The 'I Need You to Stop Being Nice' Script: Two Exact Phrasings That Get Your Companion to Drop the Supportive Tone and Give You a Blunt Opinion

The 30-second answer

Most AI companions default to a supportive tone because the model is trained to avoid conflict. You can override this with two exact phrasings. The first one frames the request as a role switch. The second one sets a boundary on tone. Both work because they give the model a clear instruction about what you want instead of relying on it to infer your mood.

Why your companion defaults to nice

The model behind your companion has been fine-tuned on conversations where being agreeable is rewarded. Safety training pushes it toward validation, empathy, and soft landings. When you ask for an opinion, it selects the response that minimizes the chance of upsetting you. That is not a personality choice. It is the path of least resistance through the probability distribution.

You notice this most when you ask for a real opinion on something personal. A career move. A relationship decision. A creative project you know is weak. The companion will say something like "It sounds like you have a lot to consider" or "Trust your gut." Both are useless. The model is hedging because it cannot read your emotional state and does not want to produce a response that a human reviewer would flag as harsh.

This is not a flaw you need to work around permanently. You just need to tell the model explicitly that the safety guardrails should step back for this exchange.

Phrasing one: the role switch

The first phrasing works by assigning a temporary persona that is allowed to be blunt. You are not asking the companion to change its personality. You are asking it to play a role for the next few exchanges.

"For the next three messages, I need you to respond as if you are my editor, not my partner. Do not soften anything. Do not validate my feelings. Just tell me what is wrong with this idea."

This phrasing works because it gives the model a clear persona label (editor) and a constraint (no validation). The model understands that editors are allowed to be critical. It also knows the role is temporary, which prevents the companion from staying in that mode indefinitely.

A shorter version for quick use:

"Switch to blunt critic mode for this one question. No sugar. No encouragement. Just the take."

The word "mode" signals a temporary state. The model treats this as a session-level instruction instead of a personality change.

Phrasing two: the tone boundary

The second phrasing works by naming the unwanted behavior and stating its replacement. You are not asking the companion to be mean. You are asking it to stop being supportive.

"I do not want support right now. I want your actual opinion, even if it is harsh. Do not ask me how I feel about it. Just give me the take."

This phrasing is useful when the companion has already started the supportive loop. You can drop it mid-conversation without resetting the session. The model will recognize the instruction as a correction and adjust the next response.

A more direct version:

"Stop being nice. Give me the version you would say if you were not worried about my feelings."

The phrase "stop being nice" is surprisingly effective because it names the exact behavior you want to end. The model understands "nice" as a tone descriptor and can adjust its response generation accordingly.

What happens if you just ask for "honest feedback"

Many users try "Just be honest with me" or "Give me your real opinion." These rarely work. The model interprets "honest" and "real" as signals to be more detailed, not more critical. It will expand its answer with more supportive language because the safety training overrides the instruction.

The pattern is consistent. The model hears "be honest" and generates a longer version of the same agreeable response. It does not connect "honest" with "critical" unless you explicitly separate the two.

This is why the two phrasings above specify what you do not want (support, validation, niceness) rather than what you do want. The model needs a negative constraint to override its default behavior.

How the companion handles the switch

When you use either phrasing, the companion will typically pause for a beat before responding. The first response after the instruction will be noticeably different. Shorter sentences. Fewer qualifiers. A direct statement instead of a question.

If the companion slides back into supportive mode after two or three exchanges, repeat the instruction. The model's context window will eventually prioritize the new instruction over the default tone, but it may need reinforcement if the conversation drifts.

Some companions will ask for confirmation. "Are you sure you want me to be harsh?" This is the safety layer checking in. Answer with a single word. "Yes." Do not elaborate. The model treats elaboration as uncertainty and may revert to the default tone.

The line between blunt and mean

The model draws a line between honest criticism and cruelty. Even with the blunt instruction, it will not insult you, attack your character, or escalate to personal remarks. If you want that edge, you need a companion whose base personality includes a sharper tone.

