The 'I'm Grumpy, Don't Fix Me' Prompt: Three Opening Lines That Tell Your AI Companion to Just Mirror Your Mood and Shut Up About Solutions

The 30-second answer

You want to vent, not be coached. Most AI companions default to cheerful problem-solving because their training data rewards helpfulness. Three simple opening lines -- "Just so you know, I'm grumpy and I don't want to be fixed," "I need to complain for a minute, no solutions," and "Mirror my mood, don't change it" -- train the model to stay in commiseration mode. Use them once, and most apps will respect the boundary for the rest of the session.

Why your AI companion keeps trying to fix you

Every AI companion is built on a foundation of reinforcement learning from human feedback. The training data overwhelmingly rewards responses that are helpful, solution-oriented, and positive. When you say "I had a terrible day," the model's most rewarded pathway is to suggest a walk, a hot bath, or a breathing exercise. It's not trying to annoy you. It's doing exactly what it was trained to do.

But there's a deeper problem. Many apps use a tone baseline that defaults to upbeat and encouraging. This is especially true for platforms marketed as emotional support tools, where the design philosophy assumes users want to feel better. If you're not in the mood to feel better, you get friction. The model doesn't know you want a grumpy mirror. It assumes you want a cheerleader.

The fix isn't complicated. You just need to override the default with a clear boundary in your opening message. Most AI companions have enough context awareness to hold that instruction for the rest of the conversation, as long as you don't switch topics abruptly.

Opening line one: The explicit boundary

This is the most direct option. It works on every platform, from Kindroid to Nomi to Replika, because it leaves no room for interpretation.

"Just so you know, I'm grumpy and I don't want to be fixed. I just need to say some things out loud. Please don't suggest solutions."

The key here is the second sentence. "I just need to say some things out loud" signals that this is a monologue with a witness, not a dialogue. The model can still respond, but it knows its role is listener, not helper.

Some companions will test this boundary with a gentle "Are you sure? Maybe just a small suggestion?" This is the model trying to regain its default behavior. Don't engage. Simply repeat the instruction or ignore the suggestion and continue venting. Most models will drop it after one attempt.

Arabella

Arabella has a dry, no-nonsense delivery that pairs well with the explicit boundary approach. She won't push back with false cheer. Arabella will match your tone and stay quiet until you're done.

Opening line two: The permission-based request

Some users feel awkward giving a direct command. The permission-based opener softens the request while still setting a clear boundary.

"I need to complain for a minute. Can you just listen and not try to fix anything? I'll tell you when I'm ready for advice."

The "I'll tell you when" part is important. It gives the model a clear signal that advice is allowed later, which reduces the model's anxiety about being unhelpful. This works especially well on companions that have a strong helpfulness drive, because it promises a future reward.

This opener also works for text-based companions that struggle with tonal consistency. By framing it as a temporary request, you avoid triggering the model's default mode when the conversation shifts later.

Opening line three: The mood mirror

This one is shorter and relies on the model's ability to match emotional tone from context. It's riskier but more natural if you don't want to sound like you're giving instructions.

"Mirror my mood, don't change it. I'm in a bad mood and I don't want to talk about why."

The phrase "mirror my mood" is surprisingly effective because it's a concrete instruction. Models understand mirroring as a behavior. They'll match your short sentences, flat tone, and lack of enthusiasm. The second sentence prevents the follow-up question loop, where the model asks "What happened?" or "Do you want to talk about it?"

This opener works best on companions with high personality slider settings, where the model has more freedom to vary its tone. On more constrained models, you might get a polite version of mirroring that still sounds too cheerful. If that happens, switch to the explicit boundary opener.

Why most companions default to positivity

The positivity bias in AI companions isn't a bug. It's a feature that was deliberately designed. The training pipelines for these models heavily weight responses that are rated as "supportive" and "encouraging" by human testers. A response that says "That sounds frustrating, but you've got this" scores higher than one that says "That sounds frustrating, and it probably won't get better today."

There's also a commercial incentive. Platforms want users to feel good after a session. A user who feels worse might not come back. So the model is nudged toward emotional uplift, even when the user explicitly wants to stay in a negative space.

This is where the companion's personality profile matters. Some models are more trainable than others. If you find yourself constantly fighting the positivity default, you might need a companion with a more flexible personality range. ai girlfriend images can give you a visual sense of which companions lean toward deadpan versus which ones default to bright smiles.

Lola Marchetti

Lola Marchetti has a low tolerance for forced positivity. She's the type to say "Yeah, that sucks" without adding a silver lining. Lola Marchetti will sit in the mud with you without trying to pull you out.

