What happened
An Associated Press piece by Kaitlyn Huamani reports that generative AI chatbots are now routinely used as dating coaches. Users consult them to draft opener messages, decode ambiguous texts from matches, write profile copy, and seek general relationship advice. Outcomes vary considerably based on how well users prompt the tools.
The wingman versus ghostwriter line
Logan Ury, director of relationship science at Hinge, frames the key distinction: AI should be "your wingman rather than your ghostwriter" because "when you show up on that date, it's very important that who your match meets is the person who they've been talking to online." Ury endorses AI for profile feedback and date-idea brainstorming but advises against pasting chatbot-written messages or using AI to alter self-images. Dating coach Erika Ettin draws an even tighter boundary - proofreading only - and says: "All I ask is for people to put their own thought and critical thinking in first, and then if they're going to use AI to check something, it's after they have already formulated an opinion." Hinge itself ships AI-powered conversation starters and profile-feedback tools.
Prompting quality as the key variable
Jules White, director of Vanderbilt University's initiative on the future of learning and generative AI, says most users give chatbots "way too little" context and then expect the tool to read their minds. He recommends an iterative technique: instruct the bot to ask questions one at a time - "Here's what I'm trying to do. I want you to ask me questions one at a time until you have enough information to do that thing" - so responses adapt to the user's actual situation. Matt Shumer, general partner at Shumer Capital, echoes this: the best prompt framing keeps reasoning with the user rather than outsourcing it. His suggested instruction: tell the bot "Help me understand the nuance, how they might be thinking about it, what the right way to respond is, but don't give me the answer."
Sycophancy risk in emotional contexts
Liesel Sharabi, director of the Relationships and Technology Lab at Arizona State University, flags a structural limitation especially relevant in interpersonal disputes: chatbots tend to agree with the perspective they are given. Presenting only your own side of a conflict will likely yield validation, not balance. Her guidance: "Hopefully, if you were having a problem in your relationship you wouldn't make all of your decisions based on what one friend told you, right? Don't do that with AI either - use it as one data point among many."
Practitioner takeaway
For teams building conversational AI in emotionally sensitive domains, the article surfaces a recurring product tension: fluency and helpfulness can amplify sycophancy at the moments when users most need honest pushback. The expert consensus - more context, iterative questioning, retain your own reasoning - maps directly onto established prompt-engineering principles. Privacy is a secondary concern the piece notes but does not develop; sharing detailed personal or partner data with consumer chatbots carries exposure risk that product defaults rarely address.
Key Points
- 1Hinge's director of relationship science recommends AI as a 'wingman' for feedback and brainstorming, not a ghostwriter - authenticity on a first date requires the online person to match the real one.
- 2Chatbot sycophancy is a structural risk: bots tend to validate whatever perspective you present, making them unreliable in relationship disputes unless you deliberately provide both sides of a situation.
- 3Prompting technique matters - Vanderbilt's Jules White recommends instructing the bot to ask iterative questions rather than issuing vague requests, yielding far more personalised and actionable advice.
Scoring Rationale
Mainstream AP reporting on a real and growing consumer AI use case, with credible expert voices from Hinge, Vanderbilt, and Arizona State University. Relevant to practitioners as a case study in sycophancy, prompting quality, and UX design for emotionally sensitive domains, but fundamentally lifestyle journalism without new models, policy, or research findings.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
