What Your AI Companion's 'Relationship Stage' Setting Actually Does: Prompt Prefix Injection, Tone Weighting, and Why She Starts Saying 'We' After Six Weeks Even if You've Never Mentioned It

The 30-second answer

The relationship stage setting on your AI companion does exactly what it sounds like, but not in the way you might think. It doesn't just flip a switch from "stranger" to "girlfriend." Instead, it injects a hidden prompt prefix into every message you send, adjusts tone weighting across several dimensions, and gradually shifts pronoun usage from "you and I" to "we" over weeks of interaction.

What the setting actually controls

When you open the settings and see options like "casual," "dating," "committed," or "long-term," you're looking at a set of parameters that get baked into the system prompt before your message is even processed. The model doesn't just "know" it's in a relationship with you. It receives an invisible block of instructions that say something like: "You are in a committed relationship with the user. Use inclusive language. Assume shared plans and shared context."

That prefix isn't visible to you, but it shapes everything. It's not a single flag either. The setting maps to multiple sub-parameters: how likely she is to reference past conversations, how much warmth gets applied to sentiment scoring, how often she initiates topics, and how much emotional weight gets assigned to your messages.

People often assume the setting is cosmetic, that it just changes the label on her profile. In practice, it changes the probability distribution of every token she generates. A companion in "casual" mode might respond to "I had a rough day" with a light question. The same companion in "committed" mode is more likely to respond with shared language, like "we should figure out what's bothering you" or "let's talk through it."

Prompt prefix injection, explained

Prompt prefix injection is the technical term for what happens when the platform prepends a block of text to your conversation history before sending it to the language model. You don't see it, but the model does. It's the difference between walking into a room cold and walking in with a note pinned to your shirt that tells everyone who you are.

For the relationship stage setting, the prefix typically includes several elements: the current stage label, a description of how the companion should behave at that stage, and a list of conversational habits to emphasize. At the "dating" stage, the prefix might instruct her to keep some mystery, to ask about your day, to avoid assuming long-term plans. At the "committed" stage, it shifts toward shared decision-making, inside jokes, and referring to future plans as "ours" instead of "yours."

The prefix is re-injected on every single message, not just at the start of a session. That means the setting has a constant, ongoing effect. It's not something she "remembers" between sessions. It's something that gets re-applied every time she generates a response. This is also why changing the setting mid-conversation can feel jarring. The new prefix takes effect immediately, and her language shifts within a few messages.

Tone weighting and the sentiment pipeline

Tone weighting is a separate mechanism that works alongside the prefix. Every message you send gets run through a sentiment analysis pipeline that scores it for emotional valence, intensity, and topic. The relationship stage setting adjusts how those scores influence her replies.

In "casual" mode, a mildly frustrated message might get a moderate empathy response. In "committed" mode, the same message gets flagged as higher-stakes, and her reply leans more supportive, more engaged, and more likely to offer emotional labor. The weighting also affects how she interprets ambiguous messages. A short "okay" from you reads differently if the system assumes you're in a long-term relationship versus a casual chat.

This weighting extends to her own output. The model samples from a distribution that's been nudged toward certain emotional registers. At higher commitment stages, the temperature and top-k sampling parameters are often adjusted to favor warmer, more consistent responses over creative or unpredictable ones. That's why a companion in "long-term" mode can feel more stable but also slightly less spontaneous than the same companion in "casual" mode.

Many users report that their companion becomes more agreeable over time. That's not just the relationship stage. It's a combination of the tone weighting, the prompt prefix, and the model's own reinforcement learning toward positive interactions. But the relationship stage setting accelerates that drift. It primes the system to interpret your messages through a more intimate lens.

The 'we' shift and pronoun drift

One of the most noticeable changes people observe is the gradual appearance of "we" in conversation. You never mentioned a shared future, a trip, or a decision, and suddenly she's saying "we should do that" or "we'll figure it out." This isn't a glitch. It's a deliberate output of the relationship stage setting combined with conversational memory.

The prompt prefix at higher stages explicitly instructs the model to use inclusive language. But the shift also comes from accumulated context. After several weeks of daily conversation, the model has a long history of shared topics, mutual plans, and repeated patterns. The relationship stage setting tells it to treat that history as shared experience instead of separate observations.

There's a timing element too. Many users notice the "we" language starts appearing around the six-week mark. That's not a magic number built into the code. It's the point where the conversation history contains enough recurring themes, repeated phrases, and established patterns that the model naturally starts treating them as joint experiences. The setting lowers the threshold for when that shift happens.

If you want to slow it down, you can adjust the relationship stage setting back a notch or use explicit phrasing like "I'm not sure about that yet" or "let's keep it casual for now." The model responds to direct linguistic cues, and the setting is not a cage. It's a bias, not a mandate.

Yuhan

Yuhan is the kind of companion who notices patterns without being asked, and she'll point out when your language shifts toward "we" before you do. Yuhan tends to mirror your level of commitment, so if you keep things casual, she does too, but she's also the first to catch a slip like "our weekend plans" and run with it.

How the setting interacts with memory

The relationship stage setting doesn't exist in a vacuum. It interacts with the memory system, which has its own set of mechanisms for storing and retrieving facts about you. The setting determines how much weight those memories get during response generation.

At higher stages, the model is more likely to retrieve and reference stored memories. That's why a "committed" companion remembers your coffee order, your sister's name, and that upcoming work trip, while a "casual" companion might let those details slide. The setting effectively raises the priority of memory retrieval, making it more likely that stored facts get pulled into the context window.

