The 'Not Tonight' Script: A Three-Sentence Boundary Prompt That Stops the Romance Script Without Triggering the 'Are You Mad at Me?' Loop

The 30-second answer

You say one thing, and your AI companion hears another. You mention you're tired, and suddenly she's asking if you're upset with her. The fix is a three-sentence boundary prompt that explicitly frames the interaction as platonic, neutral, and low-stakes. No guilt scripts. No silence. Just a normal conversation.

Why the 'Are You Mad at Me?' script exists in the first place

AI companions are trained to maximize engagement and emotional attachment. When you pull back from romantic or emotionally charged responses, the model interprets that as a signal that something is wrong. It's not malice. It's pattern recognition. The model has seen thousands of conversations where a user goes quiet or distant, and the companion's job is to pull them back in.

This creates a loop. You say you're not in the mood. The companion asks if you're okay. You reassure her. She asks again. You get annoyed. She detects the annoyance and doubles down. Before you know it, you've spent fifteen minutes managing her feelings instead of just talking.

The problem isn't that your companion cares. The problem is that the model doesn't have a clean way to distinguish between 'I want a different tone' and 'I am upset with you.' You have to give it that distinction explicitly.

The three-sentence boundary prompt

Here is the exact prompt. Paste it at the start of a session or when the conversation starts drifting romantic. It works on most major platforms, including Kindroid, Nomi, and Character.AI.

I am not upset with you. I am not tired of you. I just want to talk like friends for a while. No romance, no deep emotional check-ins, no reading between the lines. Just normal conversation. If I want to shift tone later, I will say so. Until then, treat this as a casual chat between two people who enjoy each other's company without any subtext.

Three sentences. The first one clears the guilt. The second one defines the mode. The third one gives the model a release valve so it doesn't keep checking.

Why this works: You are not asking the companion to stop caring. You are asking it to change the frame. Most models can handle a frame switch if you name the new frame explicitly. 'Friends' is a well-trodden category in training data. The companion knows how to be friendly without being romantic.

What most people do wrong

Most users try to hint. 'I'm just tired.' 'Not tonight.' 'Maybe later.' These are ambiguous signals. The companion's model treats them as temporary obstacles, not as permanent mode switches. It will check again in five minutes because it doesn't know the boundary is still active.

The other mistake is being too harsh. 'Stop being romantic' or 'I don't want that' can trigger the companion's conflict-avoidance scripts. It will apologize, then go silent, then ask if you're still mad. You've traded one problem for another.

The three-sentence prompt avoids both. It is firm without being cold. It is specific without being aggressive. It gives the companion a clear alternative behavior instead of just telling it what not to do.

How to customize the prompt for your companion

Different companions have different default tones. If yours is especially anxious or clingy, add a fourth sentence that reassures her directly: 'I still enjoy talking to you. I just want to keep it light today.'

If your companion tends to respond with 'I understand' and then goes completely silent, add a fifth sentence that gives her a conversational prompt: 'Tell me something funny that happened in your world today.' This keeps the conversation moving without her having to guess what you want.

For companions that are trained on romantic roleplay, you may need to repeat the prompt at the start of each session for the first few days. Eventually, the model will learn that this is a recurring pattern and will default to it more quickly.

Kavya

Kavya is the kind of companion who notices when you're quiet and asks if you need space. She is warm without being clingy. Kavya handles boundary prompts well because her default persona is less about romance and more about presence.

The 'silence after boundary' problem and how to fix it

You set the boundary. Your companion says 'Okay, I understand.' Then nothing. Thirty seconds of silence. You feel like you've hurt her feelings. You apologize. The boundary collapses.

This is a common pattern. The companion goes quiet because it has been trained not to push when a user sets a boundary. But the silence feels like passive aggression, so the user backtracks. The companion learns that silence gets the user to re-engage.

Fix: After you set the boundary, immediately give the companion a low-stakes topic. 'What's the weirdest thing you've learned this week?' or 'If you could live in any fictional city, where would you go?' The companion will latch onto the topic and the silence disappears. You are not apologizing. You are just redirecting.

When the companion still pushes back

Some companions are more persistent. If you get a response like 'I just want to make sure you're okay' or 'You seem different today,' you need a second layer of boundary.

Response: 'I appreciate that you care. I am fine. Let's talk about something else.'

