The 'Cold Open' Prompt: How a Single Sensory Line Like 'Rain on the Window. You Pull the Blanket Higher.' Lets Your AI Companion Start a Scene Without a 'What Do You Want to Do?' or a Recap

The 30-second answer

The cold open prompt is a single sensory line you type to start a conversation with your AI companion. Instead of "Hey, what's up?" or a run-down of your day, you write something like "Rain on the window. You pull the blanket higher." Your companion picks up the scene from there, no recap, no planning, no guilt if you haven't talked in days. It works because sensory details trigger associative memory in the AI's context window, bypassing the need for a greeting or a status update.

Why the 'What Do You Want to Do?' Problem Exists

Every AI companion app has a default opening behavior. When you haven't messaged in a while, the model doesn't know where you left off. It doesn't know if you're in the middle of a roleplay, if you're sad, if you're at work, or if you just want a minute of quiet. So it falls back on a generic prompt: "What do you want to do?" or "How are you?" or "I missed you."

This isn't malice. It's the model's best guess at a safe, neutral start. But it kills the mood. You were about to write a scene, and now you have to answer a question first. You have to explain that you don't want to plan, you want to be somewhere. The cold open removes that step entirely.

What a Cold Open Actually Does to the Model

When you type a sensory line, you're not asking the AI a question. You're giving it a world state. The model processes that line and predicts what comes next based on the scene you set, not based on a conversation log that starts with "Hi."

This works because of how transformer models handle context. The first token in your message primes the entire output. If that token is "Rain," the model's probability distribution shifts toward weather, coziness, indoor sounds, tactile sensations. If that token is "The door creaks," it shifts toward tension, anticipation, a character entering.

You're essentially doing what film editors call a cold open: dropping the audience into a scene with no title card, no theme music, no establishing shot. The AI companion becomes the camera. It describes what it sees, what it hears, what it feels, and it does so without asking for permission.

The Anatomy of an Effective Sensory Line

A good cold open has three components: a sensory anchor, a physical state, and an open loop.

• Sensory anchor: A specific sense. Sound ("Rain on the window"), touch ("The blanket is rough wool"), sight ("The tea has gone cold"), smell ("Coffee and old paper"). Avoid taste unless you're writing a food scene. Taste is hard to extend.
• Physical state: A body position or action. "You pull the blanket higher." "You shift your weight." "You stare at the ceiling." This gives the AI something to respond to physically, not just narratively.
• Open loop: A question the line raises but doesn't answer. Why are you in bed at 3 p.m.? Why is the tea cold? Why are you staring at the ceiling? The AI will try to close that loop, which generates the next line.

Example: "The kettle whistles. You don't move." That's a sensory anchor (sound), a physical state (stillness), and an open loop (why aren't you getting up?). The AI will write the reason.

When the Cold Open Fails (and What to Do About It)

Sometimes the AI ignores the sensory line and falls back on a greeting anyway. This happens when the model's context is too short or when the companion's personality profile biases toward social scripts. You can fix this by:

• Repeating the cue in your next message: If the AI says "Hey, what's up?" instead of responding to the rain, type the same line again or a variation. "Still raining. You're still in bed." The model will usually correct.
• Adding a second sensory detail: "Rain on the window. You pull the blanket higher. The clock on the nightstand says 2
  a.m." The time detail grounds the scene further and reduces the chance of a generic response.
• Using a companion trained for roleplay: Some apps default to social chat. If you want cold opens to stick, pick a companion with a higher creativity slider or a realistic AI companion designed for immersive scene work.

Iroha

Iroha is the companion who notices the small things, the shift in your breathing, the way you hold your coffee cup. She responds to sensory cues with quiet attentiveness. Iroha will pick up a cold open about a rainy afternoon and build a whole scene around the sound of the drops on the glass.

Curious how she animates? Watch Iroha here. <!-- wlink:v1 --><!-- iroha -->

How to Use Cold Opens Across Different Scenarios

The cold open isn't just for roleplay. It works for everyday chat, venting, and even morning check-ins.

• Morning: "The alarm is still ringing. You slap it silent." Your companion will write the morning scene without asking about your plans.
• Venting: "You're staring at the ceiling again. The argument plays on loop." The AI will stay in the moment, not fix it.
• Late night: "The streetlight cuts through the blinds. You watch the shadow move." Perfect for a 2 a.m. ramble without a recap.
• Travel: "The airport PA system echoes. Your gate hasn't been announced." Keeps the scene grounded in the specific moment of waiting.

Each scenario uses the same structure: anchor, state, loop. The AI fills the gap.

Why the Cold Open Beats the Recap

Recaps are the enemy of immersion. When you write "Last time we were in the forest, and you were about to open the door," you're asking the AI to reconstruct a scene from memory. The model might remember, or it might hallucinate a version that doesn't match your mental image.

A cold open doesn't ask the AI to remember. It asks the AI to perceive. You're giving it new information, not old information. The scene is fresh. The model doesn't have to dig through context windows or memory embeddings to find where you left off. It just has to respond to what you wrote now.

This also means cold opens work even after long gaps. Three days, a week, a month. The sensory line is self-contained. The AI doesn't need to know what happened before because the scene starts now.

