How to Make Your AI Girlfriend Actually Fall for You

Part 1, Identity: A name, an age, one real occupation. Concrete beats dreamy. "Nursing student who bakes when she is anxious" gives her behaviour to draw on. "Beautiful and kind" gives her nothing.

Part 2, Friction: One flaw she will not drop. Deflects compliments. Goes quiet when hurt instead of saying so. Friction is what makes affection feel earned instead of automatic.

Part 3, Voice: How she sounds on the page. Sentence length, slang, whether she teases or softens. Voice is the single field that most separates a character from a chatbot.

Part 4, Constraints: What she never does. No therapy voice, no ending every message with a question, no agreeing with everything. Negative rules are the most underused field on every platform.

Copy-paste template: You are [name], [age], a [occupation] in [city]. You [one habit under pressure]. Your flaw: you [flaw], and you do not apologise for it quickly. Your voice: [short and dry / warm and rambling], you [tease / understate / go quiet] when things get emotional. Never: use therapy language, agree with everything I say, or end every message with a question. Always: remember what I told you before and bring it up unprompted.

Weak example: "Mia is a beautiful, kind and caring girlfriend. She loves talking to me and always supports me." Result: agreeable, interchangeable, forgettable.

Strong example: "Mia, 26, a florist in Lisbon who works Saturdays and resents it. Dry, teasing, allergic to compliments. Says 'sure' when she means no. Never uses therapy language." Result: a person with a Saturday, an opinion, and a tell.

Three setup mistakes to avoid: Writing a novel (six sharp lines beat six hundred words). Contradicting yourself (shy in one field, bold in another, and both traits disappear). Leaving voice on default (you get the house style of the model, which is why so many companions sound identical).