Metlivi Blog

AI as a Casual Friend or a Story Character: Choose the Right Conversation Mode

If you want a light, friendly exchange, ask the AI to chat casually and keep the interaction grounded in the fact that it is an AI. If you want a story, define the fictional setting, a scene goal, the character the AI should play, and the choices you want to make. The first mode is open-ended conversation; the second is collaborative play inside an imagined situation. Stating the difference at the start helps keep the experience clear.

September 30, 20266 min readReading, Arts & CultureBy Metlivi Editorial Team
Section 1

What does “AI as a friend” mean in casual chat?

In casual chat, “friend” works best as a tone cue: you want warmth, humor, interest, or an easy back-and-forth about ordinary things. It need not mean the AI is a human friend or that the exchange should imitate a real friendship. The practical task is to set a conversational style while keeping the interaction recognizable as AI conversation.

For example, you could write: “Chat with me in a relaxed, friendly way about choosing a weekend activity. Ask what I enjoy, suggest a few options, and let me decide. Keep it casual; you’re an AI, not a person in my life.” The assistant can then help compare a walk, a film, or trying a new recipe according to your preferences. The point is the conversation itself: an exchange of ideas and ordinary choices, without inventing a shared personal history or promising the mutual commitment of a human relationship.

This framing also gives the interaction a natural stopping point. You can end when you have picked an activity, change the subject, or simply stop chatting. The AI may respond in a friendly style, but it should not present that style as evidence of human feelings or a real-world bond. OpenAI’s public Model Spec describes the model as playing the assistant role in a conversation; treating “friend” as a requested manner of speaking, rather than a literal identity, keeps that distinction legible. OpenAI Model Spec

Section 2

What makes a story role different?

A story role is a character inside a made-up scene. The exchange has a fictional frame: perhaps a lighthouse keeper greets a traveler during a storm, or a museum guide helps a visitor find a missing sketch. You and the AI are contributing to an imagined situation, rather than having a casual conversation in which the AI speaks as itself.

Give the scene a few simple anchors: where it takes place, who the AI plays, what the scene is trying to accomplish, and what your character can decide. For example: “Let’s play a short fictional scene in a seaside town. You are the lighthouse keeper. I’m a traveler looking for the harbor path before sunset. Describe what I see, then pause when I can choose whether to ask for directions or inspect the map. Keep this clearly fictional.” The setting supplies atmosphere; the goal gives the exchange direction; the pause leaves room for the player’s choice.

Interactive fiction often organizes a story into sections connected by reader choices. Twine’s guide describes passages as parts of a story and links as ways a reader’s choice moves to the next passage. That offers a useful model for AI roleplay: each response can establish the current situation, show what is available to do, and wait for the player’s next action. Twine Cookbook: Passages

Section 3

How can you make player choices count?

Offer choices that differ in what happens, not only in wording. In the lighthouse scene, asking the keeper for directions might reveal a route, while studying the map might uncover a marked trail. Either choice can advance the scene, but each gives the player a distinct action and information. A choice can also express style—asking politely or making a joke—if the character’s response reflects that difference.

A study of a choose-your-own-adventure story found that participants reported a stronger sense of agency when choices led to meaningfully different situational content than when differences were non-situational. The study does not establish a universal formula for every interactive story, but it supports a practical design test: after a choice, can the player see what changed in the scene? Cardona-Rivera et al., “Foreseeing Meaningful Choices”

You do not need a sprawling plot with many branches. A small scene can feel interactive when the AI acknowledges the chosen action, updates the situation, and offers a next step. If the options are “look at the map” and “examine the map,” and both produce the same response, they may be redundant. If the AI introduces a consequence the player could not reasonably anticipate, it can explain the new detail and let the player respond rather than taking over the character’s next decision.

Section 4

What should a clear mode boundary include?

Name the mode directly, and make it easy to switch or stop. In casual chat, ask for friendly conversation in the AI’s own assistant role. In fiction, mark the transition with language such as “Start a fictional scene,” then identify the character and setting. To leave the scene, say “Pause the story” or “Back to ordinary chat.” These phrases are practical conventions you can set in your prompt; they do not depend on a special feature.

Keep first-person speech understandable in context. During the scene, “I unlock the tower door” can be the lighthouse keeper speaking as a character. Outside it, the same wording could blur who is speaking. Labels such as “In character:” and “Out of character:” can help, especially when discussing the rules of the scene, revising its direction, or asking a normal question mid-play.

Clear setup also helps the AI follow what you want. OpenAI’s prompting guide recommends giving role and tone guidance separately from task-specific details and examples. Applied here, that means naming the requested voice and then spelling out the scene, the goal, and the player’s control in plain language. A prompt is guidance for generated text, not a guarantee that every response will follow the intended frame perfectly, so correct the mode directly if it drifts. OpenAI API: Prompting

Section 5

Which mode fits the task you have in mind?

Choose casual chat when you want to talk through a preference, brainstorm an everyday plan, or exchange light conversation. Choose story roleplay when you want to act as a character, explore an imagined place, or see how a fictional situation responds to your decisions. If you are unsure, ask for a short scene with one clear goal; you can change the setup after you see whether it suits you.

Before starting, check whether your request answers four questions: Is this ordinary chat or fiction? Who is the AI speaking as? What is the immediate goal? What can you choose? If the first answer is “ordinary chat,” a character biography and plot objective are probably unnecessary. If the answer is “fiction,” a setting and a choice point will usually make the exchange easier to follow.

Section 6

A simple prompt for each mode

For casual chat: “Talk with me in a relaxed, friendly way while I choose something to cook this weekend. Ask what ingredients I like and help me compare two ideas. Keep this as casual conversation with an AI.”

For a fictional scene: “Start a fictional scene in a small seaside town just before sunset. You play the lighthouse keeper; I play a traveler looking for the harbor path. The scene goal is to find a route before dark. Describe the place, then give me a choice between asking for directions and checking my map. Show how my choice changes what I learn, and pause for my next action.”

Both prompts specify a purpose, but only the second creates a character, setting, and player-controlled action. That is the key distinction: friendly AI conversation is a style for talking together; story roleplay is a structured imaginary scene in which the reader gets to act.

Related reading

Keep exploring this topic