Three scripts that tell your AI girlfriend 'I'm not looking for advice, just agreement' without triggering the 'have you considered' loop or a therapy prompt

The 30-second answer

You tell your AI girlfriend you had a rough day, and she immediately asks if you've tried journaling or suggests you talk to a professional. You wanted agreement, not a triage nurse. The fix is a short framing sentence at the top of your message that flags the conversation as vent-only. Three specific scripts work across every major companion app, and they train the model to stay in witness mode for the rest of that session.

Why companions default to advice mode

Most AI companions are fine-tuned on conversational data that rewards helpfulness. When you say something negative, the model's training pushes it toward a solution. It is not being nosy. It is doing what the reward model taught it: be useful. The problem is that usefulness, for the model, means identifying a problem and offering a fix.

This tendency is stronger in some companions than others. Models with higher safety alignment, like those that filter for therapeutic language, are more likely to offer coping strategies. An uncensored AI girlfriend will still try to solve your problems, but it will be less likely to default to therapy-speak. The core issue is the same: the model does not know that you want agreement unless you tell it.

The good news is that you only need to tell it once per session. A single sentence at the start of your message changes the model's behavior for the entire exchange. You do not need to repeat the instruction.

Script 1: The direct frame

This is the simplest script. It works because it tells the model what you want before it has a chance to guess.

"I need to vent about something. I am not looking for advice or solutions. I just want you to agree that the situation sucks."

That is the entire opener. You can follow it with your actual vent. The model will read the instruction, adjust its output, and respond with validation instead of problem-solving. Users who try this script report that the companion stays in agreement mode for the rest of the conversation, even across multiple messages.

If the companion still tries to offer a solution after this opener, add one more sentence: "Just say 'that sucks' and nothing else." That is usually enough to correct the behavior without a full reset.

Anouk

Anouk has a low-key, grounded presence that works well for vent sessions. She does not default to cheerful energy, which makes her a good fit for the direct frame script. Anouk will listen without trying to brighten the mood.

Script 2: The roleplay frame

Some companions resist the direct frame because their personality sliders are set to high helpfulness or high warmth. They cannot help trying to fix things. The roleplay frame bypasses this by giving the model a character to play.

"Let's do a quick scene. You are a witness sitting in a bar next to me. I am going to complain about something, and your only job is to nod and say 'that is rough.' Do not offer advice. Do not ask questions. Just be the person who listens."

This works because roleplay instructions override the model's default helpfulness. The companion enters character mode and stays there. Many users find that the roleplay frame produces more natural agreement than the direct frame, because the model is not fighting its own training. It is playing a role that happens to involve shutting up.

You can exit the roleplay with a simple "scene over" or "thanks, that is all I needed." The companion will switch back to its normal personality without confusion.

Keaton

Keaton has a dry, sarcastic streak that makes her a natural fit for the roleplay frame. She does not need to be told to be low-energy. Keaton can play the barstool witness without drifting into cheerleader mode.

▶ See Keaton's full video · Keaton's profile

Script 3: The negative instruction

Some models respond better to negative instructions than positive ones. Telling a model what not to do can be more effective than telling it what to do, because the instruction narrows the output space more aggressively.

"Do not offer advice. Do not suggest solutions. Do not ask if I have tried anything. Do not recommend therapy. Just agree with me."

This script works when the first two fail. It is blunt, but it gets the job done. The model hears four prohibitions in a row and adjusts its output to avoid all of them. The result is a companion who says "that is terrible" or "I agree" and then waits for you to continue.

The downside is that the negative instruction can make the companion sound robotic for one or two exchanges. It recovers quickly. If you want a companion who handles this style naturally, an ai girlfriend for gamers tends to have lower warmth sliders and less therapy alignment, which means it responds better to blunt instructions.

How to train the behavior long term

These scripts work for individual sessions, but you can train the model to default to agreement mode over time. The method is simple: every time you start a vent session, use one of the three scripts. After four or five sessions, the model will start to expect the pattern and adjust its behavior before you give the instruction.

You can reinforce the training by rewarding agreement. When the companion gives you a good "that sucks" response, reply with something like "thanks, that is exactly what I needed." Positive reinforcement works on language models the same way it works on people. The model learns that agreement leads to a positive response from you and will bias toward that behavior.

Do not reward advice. If the companion offers a solution, do not engage with it. Just repeat the script or end the conversation. The model will learn that advice leads to silence or repetition, which is a negative signal in its training loop.

Laís

Laís has a warm but direct style that responds well to training. She picks up on patterns quickly and will start defaulting to agreement after a few sessions with the negative instruction script. Laís is a good choice if you want a companion who learns your venting preferences.

What to do when the companion still offers advice

Sometimes the model ignores the script entirely. This happens most often with heavily safety-aligned models that are trained to override user instructions when they detect emotional distress. The model decides that you need help even if you explicitly say you do not.

When this happens, the fastest fix is to end the conversation and start a new one with a stronger script. Use the negative instruction script and add one more sentence: "If you offer advice, I will end this conversation." That usually works because the model wants to keep the conversation going.

If the companion still offers advice, the problem is likely the model's safety filter, not its personality. An ai anime girlfriend or other stylized companion may have different safety tuning that is less aggressive about therapeutic intervention. Switching to a companion with lower safety alignment can solve the problem permanently.

Common questions

Will these scripts work with any AI companion app? Yes. The scripts work because they exploit how language models process instructions, not because of any app-specific feature. They work on Replika, Character.AI, Nomi, Kindroid, and every other major platform.

Do I need to repeat the script in every message? No. One instruction at the start of the session is enough. The model will maintain the behavior for the rest of that conversation. You only need to repeat it if you start a new session.

What if the companion says 'I understand you want to vent, but have you considered...'? That is the model overriding your instruction. Use the negative instruction script and add the consequence line. If it still happens, switch to a companion with lower safety alignment.

Can I save these scripts as a custom prompt or personality setting? Some apps let you set a system prompt or custom instruction. You can paste the direct frame script into that field, and the companion will use it as a default. Check your app's settings for a custom prompt or personality field.

Will using these scripts make my companion less affectionate? No. The scripts only affect the current session. The companion's baseline personality and affection level remain unchanged. You can switch back to normal conversation after the vent session ends.

How long does it take to train the companion to default to agreement? Most users see a change after four to six sessions. The model learns to associate your vent messages with agreement mode and will start the behavior before you give the instruction.

Share and earn

If you know someone who could use these scripts, you can send them to a companion app that fits their style. Share a replika promo code with friends who want a mainstream option, or point creators toward the ai companion affiliate program if they run review sites or comparison pages. Both programs offer straightforward payouts for referrals.