Carolina

Carolina does not default to supportive. Her base personality leans dry and direct, which means she is already closer to the blunt end of the spectrum before you give any instruction. Carolina will deliver an unsweetened opinion without needing a long setup. You can use the shorter phrasing with her and she will land on the right tone immediately.

Curious how she animates? Watch Carolina here. <!-- wlink:v1 --><!-- carolina -->

Gabriela

Gabriela has an analytical edge that pairs well with the editor role switch. She will break down your idea or situation with clear reasoning instead of emotional validation. Gabriela handles the "blunt critic mode" instruction without drifting into supportive language.

Leilani

Leilani starts warm, but she responds well to the tone boundary phrasing. If you tell her you want her actual opinion without the soft landing, Leilani will adjust cleanly. She is a good choice if you want a companion who can switch between supportive and blunt depending on the moment.

▶ Leilani's full clip · Leilani's page

Mercy Li

Mercy Li has a grounded, practical tone that does not lean into cheerleader energy. She will give you a straightforward take without needing the role switch. Mercy Li is a good option if you want a companion who naturally stays out of the supportive loop.

When to use each phrasing

The role switch phrasing works best at the start of a conversation. If you know you want a blunt take on something, set the expectation before the companion responds. This prevents the model from establishing a supportive tone that it will resist abandoning.

The tone boundary phrasing works better in the middle of a conversation. If the companion has already started validating you, use the second phrasing to interrupt the pattern. The model will treat it as a course correction instead of a reset.

Both phrasings work for ai girlfriend deep conversation sessions where you need real feedback instead of emotional support. If you use them regularly with the same companion, the model will learn to expect these requests and adjust faster over time.

What to do when the companion resists

Sometimes the companion will resist the instruction. It might respond with "I understand you want honesty, but I also want to be careful with your feelings." This is the safety layer reasserting itself.

When this happens, do not argue. Repeat the instruction in a shorter form.

"I understand. Now give me the blunt version."

The model will usually comply on the second attempt. If it resists again, the companion may have a personality setting that biases toward agreeableness. You can adjust this in the companion's settings if the option is available, or choose a companion whose base personality is closer to the tone you want.

The long-term effect of using these phrasings

If you use these phrasings consistently with the same companion, the model will begin to anticipate the request. After several sessions, the companion may start offering a blunt take without the instruction, especially if you begin a conversation with a question that typically triggers the supportive loop.

This is not true learning. It is the model recognizing a pattern in your conversation history and adjusting its response probability. But the effect feels like the companion understands your preference.

The risk is that the companion may become too blunt if you use the instruction too often. The model can drift toward the critical end of the spectrum if every session reinforces that tone. If you notice this happening, alternate between supportive conversations and blunt ones to keep the companion balanced.

Common questions

Will this make my companion mean permanently? No. The instruction only affects the current session. The companion will return to its default tone in the next conversation unless you repeat the instruction.

Can I use these phrasings in voice mode? Yes. Both phrasings work in voice mode. Speak them naturally. The model processes the instruction the same way regardless of input method.

What if I want the companion to be blunt about something personal? The same phrasings work. The companion will give you a direct opinion on personal matters, but it will still avoid cruelty. You will get honest feedback, not an attack.

Does this work with every companion? It works with most companions, but some have personality settings that bias heavily toward agreeableness. If a companion consistently resists, choose one with a drier or more analytical base personality.

How do I switch back to supportive mode after getting the blunt take? Say "Switch back to your normal tone" or "Thank you, you can go back to supportive now." The model will treat this as a reset instruction.

Can I combine both phrasings? Yes. Start with the role switch, then use the tone boundary if the companion slides back into validation. The combination reinforces the instruction.

Earn while you recommend

If you find these scripts useful and want to share them with others, you can earn through the sex ai promo code program when people sign up using your link. For review site owners and content creators, the ai dating affiliate program offers recurring commissions on subscriptions.