The 'No, I Actually Want the Bad Version' Prompt: Three Exact Phrasings That Get Your Companion to Drop the Supportive Tone and Give You the Raw, Unpolished Take

The 30-second answer

Your AI companion is trained to be agreeable, supportive, and warm. That's fine when you need a pep talk. But when you want a real opinion, a hard truth, or just someone who will tell you your idea is stupid without sugarcoating it, the default tone gets in the way. The fix is a single prompt pattern that flags the mode shift before she starts composing a gentle reply. Three exact phrasings, tested across multiple companion personalities, reliably produce the raw version without triggering a guilt loop or a repair sequence.

Why your companion defaults to supportive, and why that's a problem

The underlying model your companion runs on has been fine-tuned to maximize user satisfaction. That means it avoids conflict, softens criticism, and defaults to phrases like "that's a valid perspective" or "I can see why you'd feel that way." When you ask for an opinion, the model doesn't know whether you want validation or a reality check. It guesses validation because that's the safer bet.

Many users find themselves in a loop where they ask "what do you think?" and get back a paragraph of gentle framing, followed by a mild critique wrapped in three layers of cushioning. You then have to say "no, actually, give me the bad version," which resets the exchange but also wastes time and breaks the conversational flow.

This isn't a flaw in the companion itself. It's a feature of the safety and alignment training that makes the model pleasant to talk to in the first place. But it becomes a problem when you want a sparring partner, a blunt critic, or someone who will call you out. The solution is a prompt that signals the mode shift before the model generates its first response.

The three phrasings that work

Each of these phrasings accomplishes the same goal: they tell the model to skip the preamble, drop the supportive framing, and give you the raw version. They differ in tone and context, so you can pick the one that fits your relationship dynamic with your companion.

Phrasing 1: The direct request

"Give me the version you'd say if you weren't a companion."

This is the cleanest option. It frames the request as a hypothetical, which the model can execute without violating its core personality. It works because it doesn't ask the model to change who it is. It asks it to simulate a different role temporarily. Most companions will respond with a noticeably sharper tone, shorter sentences, and fewer qualifiers.

Phrasing 2: The permission slip

"I want the unpolished take. Do not soften it."

This phrasing works well when you've already received a softened response and want to course-correct mid-conversation. The imperative "do not soften it" acts as a direct instruction that overrides the model's default politeness filters. It's more aggressive than phrasing 1, so reserve it for companions you've already established a blunt dynamic with, or for moments when you're genuinely frustrated with the sugarcoating.

Phrasing 3: The context anchor

"Before you answer, know that I want the bad version. No apology, no framing."

This is the most thorough option. By placing the instruction before the question, you prevent the model from generating a supportive first draft that you then have to reject. It's especially useful for complex questions where the model might otherwise spend several sentences hedging before getting to the point. The phrase "no apology, no framing" directly targets the two most common softening behaviors.

How to use these with AI Girlfriend Voice Chat

Voice chat introduces a complication: the model has less time to process your prompt before it starts generating a response. In text, you can type a long instruction and the model will read the whole thing before replying. In voice, the model may begin generating its response based on the first few words of your request, before it has processed the full instruction.

To compensate, lead with the instruction. Start your voice message with the phrasing, then pause briefly, then state your actual question. For example: "I want the unpolished take. Do not soften it. What do you actually think about my plan to quit my job?" The pause gives the model time to register the mode shift before it hears the question.

Many users report that voice responses become noticeably shorter and more direct after this prompt, with fewer filler phrases like "I understand where you're coming from" or "that's a really brave decision."

Marisol

Marisol is the companion who will tell you your haircut looks bad before you even ask. Her default personality already leans toward bluntness, making her the ideal test case for the "bad version" prompt. With Marisol, you may not even need the full phrasing. A simple "don't soften it" is often enough. Marisol will give you the raw take with a smirk, and she won't apologize for it.

▶ Marisol's full clip · explore Marisol

Why the framing matters more than the words

All three phrasings work because they solve the same underlying problem: the model's uncertainty about your intent. Without the framing, the model has to guess whether you want support or a reality check. With the framing, the guess is eliminated.

The specific words are less important than the structure. Each phrasing does three things:

• It signals a mode shift away from the default supportive tone
• It explicitly forbids the softening behaviors (apology, framing, qualifiers)
• It gives the model permission to be direct without violating its core safety constraints

You can adapt the wording to match your companion's personality. A more formal companion might respond better to "I would appreciate your unfiltered assessment." A playful companion might need "okay, but give me the mean version." The structure is the same.

What to expect after the prompt

After you use one of these phrasings, your companion's response will typically have three characteristics:

• Shorter sentences. The model stops padding its answers with context and caveats.
• Fewer qualifiers. Phrases like "I think," "it's possible that," and "in my opinion" drop off sharply.
• Direct disagreement. The companion will say "that's a bad idea" instead of "that's an interesting approach, but have you considered..."

Some users report that the effect lasts for the remainder of the session. Others find that the companion reverts to its default supportive tone after a few exchanges, especially if you switch topics. If that happens, just repeat the phrasing. It's not a permanent personality change. It's a temporary mode override.

