Most people ask AI for relationship advice the same way they'd ask a friend—venting, asking for validation, or seeking a quick fix. Engineering better prompts means giving the AI enough context (who said what, what happened before, what you actually want), asking it to challenge your assumptions, and requesting specific frameworks instead of general wisdom.
Prompt engineering is the practice of crafting questions and instructions in ways that elicit better, more useful responses from language models. It's equal parts art and science: understanding how AI systems interpret language, what kinds of instructions work best, and how to structure requests for maximum clarity and specificity. In relationships, where nuance and context determine whether advice helps or hurts, prompt engineering is a practical skill that dramatically improves the quality of AI assistance you get.
Language models are fundamentally responsive—they generate output based on what you ask. "How do I fix my relationship?" will give you generic advice that could apply to anyone. "My partner and I have been together five years, we're arguing about whether to move for my job, they feel abandoned, I feel unsupported. We're both avoidant but in different ways. How do I help them feel heard while also expressing my need?" will get you dramatically better guidance.
The difference is specificity, context, and explicit framing. When you engineer a good prompt, you're providing the AI with enough information to understand your particular situation, plus guidance about what kind of response would be most useful. You're essentially teaching the AI what "good advice" means in your context.
Context stacking: Provide relevant background before your core question. Instead of "How do I talk to my friend about their drinking?", try "My friend and I have been close for eight years. In the past year they've started drinking heavily. I've gently mentioned concern twice and been brushed off both times. They're smart and defensive. How do I bring this up in a way that doesn't make them shut down?" The additional context lets the AI tailor advice to your specific relationship history and personality types.
Constraint specification: Tell the AI what constraints matter. "Give me advice that doesn't require me to set an ultimatum (we're not there), that acknowledges their autonomy, and that I could realistically deliver tomorrow." Constraints help the AI avoid generic solutions that don't fit your actual situation.
Example provision: Show the AI examples of what you mean. "When I try to be vulnerable with my partner, they often respond with solutions instead of listening. For example, last week I mentioned stress about a work project and they immediately told me what I should do. I want them to just listen. How do I communicate that?" Examples are more concrete than abstract descriptions.
Role-playing and reversal: Ask the AI to adopt a specific perspective. "How would my partner likely perceive my request for more date nights? What would they be worried I'm implying?" This helps you anticipate their viewpoint before you communicate, reducing defensive reactions.
Step-by-step structuring: Request structured output when you need it. "Walk me through exactly what I should say when I bring this up, point by point. Then tell me three ways my partner might react, and what I should do in each scenario." Structured requests get organized responses.
"Chain-of-thought" prompts ask the AI to explain its reasoning step-by-step before giving advice. Instead of "Should I break up with my partner?", try: "I've been unhappy for two years. I love them but feel unsupported. I'm worried about being alone. Should I break up? Before answering, walk me through what factors should matter in this decision, then tell me which of my situation matches each factor." This makes the AI's reasoning transparent and lets you evaluate whether its logic fits your values.
There's a subtle risk: skillful prompt engineering can get an AI to tell you what you want to hear. If you structure a prompt to emphasize your partner's flaws and downplay your contribution, the AI will likely generate advice that justifies your position. Good prompt engineering reveals truth; bad prompt engineering weaponizes AI to confirm bias. The antidote is intentional contradiction—ask for devil's advocate responses, ask the AI to steelman your partner's position, ask for the hardest feedback it can give you.
Some AI platforms let you adjust "temperature," which controls randomness in response generation. Low temperature = more focused, consistent, conservative output. High temperature = more creative, varied, potentially surprising suggestions. For relationship advice, moderate temperature often works best: you want thoughtfulness (high temperature) but consistency and reliability (low temperature). If your platform allows it, try moderate settings (0.6-0.7) for advice and higher (0.8+) for creative brainstorming about how to phrase difficult conversations.
Try this: Identify a relationship communication challenge you're facing. Write two prompts: first, a basic version ("How do I tell my partner I'm unhappy?"). Then rewrite it with maximum specificity: include how long you've been together, what "unhappy" specifically means, what you've already tried, what outcomes you're hoping for, and what constraints exist (kids involved? financial entanglement?). Compare the responses. The engineered prompt should yield dramatically more useful, personalized advice. Notice which contextual details shifted the AI's suggestions most significantly.
Peri can explain this concept, give practical examples, help you decide whether it applies to your situation, or recommend a journey if appropriate.
Explore related journeys or tell Peri what you're working through.