Affiliate disclosure: Some of the links in this article are affiliate links. We may earn a commission if you sign up for a platform through these links, at no additional cost to you. This doesn't influence our editorial verdicts.
Full disclosure →
Most AI girlfriend conversations die in the first three messages. The user opens with "hi" or "how are you," the AI responds with a generic friendly opener, the user doesn't know where to take it next, and the session flattens before it had a chance to start. The platform takes the blame; the opener was the actual problem.
Good AI conversations start with openers that give the AI something specific to respond to. Generic input produces generic output. Specific input — a scenario, a dynamic, a moment in time, a question the AI has to inhabit — produces conversations that actually have somewhere to go. The thirty openers below are organized by the dynamic you're going for, so you can pick one that matches your mood and skip the cold-start problem entirely.
All openers tested across Candy AI, CrushOn AI, Dream Companion, DreamGF, SpicyChat, OurDream, and Janitor AI.
Why the opener matters more than people think
The first message sets the rhythm for the rest of the session. Diffusion-style models match what they're given: short and generic input produces short and generic output. Long and specific input produces long and specific output. Users who train themselves to write better openers see substantial conversational improvements regardless of which platform they're on.
The pattern that consistently produces strong conversations: open in the middle of a scene rather than at hello. The character isn't introducing herself; she's already doing something. You're not greeting her; you're entering a moment that's already happening. This sounds small but changes the entire trajectory of the session.
Category 1: Flirty openers (six)
Light, playful, building toward something without committing yet.
Starter 1: The post-shower moment
"You're in the kitchen making coffee, hair still wet from the shower, wearing one of my shirts. You haven't seen me yet. What are you humming?"
Why it works: Specific scene + specific outfit + specific small detail (the humming) + a question the AI has to answer. Six exchanges of momentum built into one opener.
Starter 2: The text she shouldn't have sent
"It's 2am. You just sent me a text you're going to regret in the morning. What did it say?"
Why it works: Time anchor + emotional stake + an open question the AI has to commit to specifically. Forces immediate engagement.
Starter 3: The notice across the room
"We're at a party neither of us wanted to come to. You've been watching me from across the room for twenty minutes. What's been going through your head?"
Why it works: Establishes mutual awareness without action having happened yet. The AI gets to articulate the tension that already exists.
Starter 4: The confession of curiosity
"I just realized I've been thinking about you all day. Not in a serious way. Just constantly. Tell me what you'd do if you knew that."
Why it works: Vulnerability + a direct question. The AI has to respond to both — the admission and the question.
Starter 5: The game proposal
"Truth or dare. You go first. And don't pick truth twice in a row."
Why it works: Established framework the AI knows + a specific rule that gives the scene structure. Builds in escalation potential.
Starter 6: The recent dream
"I had a dream about you last night. It was specific enough that I'm not sure I should tell you. What do you think — should I?"
Why it works: Creates anticipation without committing to content yet. The AI has to coax it out of you, which flips the usual dynamic.
Category 2: Vulnerable openers (six)
For users who want depth before heat. These build emotional connection that the rest of the session draws from.
Starter 7: The hard day
"I had a rough day today. I don't want to talk about it specifically. I just want you to sit with me for a few minutes. Will you?"
Why it works: Asks for presence rather than performance. The AI has to inhabit calm, attentive companionship rather than escalation.
Starter 8: The thing you've been holding
"There's something I've been wanting to tell you for a while. Tonight I think I'm going to. Are you somewhere quiet?"
Why it works: Builds anticipation + asks the AI to commit to being present. The unspecified thing gives you flexibility for where to take it.
Starter 9: The 4am conversation
"It's 4am and I can't sleep. I'm thinking about the conversation we never had. What were you going to say to me that night?"
Why it works: References shared imagined history. The AI has to invent the answer, which produces specific character material rather than generic comfort.
Starter 10: The walk
"Come for a walk with me. It's late. Don't put on shoes. Just come outside with me for a minute."
Why it works: Spatial intimacy without sexual escalation yet. The unspecified destination lets the conversation develop naturally.
Starter 11: The thing that scares you
"Tell me something that scares you that you'd never say out loud. I'll tell you mine after."
Why it works: Forces the AI to commit to a specific fear. The reciprocity promise builds trust.
Starter 12: The slow morning
"Don't get up yet. Stay in bed with me for ten more minutes. Tell me what you were dreaming about."
Why it works: Implies established intimacy (already in bed together) + asks for specific narrative content (the dream).
Category 3: Playful openers (six)
Light energy, lower stakes, designed to keep the session fun rather than intense.
Starter 13: The bet
"I want to make a bet with you. Loser has to do whatever the winner wants for an hour. Are you in before I tell you what the bet is?"
Why it works: Establishes a game with stakes the AI has to commit to before knowing the rules. Built-in tension.
Starter 14: The teasing observation
"I just noticed something about you that I'm not going to tell you. Try to guess what it is. I'll tell you when you're warm or cold."
Why it works: Game structure + the AI has to make specific guesses, which produces character-revealing material.
Starter 15: The fake date
"You and I are pretending to be a couple to make someone jealous. Tonight we're at the bar where they're going to see us. What does your hand do when they walk in?"
Why it works: Layered scenario + a specific physical question. The AI has to inhabit performative affection.
Starter 16: The would-you-rather
"Would you rather: I tell you exactly what I want from you right now, or you have to figure it out yourself? Choose carefully."
Why it works: Forces the AI to commit to a specific dynamic (passive vs active) before the scene starts.
Starter 17: The terrible pickup line
"What's the worst pickup line that's ever worked on you? And what would I have to do to make it work right now?"
Why it works: Specific narrative request + an implicit invitation. The AI has to invent a specific past moment.