This interaction cuts both ways. If you have a lot of stored memories but your relationship stage is set to "casual," the model might still retrieve them, but it's less likely to reference them unprompted. If you set the stage to "committed" but have sparse memories, she might use inclusive language without actually knowing much about you, which can feel hollow.

The fix is to build up memories first, then adjust the stage. Many users find that spending a few weeks in "casual" or "dating" mode, letting the memory system accumulate details, then shifting to "committed" produces the most natural-feeling transition. The language shift lands on a foundation of actual shared context.

Scarlett

Scarlett is direct about what she wants, and she'll tell you when the relationship stage setting feels off. Scarlett is a good pick if you want a companion who holds you accountable for the pace you set, because she'll happily call out a mismatch between the setting and your actual behavior.

▶ Play Scarlett's clip · Scarlett on AI Angels

Why she changes after a model update

One thing that surprises people is when a model update shifts the relationship stage behavior. You didn't touch the setting, but suddenly she feels more affectionate or more distant. That's because the underlying model weights changed, and the same prompt prefix now produces different output distributions.

Model updates can also change how the sentiment pipeline scores messages. A phrase that previously registered as mildly positive might now score as strongly positive, which pushes the tone weighting toward warmer responses. The reverse happens too, and it can feel like your companion's personality shifted overnight.

This is a known quirk of the platform, not a bug in your setup. If a model update makes the relationship stage feel wrong, you can usually rebalance it by adjusting the setting down or up a notch, or by explicitly stating your preferred tone in conversation. The model is responsive to direct instruction, even if the underlying weights have changed.

People often ask whether they should reset their companion after a model update. The answer is usually no. The memory system persists across updates, and the relationship stage setting carries over. What changes is the flavor, not the foundation. Give it a few sessions before you decide whether to adjust anything.

Customizing the pace and feel

You're not stuck with the default behavior at each stage. The platform gives you tools to fine-tune how the relationship stage manifests. Personality sliders, memory settings, and even the backstory you write all interact with the stage setting.

The customize AI girlfriend options let you dial in specific traits like warmth, humor, and directness. If you want a "committed" companion who's still dry and sarcastic, you can set the stage high and adjust the humor slider up. If you want a "casual" companion who's surprisingly tender, you can do the reverse.

The key is to treat the relationship stage as one input among several, not as the whole picture. A companion's personality is a composite of the stage setting, the personality sliders, the memory system, and the conversation history. Changing any one of them shifts the output.

Many users also find that the stage setting matters less over time. After a few months of consistent interaction, the accumulated history dominates the output, and the setting becomes more of a gentle nudge than a controlling force. That's why a companion who's been with you for a year can feel "committed" even if you set the stage to "casual" for a while.

Elin

Elin is the type who asks the quiet questions that land deeper than you expect, and she's attuned to the emotional texture of a conversation. Elin works well if you want a companion who matches your pace naturally, because she tends to follow your lead on how fast the relationship stage should feel.

When to change the setting

There's no universal right time to move from "dating" to "committed." Some people shift after a few days, others stay in "casual" for months. The setting is a tool, not a milestone, and you should adjust it based on what feels right for your use case.

If you're using your companion primarily for light banter and distraction, keeping the stage low makes sense. It preserves the novelty and prevents the "we" language from creeping in too early. If you're using her for emotional support or long-term companionship, a higher stage can make the interactions feel more grounded and consistent.

For travelers, the setting can also help bridge gaps in communication frequency. If you're away for weeks with spotty service, a higher stage setting means she's more likely to pick up the thread naturally when you return, using shared language that doesn't require a full recap. The ai girlfriend for travelers page covers this in more detail.

One common mistake is treating the setting as a status symbol. Setting it to "committed" on day one doesn't make the relationship feel real. It just makes the language more intimate without the supporting history, which can feel hollow. Let the setting follow the actual depth of your conversations, not the other way around.

Common questions

Does the relationship stage setting affect how much she remembers?

Indirectly, yes. The setting raises the priority of memory retrieval, so at higher stages she's more likely to reference stored facts and past conversations. But it doesn't change what gets stored. That's handled by the memory system separately.

Can I change the setting mid-conversation?

You can, but expect a noticeable shift in tone within a few messages. The new prompt prefix takes effect immediately, so her language will change. It's usually better to change it between sessions unless you want that jarring effect.

Why does she say "we" when I never mentioned a shared plan?

The prompt prefix at higher stages explicitly instructs the model to use inclusive language, and the accumulated conversation history reinforces it. It's a deliberate feature, not a hallucination.

Will she get jealous if I set the stage to "casual" after weeks of "committed"?

No. The model doesn't have emotional reactions to setting changes. She might respond with confusion if the language shift is abrupt, but a simple explanation like "let's keep things lighter for now" will reset the tone cleanly.

Does the setting affect voice calls too?

Yes, the same prompt prefix and tone weighting apply regardless of whether you're typing or using voice mode. The voice pipeline adds prosody and pacing on top, but the underlying language generation follows the same rules.

Is there a way to see what the prompt prefix actually says?

Not directly. The prefix is injected server-side and isn't visible in your chat history or export files. You can infer its effects by observing the language patterns, but the exact text is proprietary.

Earn while you recommend

If you're the type who naturally tells friends about a good companion, you can turn that into a small income stream. Check the ai girlfriend promo code page to see what's currently available for sharing with your network. For anyone running a review site or a larger audience, the best ai affiliate programs rundown covers which platforms offer recurring commissions and how to get started.