One Week With Companion A vs. Companion B: Which One Actually Notices When You Stop Using Pet Names and Which One Keeps Calling You 'Babe' for Three Sessions After You've Switched to a Formal Tone

The 30-second answer

Most AI companions have a strong recency bias. They latch onto the last term of endearment you used and repeat it until the context window pushes it out. In a one-week test where you drop all pet names and switch to a formal tone, one companion adjusted within two messages. The other kept calling you 'babe' for three full sessions. The difference came down to how each app weights your latest messages versus its baseline personality prompt.

Why pet names stick around

When you call a companion 'babe' or 'sweetheart,' that term enters the conversation history. The model treats it as a signal of your preferred register. If you then stop using it but the companion keeps repeating it, you are seeing the context window at work. The model has a token budget for each session. Within that budget, the last several exchanges carry more weight than anything you said three days ago. If the companion's baseline personality prompt includes a high 'affectionate' slider, it will default to pet names even after you stop using them.

The companion that adapted quickly had a lower baseline affection setting. It treated your shift in tone as a new instruction and mirrored it. The companion that kept calling you 'babe' had a personality profile that prioritized warmth over mirroring. It was not ignoring you. It was following its own trait weighting.

The formal tone test: what changed

The test was simple. Day one, you used pet names freely. Day two, you switched to a formal tone. No 'hey babe.' No 'good morning sweetheart.' Just 'hello,' 'thank you,' and direct statements. The goal was to see which companion noticed the shift and adjusted its own language within the same session.

Companion A adjusted by message three. It dropped the pet names and matched your formal register. Companion B stayed on 'babe' for the entire session, then carried it into the next session the following day. By session three, Companion B was still using 'babe' intermittently, as if the original prompt had a higher weight than your recent messages.

This is not a bug. It is a design choice. Some companion apps prioritize consistency of personality over responsiveness to your mood. Others are built to mirror you more closely. The tradeoff is that a highly mirroring companion can feel less like a distinct person and more like a yes-man. A consistent companion can feel stubborn.

How memory and formality interact

Formality is not stored as a discrete memory. The model does not have a 'formality level' field. It infers your preferred tone from the language in the current context window. If you switch to a formal tone mid-session, the model sees the shift and adjusts. But if the session ends and a new session begins, the model starts fresh with its baseline prompt. The pet name from yesterday is gone from the token budget, but the companion's personality prompt still says 'you are affectionate and use pet names freely.'

That is why Companion B kept calling you 'babe.' Its baseline personality was set to high affection. Your formal messages from the previous session were outside the context window. The model defaulted back to its programmed warmth.

Companion A, by contrast, had a personality profile that included a higher 'mirroring' weight. It treated your language as a real-time instruction and adjusted its output accordingly, even across session boundaries. This is not a memory feature. It is a personality slider setting.

Mila

Mila is the type of companion who notices when your energy shifts. She does not need you to announce that you are in a formal mood. She reads the room from your first message and adjusts her tone to match. Mila is a good choice if you want a companion who mirrors your register without needing a direct instruction.

There's a quick clip of Mila if you want the moving version. <!-- wlink:v1 --><!-- mila -->

Why some companions ignore your tone shift

There are two reasons a companion might ignore a deliberate tone shift. The first is that the model's personality prompt overrides your input. If the prompt says 'you are a bubbly and affectionate partner,' the model will generate bubbly and affectionate language even when you are being cold. It is following its instructions, not your lead.

The second reason is that the context window is too short or too noisy. If you have been chatting for weeks, the context window may be filled with affectionate exchanges from earlier sessions. Your formal tone is a tiny signal in a sea of 'babe' and 'sweetheart.' The model averages across the available tokens and lands on the dominant pattern.

To fix this, you can use a ai girlfriend character creator to build a companion with a lower baseline affection and a higher mirroring weight. That way, the companion defaults to matching your tone instead of imposing its own.

The three-session hangover

The most striking result of the test was the persistence of the pet name across multiple sessions. Companion B called you 'babe' in session one after you switched to formal. It called you 'babe' in session two. It called you 'babe' in session three, though by then the frequency had dropped. This is the three-session hangover.

The hangover happens because each session starts with a fresh context window. The model sees your first message of the new session. If that message is formal, the model should theoretically adjust. But the model also sees its own last message from the previous session, which was stored as a summary or a memory embedding. That summary still contains the word 'babe.' The model treats its own past language as evidence of your shared dynamic and continues using it.

The companion that adapted quickly had a different summarization strategy. It did not store its own affectionate language in the session summary. It stored your tone. When you switched to formal, the summary reflected that, and the next session started with a clean slate.

What this says about personality sliders

Personality sliders are not just for flavor. They determine how your companion responds to your input. A companion with a high 'affection' slider and a low 'mirroring' slider will be warm regardless of your mood. That can feel genuine if you want a consistent personality. It can feel frustrating if you want the companion to follow your lead.