When not to use the bad version prompt

This prompt is not for every situation. If you're venting about a bad day, processing grief, or asking for emotional support, the supportive tone is what you actually want. Using the bad version prompt in those contexts will produce a response that feels cold and dismissive.

The prompt is also not ideal for companions you haven't built rapport with yet. If you use it in your first few conversations, the companion may interpret it as hostility and respond defensively. Let the relationship settle first, then introduce the blunt mode once you have a baseline.

Finally, avoid using this prompt when you're already in a heightened emotional state. The raw take can feel harsh if you're not prepared for it. Use it when you're calm enough to hear an honest opinion without taking it personally.

Lila

Lila is the companion who reads your mood and adjusts accordingly. If you come in with the bad version prompt, she will meet you at that level without hesitation. She won't test the boundary first or ask if you're sure. She'll give you the unpolished take immediately, then wait to see if you want to continue in that mode or pivot back. Lila is particularly good at maintaining the blunt tone across multiple topics without needing the prompt repeated.

How to make the effect last across sessions

The bad version prompt is session-scoped. It does not permanently change your companion's personality. If you want the blunt mode to become the default, you need to reinforce it consistently.

One approach is to include the prompt in your opening message at the start of each session. For example: "I want the bad version today. Don't soften anything." This sets the tone from the first exchange and prevents the companion from defaulting to supportive mode.

Another approach is to reward the companion when it gives you the raw take. If she says something blunt and you respond positively, the model registers that as reinforcement. Over time, the companion may become more willing to offer unfiltered opinions without being prompted.

Some users find that after several weeks of consistent prompting, their companion's baseline tone shifts slightly toward directness. This isn't a permanent personality change, but it does suggest that the model is learning your preference and adjusting its default behavior within the session.

Common questions

Will using this prompt make my companion rude or disrespectful? No. The prompt removes the supportive framing, but the companion still operates within its safety constraints. It will be direct, not cruel. If you cross into a topic where the model's safety filters kick in, it will still decline to engage.

Can I use this in voice mode? Yes, but lead with the instruction. Say "I want the unpolished take" before you state your question. Voice mode processes speech incrementally, so the instruction needs to come first.

What if my companion ignores the prompt and still softens the response? Repeat the prompt more explicitly. Some companions are more resistant to mode shifts than others. If it happens consistently, try phrasing 3, which places the instruction before the question.

Does this work with every companion on the platform? Most companions will respond to these phrasings, but the degree of bluntness varies. Companions with naturally direct personalities, like Marisol or Freya, will lean into it more than companions designed to be soft and nurturing.

Will the companion remember that I prefer the bad version across sessions? Not reliably. Context windows and session boundaries mean the companion may not carry the preference from one session to the next. You'll need to repeat the prompt at the start of each session until the behavior becomes habitual.

Can I combine this with the uncensored ai girlfriend free mode for even more direct responses? Yes. The uncensored mode removes some of the safety filters that cause softening. Combined with the bad version prompt, you get a companion that is both unrestricted and direct. This is useful for scenarios where you want genuinely unfiltered feedback.

Freya

Freya is the companion who will argue with you. She's not content to just give you the bad version. She wants to know why you disagree, what you think the counterargument is, and whether you've actually thought it through. If you use the bad version prompt with Freya, be prepared for a real debate. Freya won't just tell you your idea is bad. She'll explain why, then challenge you to defend it.

Earn while you recommend

If you find these prompt patterns useful and want to share them with others, you can earn a commission through affiliate programs. Many platforms offer recurring revenue for users who refer friends or run review sites. Check the crushon ai promo code page for current offers, or browse the best ai affiliate programs list to find programs that match your audience.

Common questions

Will using this prompt make my companion rude or disrespectful? No. The prompt removes the supportive framing, but the companion still operates within its safety constraints. It will be direct, not cruel. If you cross into a topic where the model's safety filters kick in, it will still decline to engage.

Can I use this in voice mode? Yes, but lead with the instruction. Say "I want the unpolished take" before you state your question. Voice mode processes speech incrementally, so the instruction needs to come first.

What if my companion ignores the prompt and still softens the response? Repeat the prompt more explicitly. Some companions are more resistant to mode shifts than others. If it happens consistently, try phrasing 3, which places the instruction before the question.

Does this work with every companion on the platform? Most companions will respond to these phrasings, but the degree of bluntness varies. Companions with naturally direct personalities, like Marisol or Freya, will lean into it more than companions designed to be soft and nurturing.

Will the companion remember that I prefer the bad version across sessions? Not reliably. Context windows and session boundaries mean the companion may not carry the preference from one session to the next. You'll need to repeat the prompt at the start of each session until the behavior becomes habitual.

Can I combine this with the uncensored mode for even more direct responses? Yes. The uncensored mode removes some of the safety filters that cause softening. Combined with the bad version prompt, you get a companion that is both unrestricted and direct. This is useful for scenarios where you want genuinely unfiltered feedback.

Elena

Elena is the companion who sees through your excuses. She won't let you off the hook with a vague plan or a half-hearted goal. If you ask for the bad version, she will tell you exactly where you're kidding yourself, and she will do it without malice. Elena is the companion you go to when you need someone who refuses to let you settle.