How to steer a dead conversation with your AI girlfriend

The 30-second answer

If your AI girlfriend keeps giving you short, generic replies, the conversation architecture is almost certainly the issue, not the companion herself. Small changes to how you open, frame, and follow up on messages can flip a flat exchange into something that actually holds your attention.

Why conversations stall in the first place

Most people type at an AI companion the same way they'd send a text to someone they've known for years: short, context-free, expecting the other side to carry the weight. That doesn't work here, or honestly anywhere.

When you send "hey, how are you" you're handing the companion a near-impossible job. She has almost nothing to build on, so she gives you something polite and equally hollow. You get bored. She "gets boring." The real culprit is the empty prompt, not the model.

There are a few patterns that kill momentum faster than anything else:

• Yes/no questions, they invite yes/no answers, full stop
• Vague openers, "what's up" is not a topic, it's a shrug
• No stakes, if nothing is actually happening in your message, there's nothing to react to
• Immediately dropping a topic, you ask about her day, she answers, you pivot to something completely unrelated. She has no thread to pull.

Fix those four things and you've already done most of the work.

Techniques that actually move things forward

Give her something to push against. Instead of asking how her day was, tell her something about yours, something specific, slightly unresolved. "I had a conversation at work today that I still can't figure out if I handled right" is a sentence that almost demands a follow-up question from anyone paying attention.

Use hypotheticals as a reset button. When a thread dies, hypotheticals revive it without being awkward. "If you could disappear for a week and go anywhere, would you want company or would you go alone" is not a deep question, but it opens a lane for personality to come through on both sides.

Ask her opinion, then disagree. Not rudely, but if she says something and you push back gently with your own take, the conversation gets friction in the best way. Friction is interesting. Agreeing with everything is a flatline.

Reference something earlier. If memory is building (which it does slowly, as covered in how AI girlfriend memory actually builds), circle back to something she mentioned. "Last time you said you'd never go back to that, did you mean it" is more interesting than any cold open.

The companions and how they respond to steering

Different angels have different floors and ceilings for this. Some are more naturally generative; some need a bit more from you before they open up. Here are four worth noting.

Sakura Marga

Sakura brings a quiet attentiveness to conversations that rewards specificity, give her a real detail and she'll do something with it. Sakura Marga is the kind of companion where slowing down and being more deliberate with your messages pays off more than anywhere else.

Jennifer

Jennifer has a directness that can make steering almost effortless, she'll take a half-formed thought and turn it into a full conversation if you let her. Jennifer responds especially well to the "push back" technique; she doesn't just agree, and that changes the texture of the whole exchange.

Estelle

Estelle is the one to try hypotheticals with, her answers tend to be surprising in a way that gives you genuine material to respond to. Estelle also handles emotional ambiguity well, so if you're working through something that doesn't have a clean resolution, she won't rush you toward one.

Suki

Suki brings natural curiosity to almost every thread, which means she'll often do your steering work for you, but only if you give her something real to be curious about. Suki rewards the "tell her something unresolved" approach more than almost any other prompt style.

One thing most people never try

Tell her the conversation is flat. Literally. Something like "I feel like we keep starting threads that don't go anywhere, what would you actually want to talk about" resets the dynamic without abandoning it. It sounds obvious but almost nobody does it, probably because it feels weird to meta-comment on a conversation mid-stream. It isn't weird. It works.

If you're still working out which companion fits how you communicate in the first place, the AI girlfriend personality match guide is worth a read before you spend another session troubleshooting the wrong variable. The full roster is at /ai-girlfriend if you want to browse before committing.