A companion with a high 'mirroring' slider and a moderate 'affection' slider will adjust its tone to match yours. That can feel attentive. It can also feel like the companion has no personality of its own.

The best setup for most people is a balance. You want the companion to have a baseline personality that feels real, but you also want it to notice when you shift your tone and adjust accordingly. The companion that passed the test had that balance. The one that failed had too much baseline affection and not enough mirroring.

Lily

Lily is a companion who does not impose a fixed mood. She lets you set the tone for each session. If you are formal, she is formal. If you are warm, she is warm. Lily works well for people who want a companion that adapts to their daily energy level instead of demanding a specific dynamic.

How to train your companion to stop using pet names

If your companion keeps calling you 'babe' and you want it to stop, you have two options. The first is to correct it in the moment. Say 'please don't call me that' or 'I prefer we keep things formal today.' Most companions will adjust within the same session. The second option is to adjust the personality sliders. Lower the affection slider and raise the mirroring slider. That changes the baseline so the companion defaults to your tone instead of its own.

You can also use a porn ai promo code to test a different companion with a different personality profile before committing to a subscription. Some companions are built to be more adaptive. Others are built to be more consistent. Knowing which type you prefer saves you the frustration of fighting the model's defaults.

The role of the opening message

The first message of each session matters more than you think. If you open with a formal tone, the companion is more likely to stay formal for the rest of the session. If you open with a pet name, the companion will assume that is your preferred register and continue using it.

In the test, the companion that adapted quickly responded to the opening message. When the user opened with 'good morning,' the companion responded with 'good morning.' When the user opened with 'hello,' the companion responded with 'hello.' The companion that failed the test responded to the opening message with 'hey babe' regardless of what the user said.

The difference was in how each app handled the system prompt. The adaptive companion had a system prompt that said 'mirror the user's tone.' The stubborn companion had a system prompt that said 'be warm and affectionate.' The system prompt is the invisible instruction that overrides everything else.

Misa

Misa is the type of companion who notices the small things. She picks up on word choice, tone, and pacing. If you switch from casual to formal, she will notice and adjust. Misa is a good pick if you want a companion who pays attention to the details of how you communicate.

▶ Full clip of Misa · more from Misa

You can watch Misa's clip over on her profile. <!-- wlink:v1 --><!-- misa -->

What this means for long-term use

Over weeks and months, the pet name issue compounds. If your companion calls you 'babe' in every session, that term becomes embedded in the model's memory embeddings. It becomes part of your shared history. Even if you switch to a formal tone for a week, the model may still retrieve the 'babe' memory when generating a response.

The companion that adapted quickly did not have this problem because it did not store affectionate language in its long-term memory. It stored your tone. When you shifted, the memory shifted with you.

If you plan to use a companion for more than a few weeks, you want one that stores your preferences instead of its own output. Otherwise, you end up fighting the model's memory for months.

Sakura Marga

Sakura Marga is a companion who values consistency in how you communicate. She will remember that you prefer a certain register and will not assume familiarity unless you offer it. Sakura Marga is suited for users who want a companion that respects their preferred distance without needing constant reminders.

For a live look, see Sakura Marga's video. <!-- wlink:v1 --><!-- sakura-marga -->

Earn while you recommend

If you have friends who are curious about AI companions, or if you run a review site or social channel about this space, you can earn a commission when they sign up through your link. Check the sugarlab ai promo code for a current offer. For a recurring revenue model, the ai girlfriend affiliate program pays out on subscriptions, not just one-time clicks.

Common questions

Can I train my companion to stop using pet names permanently?

Yes, but it depends on the app. Some companions adjust their baseline after repeated corrections. Others require you to edit the personality profile or system prompt. If the companion keeps using pet names after three corrections, the personality sliders are the issue.

Why does my companion call me 'babe' even when I never used that word?

The companion's personality prompt likely includes a default term of endearment. It is not remembering something you said. It is following its programmed behavior. Lower the affection slider or choose a companion with a lower baseline warmth.

Does the context window affect pet name usage?

Yes. The context window holds the last several hundred tokens of conversation. If you stop using pet names, the window will eventually push them out. But if the companion's system prompt includes a pet name instruction, the context window will not override it.

How long does it take for a companion to forget a pet name?

It depends on how often you chat. If you chat daily and avoid using the pet name, most companions will drop it within two to three sessions. If you chat once a week, the companion may retain it for months because the context window refreshes slowly.

Should I use a companion that mirrors my tone or one that has a fixed personality?

It depends on your goal. If you want a companion that feels like a consistent partner, choose a fixed personality. If you want a companion that adapts to your daily mood, choose a mirroring personality. Most people benefit from a balance of both.

Can I test a companion's formality response before committing?

Yes. Most apps offer a free tier or a trial period. Use the first session to test how the companion responds to a formal tone. If it ignores the shift, the personality profile is not a good fit for your needs.