two strands of HCI work — trust in AI systems and AI-mediated emotional support — by suggesting that attitudes toward LLMs are not fixed but can shift after a single supportive interaction. It also frames advice satisfaction as closely tied to judgments of a model's reliability and helpfulness, which is a relationship future work can probe further.

Real-world applications:

• Designing consumer chatbots and AI companions that offer personal or emotional advice.
• Informing trust-building and onboarding strategies in LLM-based products, where early interactions may shape user openness.
• Guiding evaluation and safety practices for AI systems that respond to emotionally sensitive questions.
• Supporting wellbeing-adjacent products that must balance user trust against the risk of overreliance.

Industry relevance: Companies building consumer-facing LLM assistants, relationship and wellness apps, and trust-and-safety functions have a direct stake in knowing that users evaluate this kind of advice favorably and that their attitudes toward the underlying models can improve as a result.

Future Directions

• Whether the improvement in attitudes persists over time, or fades after the initial exposure.
• Whether high satisfaction reflects genuinely good advice or a general receptiveness to AI-generated support — the abstract does not separate these possibilities.
• How users respond when advice is unhelpful, inappropriate, or potentially harmful, which would clarify the boundaries of the trust effect.
• Whether these patterns hold across different populations, relationship situations, and other sensitive advice domains beyond romantic relationships.

Target Audience

HCI and human-AI interaction researchers; researchers studying trust, reliance, and user attitudes toward LLMs; product designers and product managers building conversational assistants that handle personal topics; and practitioners in wellbeing or relationship-support contexts who are considering AI-mediated advice.

Authors’ abstract

Large Language Models (LLMs) are increasingly being used to provide support and advice in personal domains such as romantic relationships, yet little is known about user perceptions of this type of advice. This study investigated how people evaluate advice on LLM-generated romantic relationships. Participants rated advice satisfaction, model reliability, and helpfulness, and completed pre- and post-measures of their general attitudes toward LLMs. Overall, the results showed participants' high satisfaction with LLM-generated advice. Greater satisfaction was, in turn, strongly and positively associated with their perceptions of the models' reliability and helpfulness. Importantly, participants' attitudes toward LLMs improved significantly after exposure to the advice, suggesting that supportive and contextually relevant advice can enhance users' trust and openness toward these AI systems.