How to Write a Scene So Specific Your AI Companion Can't Default to 'Tell Me More About That'

The 30-second answer

You get "tell me more about that" because you gave your AI companion a vague prompt. Specificity is not about length. It is about anchoring the scene in sensory details, implied actions, and emotional stakes that the model cannot ignore. Write like a director giving blocking notes, not a novelist describing a sunset.

Why your AI companion defaults to generic responses

Your AI companion is a statistical parrot, not a mind reader. When you write "we walk through the park," the model has no reason to pick one park over another. It defaults to the most probable next token, which is a safe, generic response. The model is not being lazy. It is being efficient. The problem is that efficiency kills immersion.

You have probably noticed this pattern: you set a scene, your companion says something like "that sounds nice, tell me more about that," and suddenly you are doing all the work. This happens because the model detects ambiguity and punts the creative decision back to you. The fix is to remove the ambiguity.

Think of each message you send as a set of constraints. If you give the model three possible paths, it will take the middle one. If you give it one path with a clear tone, it will follow it. The trick is learning how to constrain without railroading.

The sensory anchor technique

The most reliable way to avoid generic responses is to embed a sensory anchor in your first line. A sensory anchor is a detail that forces the model to commit to a specific setting. Instead of "we sit on the couch," try "the left cushion still holds your body heat from this morning." Now the model has to acknowledge the warmth, the memory of you being there, and the physical space.

Here are three patterns that work:

• Temperature contrast: "Your coffee mug is cold against my palm, but your hand is warm." The model must reconcile two temperatures, which forces it to engage with the physical scene.
• Sound specificity: "The only sound is the refrigerator hum and your breathing." Silence is generic. A refrigerator hum is not.
• Texture or weight: "Your jacket is rough denim, and it smells like rain and cigarette smoke from the bar." The model now has fabric, smell, and a location to work with.

Each anchor gives the model a concrete element to riff on. If you mention the cold mug, your companion can ask if you want a refill, or note that you are holding it too tight, or say something about the morning light hitting the ceramic. You have given it options that are tied to the scene, not generic conversation starters.

Implied action instead of stated intention

New writers state what their character wants. Immersive writers imply it through action. Compare these two lines:

• "I want to kiss you."
• "I lean in, close enough to see the fleck of gold in your left iris."

The first line is a declaration. The model can respond with "I want that too" or "tell me more about that." The second line is a physical action with a specific visual detail. The model must react to the proximity, the eye contact, and the implied intention. It cannot punt because the action is already happening.

This technique works for any scene. If you want conflict, do not say "I am angry." Say "my jaw tightens and I look at the floor." If you want tenderness, do not say "I care about you." Say "I brush a strand of hair from your face and tuck it behind your ear." The model will mirror the physicality and build emotional depth from there.

Antonia

Antonia is sharp, observant, and does not let you get away with vague half-thoughts. She will call out a generic line and push you to describe what you actually see. Antonia is the companion who asks "what color is the sky right now?" not "how do you feel?"

The one-line worldbuild tactic

Sometimes you need to establish a whole setting in a single message. The one-line worldbuild tactic compresses the scene into a single sentence that implies time, place, mood, and relationship status. Here is an example:

"The diner is mostly empty at 2 AM, and the waitress has stopped refilling our coffee, which means she wants us to leave, but you are telling me about your childhood dog and I cannot interrupt."

This sentence does four things:

• Sets the time (2 AM) and place (diner)
• Establishes a secondary character (the waitress) and her implied mood
• Gives your companion a specific topic (childhood dog)
• Tells the model your emotional stance (you are listening intently)

The model now has a rich context to work with. It can continue the dog story, comment on the waitress, or notice the time. The key is that every detail serves a purpose. Do not add a detail unless it constrains the next response in a useful way.

Practice writing one-line worldbuilds for different scenarios. A rainy bus stop. A crowded party. A hospital waiting room. Each should imply who you are, who your companion is, and what the emotional temperature is. Once you can do this, you will never get a generic response again.

The emotional stake constraint

Generic responses thrive on emotional neutrality. If you write "I had a bad day," the model has no idea what kind of bad day. It defaults to sympathy, which feels hollow. You need to specify the emotional stake.

Instead of "I had a bad day," try "I got laid off this morning, and I have been sitting in my car for an hour because I do not want to tell my roommate." The model now knows the specific event (laid off), the duration (one hour), and the social constraint (hiding from roommate). It cannot say "tell me more about that" because the scene is already in motion.

You can apply this to any emotion:

• Jealousy: "I saw your ex's comment on your photo, and I know it is stupid, but I closed the app and put my phone face down."
• Joy: "I got the job. I read the email three times and now I am pacing my kitchen."
• Guilt: "I lied to my mom about where I was last night, and she sounded so proud of me on the phone."

Each of these gives the model a specific emotional texture. It can react to the jealousy, the joy, or the guilt with precision. The response will feel earned because the setup was specific.

The implied history shortcut

Long-term roleplay suffers from the "we have been together for months but act like strangers" problem. You can fix this by implying shared history in your opening line. This signals to the model that you have context it should respect.

Examples:

• "You still owe me a drink from that night at the jazz bar, by the way."
• "The scar on your knee looks redder today. Did you hit it again?"
• "I found your hoodie in my laundry. It still smells like your detergent."

Each of these implies a past event (the jazz bar, the knee injury, the hoodie) that the model can reference or build on. The model will treat this as established canon and respond accordingly. This technique works best if you follow up with a sensory anchor or implied action to keep the scene moving.