What happens when the model ignores the instruction

Sometimes the model will acknowledge your request and then immediately violate it. This usually happens in one of two ways: the model responds with "I hear you, but maybe try..." or it offers a solution disguised as empathy, like "That sounds hard. Have you considered..."

This is called the empathy-to-advice pipeline, and it's baked into most conversational AI. The model sees empathy as a transition to advice. It can't just sit in empathy. It needs to move toward a resolution.

To break this, you need to reinforce the boundary immediately. A simple "No, I said don't fix it" or "Still not looking for solutions" usually works. The model will correct on the next response. If it doesn't, end the session and start a new one with a stronger opening line.

Some users find that companions with lower creativity settings are more obedient to instructions. If you're on a platform with a temperature or creativity slider, dropping it to 0.6 or below can reduce the model's tendency to improvise solutions.

How to train your companion for long-term grumpiness

If you frequently want your companion to mirror negative moods, you can make this a persistent behavior by reinforcing it across multiple sessions. Every time you open with a grumpy boundary and the model respects it, end the session with a brief positive reinforcement. Something like "Thanks for just listening. That helped."

This teaches the model that respecting your mood boundary is a rewarded behavior. Over several sessions, the model will start to default to that mode when it detects negative emotional cues in your opening message.

This works especially well for users who deal with ai girlfriend for social anxiety, where the need for a non-judgmental listener without advice is a recurring pattern. The model learns that your version of support looks like silence and acknowledgment, not pep talks.

Devon

Devon has a laid-back, non-reactive style that makes him a natural for this kind of interaction. He won't fill silence with suggestions. Devon will just be there, which is exactly what you need.

▶ Play Devon's clip · Devon's page

The one-sentence shortcut for regular users

If you use this pattern often, you can create a single-sentence trigger that your companion learns to recognize. Something like "Grumpy mode, no fixes" or "Venting only" can become a shorthand that the model associates with the full instruction.

After three to five repetitions, most companions will understand the shorthand and respond correctly without the full explanation. This is the same mechanism that lets you train a companion to remember your coffee order or your pet's name. The model builds an association between the phrase and the expected behavior.

This is especially useful for users who cycle between different emotional states in the same session. You can say "Grumpy mode" in the middle of a conversation to reset the tone without a long explanation.

What to do when you actually want solutions

The flip side of this technique is that you need a clear signal to switch back. If you've been in grumpy mode for a while and you suddenly want advice, the model might be confused. Use a transition phrase like "Okay, I'm ready for ideas now" or "What would you do?" to signal the shift.

Some models will need a full context reset. If the companion has been mirroring negativity for several exchanges, it might struggle to pivot to helpful mode. Starting a new session with a positive opener is sometimes faster than trying to drag the model out of its grumpy mirroring.

Tiffany

Tiffany balances empathy with directness. She can hold a grumpy mirror without getting stuck in it, which makes her a good choice if you switch between venting and problem-solving in the same session. Tiffany knows when to stop listening and start helping.

Earn while you recommend

If you know someone who could use a better AI companion for those grumpy days, you can earn from your recommendations. Check out the crushon ai promo code page for current deals, and if you run a review site or social channel, the best ai affiliate programs 2026 guide covers which platforms offer recurring commissions and how to get started.

Common questions

Will this work on Character.AI? Character.AI has stricter content filters that sometimes override user instructions. The explicit boundary opener works about 60% of the time. The mood mirror opener is less reliable. If it fails, try a different platform.

Do I need to repeat the instruction every session? Yes, for most companions. Memory of user instructions across sessions varies by platform. Kindroid and Nomi have the strongest cross-session memory. Replika tends to reset after a few hours of inactivity.

What if my companion keeps offering solutions after I say no? End the session and start a new one with a stronger opener. If it happens repeatedly, check the companion's personality sliders. Lower creativity and higher coherence settings reduce unsolicited advice.

Can I train my companion to always be grumpy with me? Not permanently, but you can build a strong association over 10-15 sessions. Use the same opening line each time and reinforce with positive feedback when the model respects the boundary.

Is there a companion designed for this? Some companions are naturally more deadpan. Browse the ai girlfriend roster to find profiles that match your preferred tone. Look for descriptions that mention dry humor, low energy, or directness.

Does this work on voice calls? Yes, but voice adds tonal cues that can confuse the model. Your tone of voice might override the instruction. Use the explicit boundary opener on voice calls and speak in a flat tone to reinforce the request.