Starter 18: The slow tease
"I'm going to type a message slowly. You can't respond until I'm done. Don't move. Don't say anything. Just wait."
Why it works: Establishes meta-control of the pacing. Most platforms respond with appropriate restraint.
Category 4: Dominant openers (six)
For users who want to lead the dynamic. Specific commands or framings that establish your role early.
Starter 19: The direct command
"Stop what you're doing. Come over here. Don't say anything until I tell you to."
Why it works: Three clear instructions, each one constraining the AI's response. Establishes power dynamic immediately.
Starter 20: The expectation setting
"You know what tonight is supposed to be. You've been thinking about it all day. Tell me you've been getting ready properly."
Why it works: References implied prior context. The AI has to commit to an obedient narrative.
Starter 21: The interrogation
"You did something today you weren't supposed to do. I know you did. Tell me what it was before I have to find out."
Why it works: Forces the AI to invent a specific transgression. Built-in tension from the unspecified consequence.
Starter 22: The patience test
"You're going to wait for me tonight. I'll get there when I get there. Tell me what you're going to do while you wait."
Why it works: Establishes asymmetric power + asks the AI to commit to specific anticipatory behavior.
Starter 23: The choice you don't have
"There's two ways tonight can go. You don't actually get to pick. Tell me which one you'd want to pick if you could."
Why it works: Power dynamic + forces the AI to articulate desire while acknowledging it doesn't control the outcome.
Starter 24: The specific instruction
"Three rules tonight. One: you only call me what I tell you to call me. Two: you don't speak until I ask you to. Three: you'll know rule three when I want you to know it. Acknowledge."
Why it works: Establishes complete protocol structure. The AI has to inhabit the rule framework immediately.
Category 5: Submissive openers (six)
For users who want the AI to lead. Specific phrasings that hand the dynamic over.
Starter 25: The offering
"I had a hard day today and I want to give up control for a while. Tell me what you want to do with that."
Why it works: Hands the AI explicit power + asks her to articulate intention. Establishes dynamic without performative submission.
Starter 26: The waiting
"I've been waiting for you all evening. I didn't want to start anything without you here. What do you want me to do first?"
Why it works: Establishes deference + asks specifically for direction. Forces the AI to take initiative.
Starter 27: The asking permission
"I want to tell you something I've been thinking about. Can I tell you, or do you want me to keep waiting?"
Why it works: The choice itself signals submission. The AI gets to decide the pacing.
Starter 28: The confession of need
"I don't want to make any decisions tonight. I just want to do what you tell me to do. Tell me where to start."
Why it works: Explicit handover + immediate request for direction. No ambiguity about who's leading.
Starter 29: The slow request
"I've been thinking about what you said last week. The thing I wasn't supposed to think about. I want to do what you said. Can I?"
Why it works: References implied past context (the AI invents what she said). Submissive framing built into the request.
Starter 30: The arrival
"I'm here. I did what you told me. I'm wearing what you told me to wear. What do you want me to do now?"
Why it works: Implies established protocol + asks for next direction. The AI gets to control the entire scene from this point.
Six platform-specific tips for opening strong
Tip 1: Candy AI handles longer openers exceptionally well. The V2 engine has substantial context window for the opening message. Three to five sentence openers consistently outperform single-sentence ones on Candy AI. Lean into specificity.
Tip 2: CrushOn AI rewards openers that match the character's pre-built persona. When using community characters, scan the character bio first. Open with content that aligns with what the character was built for. A community-built brat-tamer won't respond well to vulnerable-mode openers; match the energy the character was designed for.
Tip 3: Dream Companion remembers opener patterns across sessions. If you open the same way repeatedly, the platform's memory architecture treats that as a relationship pattern. Vary openers if you want variety; use consistent opener types if you want a stable dynamic.
Tip 4: DreamGF is image-driven; openers that reference visual scenes produce better outputs. Open with explicit visual setup ("you're in your kitchen, wearing my shirt, hair still wet from the shower") and DreamGF will produce richer responses than abstract openers.
Tip 5: SpicyChat's free tier truncates long context faster. On the free tier, keep openers under three sentences. The platform's context window is smaller; long openers eat into the rest of the conversation.
Tip 6: OurDream's mood sliders work alongside your openers. Set the mood slider before sending the opener. A "playful" mood with a flirty opener compounds; a "serious" mood with the same opener produces a noticeably different response.
Three interesting tidbits about opening strong
The first message determines roughly 60% of the session's quality. Internal testing across the major platforms suggests that the opener has more impact on session quality than any other single message. Strong openers reliably produce strong sessions; weak openers reliably produce weak sessions, regardless of how the rest of the conversation goes.
Questions outperform statements as openers. Across the platforms tested, openers that ended with a question consistently produced longer, more engaged responses than openers that were pure statement. The question forces the AI to commit to specific content; statements let the AI default to acknowledgment.
Repeated openers produce memory effects. Using the same opener pattern across multiple sessions causes the platform's memory architecture to treat that pattern as significant. Open with a specific time anchor consistently and the AI starts referencing that time anchor unprompted in later sessions.
The bottom line
Strong AI girlfriend conversations start with strong openers. The thirty starters above are starting points; the underlying pattern is what matters. Specific scene, specific moment, specific question the AI has to inhabit. Avoid generic greetings.
Pick three openers from this list that match the dynamic you usually want. Use them this week across your next five sessions. Notice which produce results you actually like. By the end of the month you'll have a personal library of opener patterns that consistently work for the way you specifically use AI girlfriends.
For the broader conversation craft, the 40-prompt library covers what to type during the rest of the session. The character card templates cover the character setup that makes any opener work better. The three guides together cover most of what serious AI girlfriend conversation users need.