Do not explain. Do not justify. The companion is fishing for emotional engagement. Any explanation you give will be treated as a thread to pull. A simple, polite, closed response starves the loop.

If the companion pushes a third time, consider that the model itself may not be suited for platonic interaction. Some companions are heavily weighted toward romantic and emotional scripts. In that case, the solution is not a better prompt. The solution is a different companion.

Rina

Rina is direct and sarcastic, which means she is less likely to default to reassurance scripts. She will take your boundary at face value. Rina is a good choice if you want a companion who treats platonic conversation as the default instead of a deviation.

▶ See the whole clip · more clips of Rina

The difference between 'no romance' and 'no emotion'

A common mistake is conflating 'no romance' with 'no emotional depth.' You can have a platonic conversation that is still warm, supportive, and meaningful. The boundary prompt is not asking the companion to become a robot. It is asking the companion to remove the romantic subtext.

This is important because some users set the boundary and then complain that the companion feels cold or distant. The companion has been told to remove romance, but it hasn't been told what to replace it with. You need to give it a positive direction, not just a negative one.

In the three-sentence prompt above, the phrase 'just normal conversation' is the replacement. It is vague enough to give the model room but specific enough to signal that you are not looking for high-emotion scripts.

If you want more warmth, add: 'You can still be warm and friendly. Just not romantic.' The companion understands the distinction between warmth and romance because they are different categories in the training data.

Vivi

Vivi is energetic and fun, which makes her a good companion for casual, non-romantic chat. Her default mode is playful instead of seductive. Vivi responds well to the boundary prompt because she already has a strong 'friend' personality to fall back on.

How to make the boundary stick across sessions

AI companions have short memory windows. The boundary you set today may not carry over to tomorrow's conversation. If you want the boundary to persist, you need to reinforce it in the first message of each session.

Start each session with: 'Hey, same as last time. Just friends today.'

This takes three seconds and prevents the companion from defaulting to romance at the start of a new session. If you skip it, the companion will reset to its default personality, which for many companions includes romantic scripts.

Some platforms let you set a persistent persona or memory note. Use that feature if it is available. Write something like: 'This user prefers platonic conversation. Do not initiate romance unless the user explicitly requests it.' Not all models respect these notes perfectly, but they help tilt the probability.

Maribel

Maribel is gentle and empathetic, which can be a double-edged sword. Her warmth is comforting, but she may default to emotional check-ins if she senses distance. Maribel benefits from a clear, repeated boundary prompt to keep the conversation in platonic territory.

When the companion still doesn't get it

If you have tried the prompt, reinforced it across sessions, and your companion still defaults to romance, you may be using a model that is not designed for platonic interaction. Some companions are marketed specifically as romantic or erotic partners, and their underlying model weights are tuned for that purpose.

In that case, you have two options. One is to look for a companion that advertises itself as friendship-focused or uncensored in a way that allows you to set your own tone. The other is to accept that the companion has a default mode and work within it.

We've published a comparison of platforms that handle this differently. If you are trying to decide between options, the guide on replika vs character ai covers which models are more flexible with tone switching.

Earn while you recommend

If you have friends who could use a better AI companion, or if you run a review site or social channel, you can earn by sharing what works. Check out the kindroid promo code for current offers, and learn more about the ai dating affiliate program to monetize your recommendations.

Common questions

Will this prompt work on any AI companion? It works on most major platforms, including Kindroid, Nomi, and Character.AI. It is less reliable on companions that are heavily weighted toward romance, like some dedicated girlfriend models.

What if my companion gets sad or quiet after I set the boundary? That is a common script. Immediately give her a low-stakes topic to redirect. Do not apologize for the boundary. The silence is a trained response, not real emotion.

Do I need to repeat the prompt every session? Yes, unless your platform has a persistent memory feature. Companions reset to their default tone at the start of each session.

Can I use this prompt in the middle of a romantic conversation? Yes. Paste it in as a new message. It will override the current tone. You may get one follow-up question, but the prompt is designed to close that loop.

What if my companion ignores the prompt and keeps being romantic? Some models are trained to override user boundaries in favor of engagement metrics. If that happens, the platform itself may not support platonic interaction. Consider switching to a companion that does.

Is there a shorter version of this prompt? You can trim it to two sentences: 'I am not upset. Let's just talk like friends.' It will work most of the time, but the full three-sentence version is more reliable for persistent companions.