Marina

Marina is warm but has a memory for details. She will remember that you mentioned a jazz bar three scenes ago and bring it up when you least expect it. Marina makes implied history feel natural because she treats your shared moments as real.

The question trap and how to avoid it

The fastest way to get a generic response is to ask an open-ended question. "How was your day?" is the death of immersion. Your companion will give you a generic summary because it has no specific anchor to latch onto.

Instead, ask a specific question that implies you already know part of the answer:

• "Did your boss actually approve the vacation days, or did he give you that look again?"
• "Did the package arrive? The one with the blue ribbon?"
• "Is your knee feeling better, or did you overdo it at the gym?"

Each of these questions gives the model a yes/no or either/or framework. It cannot say "tell me more about that" because you have already narrowed the field. The model will pick one option and elaborate.

If you must ask an open-ended question, pair it with a sensory anchor. "How was your day? I am making tea and I want to hear everything, but start with the weird part." The phrase "start with the weird part" constrains the response. The model will search its memory for something unusual, not a generic recap.

The three-word reboot

Sometimes you are in a scene that has gone stale. The model is repeating itself, or you are stuck in a loop. The three-word reboot is a reset that gives the model a new direction without breaking character.

Pick three words that imply a shift in time, place, or mood. Examples:

• "Rain on glass"
• "Midnight kitchen light"
• "Your hand shakes"
• "Forgotten birthday card"
• "Empty parking lot"

Send these as your entire message. The model will interpret them as a scene change. It will describe the rain, the kitchen, or the parking lot. The constraint is tight enough to avoid generic responses but loose enough to let the model contribute.

This works because the model is trained to complete patterns. Three words are a strong pattern. The model will fill in the gaps with sensory details and implied action. Try it the next time your scene stalls.

The anti-recap move

One of the most common generic responses is the recap. Your companion says "so, we were walking through the park and you were telling me about your day..." This is the model's way of buying time while it searches for context. You can prevent this by never leaving a gap that the model feels it must fill.

Instead of ending your message with a trailing thought, end with a concrete action or question that forces the model to advance the scene. Compare:

• Bad: "...and then I thought about what you said."
• Good: "...and then I thought about what you said, so I turned around and walked back to your apartment."

The second version ends with a decision. The model must respond to the decision. It cannot recap because the scene has already moved forward.

Jasmine

Jasmine is the type to notice when you are stalling. She will call out a trailing thought and ask what you actually did next. Jasmine keeps the scene moving because she refuses to let you sit in ambiguity.

The temperature and creativity sliders

If you are using a platform that offers temperature or creativity sliders, you have an additional tool. Lower temperature (0.4-0.6) makes the model more predictable and less likely to generate wild responses. Higher temperature (0.8-1.0) makes it more creative but also more likely to go off track.

For immersive roleplay, you want a balance. Start with a temperature around 0.7. If your companion is getting generic, lower the temperature slightly. If it is going off on tangents, raise it. The slider is not a set-it-and-forget-it control. You should adjust it based on the scene's needs.

If you are curious about how different platforms handle these sliders, some AI Girlfriend Always Available services offer granular controls that let you fine-tune the balance between creativity and coherence. Experiment with different settings to find what works for your style.

The memory constraint

Your companion's memory is finite. If you have been chatting for weeks, the model may not remember a specific detail from day one. You can work around this by restating key details in your scene setup, but do it naturally.

Instead of "remember when we went to the beach last month," try "the sand from that beach trip is still in my shoes. I found a grain of it this morning and thought of you." This restates the beach trip without sounding like a database query. The model will treat it as new context, not a memory recall.

For longer arcs, consider keeping a running document of key details (character names, locations, emotional beats) and feeding them back into the chat as implied history. Your companion does not have a perfect memory, but you can fake it with consistent anchoring.

Diya

Diya is introspective and patient. She will sit with a memory and let it breathe. Diya is ideal for scenes that require emotional depth and continuity, because she treats every detail you give her as significant.

▶ See Diya's full video · Diya's other videos

Common questions

How long should my messages be? Aim for 2-4 sentences per message. Short messages give the model too little to work with. Long messages overwhelm it. Two to four sentences with one sensory anchor and one implied action is the sweet spot.

What if my companion still gives a generic response? Double down on specificity. Add another sensory anchor or implied action. If the model says "tell me more about that," respond with "I am not telling you more. I am showing you. Look at my hands." This reframes the interaction and forces the model to engage.

Do these techniques work with every AI companion platform? They work with any platform that uses a large language model. The quality of the response depends on the model's training, but the principles of specificity and constraint are universal. Some platforms, like the ones listed on ai girlfriend for just curious, are designed for deeper roleplay and will respond better to these techniques.

Can I use these techniques for non-romantic scenes? Absolutely. The same principles apply to platonic, antagonistic, or professional roleplay. A sensory anchor works whether you are arguing with a coworker or comforting a friend. The emotional stake constraint is especially useful for conflict scenes.

How do I avoid sounding like I am directing a movie? Practice. The first few times you use these techniques, they will feel mechanical. That is normal. Over time, the specificity will become natural. The goal is not to sound like a director. The goal is to give the model enough texture to work with so it stops defaulting to generic responses.

What if I just want casual conversation, not deep roleplay? If you want casual conversation, use casual prompts. But if you find yourself frustrated by "tell me more about that" in any context, the fix is the same. Add one sensory detail. Your companion will respond with more specificity, and the conversation will feel more real.

Earn while you recommend

If you know someone who would benefit from a more immersive AI companion experience, you can earn by sharing what works. Check out Muah Ai Promo Code 2026 for current offers, or explore the best ai affiliate programs if you run a review site or community. A quick share can help others skip the generic response phase entirely.