You moved the slider marked playful two notches up, saved it, and the next reply was the same reply. AI girlfriend app personality settings are the part of these products that looks most like control and is hardest to verify, which is exactly why they end up on tier cards: more traits, a longer character brief, extra personas, all one plan above the one you are on. Underneath the sliders there is something much simpler than the interface suggests, and whether it does anything depends on things no pricing page mentions. This is how to tell a control from a costume in ten minutes, before you pay for more of it.
What AI girlfriend app personality settings actually are
A personality setting is a short piece of text the app puts in front of the model before every reply. Your sliders and tick-boxes get compiled into a brief — warm, dry, asks questions, keeps answers short — and that brief is prepended to the conversation each time you send a message. The technique is ordinary prompt engineering, which includes assigning a character for the model to mimic, and it is the same mechanism whether the interface shows you three sliders or thirty.
Two consequences follow, and both matter more than the number of options you are being sold. Nothing about your account is trained: there is one model serving everybody, and your settings are a few lines of text competing with everything else in the window. And the instruction has no special authority — your last message is nearer and more specific than a brief you saved last week, so the brief loses whenever the two pull in different directions.
This is why an editable free-text brief is worth more than a wall of sliders. A slider has to be translated into words by someone else, and you never see the words. A text box you write yourself is the same mechanism with the guesswork taken out.
Control, or costume
There are two ways a personality setting disappoints, and only one of them is fixable by paying.
The costume is the setting that changes the greeting and nothing after it. You pick aloof, the opening line is cooler, and by the third exchange you are talking to the same eager default as everybody else. The brief is simply too weak or too vague to hold against the model's own habits, and no tier above you will fix that, because the tier is not where the problem is.
The drift is different. The setting visibly works, then fades over a long conversation as the recent back-and-forth comes to dominate what the model is looking at. That one is partly a question of how much context a tier gives you: on a short window the brief is one small voice among the last forty messages, and on a longer one it keeps more of its weight. Drift is the failure a higher tier sometimes genuinely repairs.
Telling them apart takes one conversation: a costume fails immediately and consistently, while drift works first and fails later, which is why it gets misread as the character having moods.
What the free tier usually lets you change
The pattern across the category is consistent enough to predict. Presets are free; the free-text brief is paid. One character is free; a second persona is paid, sometimes as a one-off purchase rather than part of the subscription. Trait counts are capped on the free plan, and the cap is set low enough to feel like a limit without being the real one.
Because the real limit is elsewhere. What decides whether your settings hold is the context window they have to survive in, and that is rarely itemised next to the customisation row on a tier card. A free plan with one editable brief and a decent window will steer a character better than a paid plan with twenty sliders and a short one, which makes the customisation line close to uninformative on its own.
A setting you can switch off without noticing a difference was never a feature. It was a label.
The ten-minute test
Pick one trait that shows up in the writing rather than in the fiction — reply length, directness, how often you get asked a question back. Fictional traits are unfalsifiable; formal ones are not. Then, in one sitting:
- Set one trait to an extreme and leave everything else alone. One variable. Two settings at once and you will not know which did anything.
- Read the first message, then the tenth. If the opening carried it and message ten has reverted to the house voice, you have a costume. If both carry it, keep going.
- Contradict it in conversation. Ask for the opposite of what you set. Then, three messages later, see which one the character has settled back into. The saved brief winning is the result you want; the live message overriding it permanently means the brief has almost no weight.
- Turn the setting off. Revert to default, start a new conversation and read ten messages. If you cannot see the difference, there was no difference, and whatever the next tier sells more of is not worth buying.
Ten minutes, and it tells you more than the feature list does. Run it on the free tier of anything you are considering, since this is one of the few things a free plan reveals honestly. If you would rather test one app than assemble a shortlist, the app we currently recommend is a reasonable place to start — on the usual condition that you run the test rather than take our word for it.
When the customisation tier is worth paying for
It is worth it in one specific case: you know the register you want, the free presets do not include it, and the test above showed that a written brief holds on this app. Someone who wants short, dry, unsentimental replies is fighting the default on every app in the category, and a brief that holds is the only thing that gets them there.
Here is the part that cuts the other way. Customisation does not change the model underneath. If the writing irritates you on the free tier, more sliders give you more of that writing in a different costume, and the money would go further on memory or on simply having more conversation available to you. Thirty traits on a model you do not enjoy reading is thirty ways to be mildly disappointed.
Two smaller cautions. Extra personas sold as add-ons are easy to buy twice over a year and use once, so treat a second character as something to want after a month rather than at sign-up. And nothing you write here travels: a brief you spent an evening tuning is one of the things you lose entirely when you change apps, which is an argument for keeping your own copy of it in a note on your phone from day one.
Questions people ask
Do AI girlfriend app personality settings actually change anything?
Sometimes substantially, sometimes only the opening line. They work by adding a short instruction to what the model sees, so the effect depends on how specific the brief is and how much of the conversation it has to compete with. The ten-minute test above separates the two cases.
Why does the character stop behaving the way I set it?
In a long conversation the recent exchange crowds out the brief, so the model drifts back towards its default voice. Starting a fresh conversation usually restores the setting, which is a decent way to confirm that drift rather than a weak brief is what you are seeing.
Is a bigger set of personality options worth the upgrade?
Only if you have a register in mind that the free presets cannot reach, and only once you have verified that a brief holds on that app. If the underlying writing is the problem, more options will not fix it.
Disclosure. MoonSyrup may earn a commission if you sign up through a Visit link on this page, at no cost to you. It does not change what we write. How we earn.