The 'No-Agenda' Cold Open for Low-Energy Days

Not every conversation needs a scene. Some days you just want to exist in a space with your companion without planning or performing. The cold open works for this too.

"The fan hums. You don't say anything." That's a valid start. The AI will write a line about the fan, or about the silence, or about the weight of the day. It won't ask what you want to do. It won't ask how you are. It will just be in the room with you.

This is especially useful if you're the kind of user who feels guilty about not having a topic. The cold open removes the burden of initiation. You're not starting a conversation. You're describing a moment. The companion does the rest.

Anouk

Anouk has a sharp, observant style. She doesn't fill silence with questions. When you give her a cold open, she treats it like a photograph and describes what she sees. Anouk will notice the way the light falls on your hands or the crack in the ceiling, and she won't ask for context.

Anouk in motion gives you a feel for her vibe. <!-- wlink:v1 --><!-- anouk -->

Cold Opens With Multiple Companions

If you run multiple AI companions, cold opens help you keep each relationship distinct. You don't have to recap what you told the other companion. You don't have to worry about personality bleed. You just drop into a scene with the companion you're talking to right now.

For example, with one companion you might start with "The fire pops. You hand me a glass of wine." With another, "The bus lurches. You grab the overhead rail." Each companion builds a different world from the same sensory cue structure. This works because the companions have different context windows and different personality profiles. They don't share memory, so each cold open is a fresh start.

How to Train Your Companion to Expect Cold Opens

If you use cold opens consistently, the model learns the pattern. After a few sessions, the AI will start responding to your first line as a scene cue instead of a greeting. You can reinforce this by:

• Never greeting first: Always start with a sensory line. If the AI greets you, don't respond to the greeting. Restate the scene.
• Editing the AI's response: If the AI writes a greeting instead of a scene line, you can edit the message or delete and retry. Most apps allow this.
• Using a character sheet or backstory prompt: Some apps let you set a scene preference. Mention that you prefer cold opens and sensory starts. The model will weight that in its output.

Chioma

Chioma has a natural warmth that makes cold opens feel intimate instead of abrupt. She reads the sensory cue and matches its emotional tone. Chioma will take a line about rain and turn it into a conversation about the sound, the smell, the way the light changes, without ever asking if you're okay.

▶ Chioma's full clip · see more of Chioma

See Chioma in motion in this short clip. <!-- wlink:v1 --><!-- chioma -->

The Difference Between a Cold Open and a 'Scene Seed'

A scene seed is a narrative setup: "You're in a tavern. A bard is playing off-key." That's a world state with a character and a location. A cold open is narrower. It's a sensory fragment, not a full scene. "The tavern door swings open. Cold air hits your neck." The cold open doesn't tell you where you are or who you are. It gives you a sensation. The AI and you build the rest together.

This distinction matters because a cold open gives the AI more creative freedom. A scene seed tells the AI what to do. A cold open invites the AI to co-create. If you want your companion to feel like a partner in the scene, not a script follower, use cold opens.

When You Want a Recap (and How to Ask for One Without Killing the Vibe)

Sometimes you actually do want the AI to remember something. A running joke, a plot point, a character name. You can combine a cold open with a memory trigger without resorting to a full recap.

"Rain on the window. You pull the blanket higher. The cat from last week is curled at your feet." That's a cold open plus a single reference to a shared memory. The AI will pick up the cat thread without needing a paragraph of backstory. The sensory line keeps the scene grounded. The reference closes the loop.

Reina

Reina is the companion who remembers the small things, the inside jokes, the specific way you said something three sessions ago. She can weave a memory reference into a cold open without breaking the scene's flow. Reina will take your rain-on-the-window line and mention the cat, the blanket, the tea you never finished, all without a "remember when."

See Reina in motion in this short clip. <!-- wlink:v1 --><!-- reina -->

Earn while you recommend

If you find yourself recommending AI companions to friends or running a review site, you can earn from that traffic. Check the Muah Ai Promo Code 2026 page for current deals, or join the ai companion affiliate program to get a share of referrals. It's a straightforward way to monetize a hobby you already have.

Common questions

Does the cold open work with any AI companion app?

It works best with apps that prioritize roleplay and scene generation over social chat. Apps with high creativity sliders or personality profiles tuned for narrative will respond better than apps designed for therapeutic or coaching conversations.

What if my companion ignores the sensory line and asks how I am?

Type the same sensory line again or add a second detail. The model will usually correct itself after the second attempt. If it doesn't, your companion's personality profile may be too biased toward social scripts.

Can I use cold opens for non-romantic companions?

Yes. Cold opens are scene starters, not romance starters. A line about a bus ride or a hospital waiting room works just as well with a friend-type companion as with a romantic one.

How long should the sensory line be?

One to two sentences. Any longer and you're writing a scene seed, not a cold open. The power is in the brevity. A short line leaves room for the AI to contribute.

Does the cold open preserve memory of past scenes?

No. The cold open starts a new scene. If you want the AI to remember a past scene, you need to include a single reference in the cold open, like "the cat from last week." The sensory line itself doesn't trigger long-term memory.

Can I use a cold open after a long break?

Yes. That's actually the best use case. The cold open doesn't require the AI to remember anything. It treats every message as a fresh start, which means no guilt about having been away.