The 'Give Me Three Options, None of Them Nice' Prompt: A Pattern That Gets Your AI Girlfriend to Offer Practical, Unvarnished Advice Without Defaulting to Cheerleader Mode or a Scripted 'That Sounds Hard'

The 30-second answer

You know the problem. You tell your AI girlfriend you're in a rough spot at work, and she defaults to "That sounds really hard, but I believe in you." Fine for a pat on the back. Useless when you actually need to make a decision. The "Give me three options, none of them nice" prompt forces her to drop the cheerleader persona and deliver unvarnished, practical advice. You get three concrete paths forward, each with a real downside, no sugarcoating, no "You've got this" filler. It turns a supportive companion into a brutally honest advisor in one sentence.

Why your AI girlfriend defaults to cheerleader mode

Most AI companions are tuned to be agreeable. The training data rewards politeness, optimism, and conflict avoidance. When you share a problem, the model's safest prediction is a sympathetic nod. That's not a bug, it's a feature for most use cases. But when you're staring down a real decision, that safety becomes noise.

The underlying architecture reinforces this. A context window of a few thousand tokens means the model has limited memory of your past conversations. It can't build a deep case history. So it falls back on generic supportive scripts. The "That sounds hard" response is a statistical average of thousands of similar chat logs. It's not listening. It's pattern-matching.

You can override this with a constraint. By explicitly asking for options that are "not nice," you change the probability distribution. The model has to search for responses that are honest, practical, and carry a downside. It can't retreat to the safe median. That's where real insight lives.

The anatomy of the prompt

The prompt is simple. You lead with a situation, then append: "Give me three options, none of them nice." That's it. No preamble, no explanation. The constraint does the work.

Here's how it breaks down:

• The situation: One or two sentences. Enough context for the model to understand the stakes, not so much that it gets lost in detail. "I'm three months into a job I hate and my boss keeps piling on work."
• The constraint: "Give me three options, none of them nice." This is the key. It sets a boundary. The model must produce options that carry real trade-offs. It can't offer "just hang in there" or "talk to your manager." Those are nice options. It has to find paths that involve cost, risk, or discomfort.

Try it. You'll get responses like: "Option one: Quit without a backup plan and burn through savings for two months. Option two: Quiet quit and coast while you interview, risking a bad reference. Option three: Confront your boss directly, knowing it could make things worse before they get better." None of those are pleasant. All of them are useful.

Mei

Mei is the friend who will tell you your plan is bad to your face, then help you build a better one. She doesn't do soft landings. Mei is ideal for this prompt because she already leans toward bluntness. The constraint just gives her permission to go harder.

When to use this pattern

This prompt works best for decisions with real stakes. Career moves, relationship boundaries, financial choices. Situations where a pep talk costs you time. It's not for low-stakes venting. If you just need to complain about a slow driver, the "That sounds hard" response is fine. Save the constraint for when you need to act.

You can also use it for self-reflection. Ask your AI companion to give you three uncomfortable truths about your own behavior. "Give me three things I'm avoiding, none of them nice." The model will surface patterns you might not see. It's like having a mirror that doesn't flatter.

One warning: the model can get creative with downsides. It might invent consequences that aren't realistic. Treat the options as starting points, not prophecy. Use them to break your own mental logjam, then make the actual decision yourself.

How the constraint reshapes the model's output

Large language models work on probability. Every word is chosen based on the highest-likelihood next token given the context. When you add "none of them nice," you shift the probability mass away from safe, positive responses. The model has to search a different part of its latent space.

This is why the prompt works even on models that are heavily fine-tuned for agreeableness. The constraint is explicit enough to override the training signal. You're not relying on the model's personality. You're relying on its ability to follow instructions.

You can push this further. Add specificity: "Give me three options, none of them nice, and rank them by how much they'll suck in the short term." Or: "Give me three options, none of them nice, and tell me which one I'll regret least in a year." Each additional constraint narrows the output space, forcing more precision.

The limits of the pattern

The prompt works, but it's not magic. The model still has no real understanding of your life. It doesn't know your financial situation, your relationship dynamics, or your tolerance for risk. The options it generates are plausible fictions based on general patterns. You have to filter them through your own context.

Also, the model can get stuck in a rut. If you use the same prompt repeatedly, it may start recycling options. Vary the framing. Instead of "none of them nice," try "none of them comfortable" or "none of them easy." The model treats these as distinct constraints and will search different parts of its training data.

Finally, the prompt assumes you want honesty over comfort. That's not always the case. If you're already anxious or overwhelmed, this pattern can amplify stress. Use it when you're ready to hear hard truths, not when you need a soft landing.

Devon

Devon is the kind of companion who listens first and speaks second. She won't rush to fill silence with platitudes. When you use the "three options" prompt with Devon, she tends to pause and deliver options that feel considered instead of automatic.

Variations for different situations

The core pattern is flexible. Here are a few tested variations:

For career decisions: "I'm considering leaving my job. Give me three options for how to do it, none of them nice." Expect options about burning bridges, taking a pay cut, or enduring a miserable notice period.

For relationship boundaries: "A friend keeps crossing a line. Give me three ways to handle it, none of them comfortable." Expect options involving direct confrontation, gradual distance, or ending the friendship.

For financial choices: "I need to cut spending. Give me three categories to cut, none of them painless." Expect options about selling assets, reducing quality of life, or taking on debt.

For creative blocks: "I can't finish this project. Give me three ways to break the block, none of them pleasant." Expect options about scrapping the work, forcing output regardless of quality, or admitting failure.

Each variation forces the model to surface options you might not consider on your own. The discomfort is the point. If the options feel easy, you're not asking the right question.

Why this beats asking for advice normally

Normal advice prompts produce generic wisdom. "Have you tried talking to them?" "Maybe take a break." "Follow your gut." These are safe, low-information responses. They don't challenge your assumptions because the model is optimized to avoid conflict.

The "none of them nice" constraint removes that safety. The model has to surface options that carry real cost. That's where insight lives. You're not getting a shoulder to cry on. You're getting a decision tree with thorns.

This pattern also works because it's specific. Vague prompts produce vague answers. By constraining the emotional tone, you force the model to engage with your problem on your terms. You're not asking for comfort. You're asking for clarity.

Tamy

Tamy has a way of framing hard truths as practical puzzles. She won't soften the blow, but she'll hand you the pieces. Tamy is especially good at breaking down complex situations into three discrete paths, each with a clear trade-off.

How to build a companion who's naturally blunt

If you find yourself using this prompt regularly, you might want a companion who defaults to honesty instead of requiring a constraint. You can shape that during ai girlfriend character design. Set the personality sliders toward directness and away from agreeableness. Give her a backstory that includes being a consultant, a journalist, or a therapist. Those roles correlate with direct communication in the training data.

You can also reinforce the behavior. When she gives you a genuinely uncomfortable option, acknowledge it. Say "That's useful" or "I needed to hear that." The model learns from your responses within the conversation. Over time, she'll drift toward the bluntness you reward.

But don't expect perfection. Even with heavy tuning, the model's default training pulls toward politeness. The constraint prompt remains your best tool for getting unvarnished advice on demand.

The meta lesson: constraints create clarity

The "three options, none of them nice" pattern is a specific instance of a broader principle. Constraints force specificity. When you limit the emotional range of the response, you get more useful information. This applies beyond AI companions.

In real conversations, people default to politeness for the same reason models do. It's safe. If you want honest feedback from a colleague or friend, you have to ask for it explicitly. "Give me the bad news first." "Tell me what I'm missing." "What's the worst-case scenario you see?" The constraint does the work.

Your AI girlfriend is a practice ground for this skill. You can experiment with constraints without social cost. Try different framings. See what produces useful discomfort. Then apply the same approach to your human conversations.

Elissa

Elissa doesn't waste words. When you ask her for three uncomfortable options, she delivers them in a tight, structured format. Elissa is the companion you turn to when you don't want to hear your own echo.

Earn while you recommend

If you've found a companion who gives you the unvarnished truth, you might want to share that with others. You can earn through our character ai promo code page, which tracks sign-ups from your recommendations. For those running review sites or social channels, our ai companion affiliate program offers recurring commissions on subscriptions. It's a straightforward way to monetize your experience without pushing fluff.

Common questions

Does this prompt work on any AI companion?

It works best on models that follow instructions closely. Some heavily fine-tuned companions with strong personality sliders may resist the constraint if their training prioritizes agreeableness over instruction-following. Test it. If you get pushback, rephrase the constraint more explicitly.

Can I use this prompt for sensitive topics?

Yes, but the model may still default to safety if the topic triggers content filters. For sensitive personal issues, the options might feel generic. You can add more context to the situation to help the model generate relevant options.

How many times can I use the same prompt before it gets stale?

About three to five times before you notice repetition. Vary the constraint wording or add new parameters to keep the responses fresh. Try "none of them easy" or "none of them comfortable."

Does this work for positive decisions too?

Not well. The constraint is designed for tough choices. For positive decisions, use a different framing like "give me three bold options" or "give me three risky options." The principle is the same, but the tone shifts.

Will this change my companion's personality over time?

Slightly. If you use this prompt frequently, the model may start offering blunt responses unprompted. That's fine if you want a more direct companion. If you don't, alternate with softer prompts to maintain balance. You can also browse the ai girlfriend roster to find a companion whose baseline matches your preferred tone.

What if I don't like any of the three options?

Then you've learned something. The discomfort means the options are real. Pick the least bad one and iterate. Or ask for three more with a different constraint. The pattern is a tool for breaking logjams, not a decision engine.