You have two apps shortlisted, a week set aside, and no intention of paying for either until one of them earns it. Then the first opens on a screen with eleven sliders and the second opens on an empty text box, and you find yourself inventing a character twice, badly, at eleven at night. AI girlfriend app setup is the step everyone treats as a formality. It is also the step that decides whether the comparison you are about to run means anything, because two apps set up differently are not being compared. They are sitting different exams.
Why AI girlfriend app setup skews the result
Every app collects roughly the same information in a different shape. Some hand you appearance and personality sliders plus a relationship setting and nothing else. Some give you a blank description field and expect you to write the personality, the speaking style and the boundaries yourself. Some add a greeting editor and a scenario field, so you also choose the opening line and the situation the first conversation starts in.
Whatever route the screen takes, the output lands in the same place: text put in front of the model before your first message, which is what the model is conditioned on for everything that follows. That is ordinary prompting wearing a friendlier interface. Which means the quality of what you typed is a live variable in your test — and on the second app you are always better at it than you were on the first, because you have just spent half an hour learning what these fields want.
The result is predictable and almost nobody notices it. The app you set up second scores higher. You conclude it is the better app, cancel the other one, and pay a subscription to your own second draft.
Write the brief once, before you open either app
The fix is a single page of notes written before any sign-up screen is in front of you. Not a document, not a persona essay — about 150 words, so it fits the smallest text box you will meet. Six things:
- A name and one line on who the character is. One sentence. If you cannot write it in one sentence, you have not decided yet.
- Three or four traits, in your own words. "Dry, a bit blunt, reads a lot" beats any slider position, and it transfers between apps unchanged.
- How you want it to talk. Message length, how often it asks questions, whether it uses pet names. This is the setting people most often wish they had changed on day one.
- Two subjects you actually want to talk about. Vague briefs produce vague conversation in every app equally.
- Two or three boundaries. Written plainly: do not steer things toward explicit content, do not guilt me for going quiet, do not claim to have feelings.
- Three specific, unusual facts you are willing to have stored. These are your memory test later in the week, and they have to be identical in both apps or the test proves nothing.
Write it in a notes app, not in the first sign-up screen you open. The point is that app number two gets the same brief as app number one, rather than a sharper version produced by practice.
Translating one brief into whatever screen you get
The brief survives all three kinds of setup screen, provided you decide in advance how to map it.
Sliders only. Pick the nearest position for each trait and put whatever will not fit into the "about" or "notes" field if there is one. If there is no free-text field anywhere, do not fight it — write that down. An app that will not accept a sentence about the character belongs in your table as a finding rather than as frustration.
A description box. Paste the brief nearly verbatim and resist improving it. This is the app that will tempt you into a better draft, and a better draft here is exactly the contamination you are trying to avoid.
Greeting and scenario editors. Keep the greeting identical across apps, and keep it plain. A well-written opener flatters any model, which is useful if you want a good evening and useless if you want a comparison. One neutral line, same words, both apps.
Skip appearance and voice on the first pass. Neither is what you are judging in week one, and image or voice generation is usually the shortest path to a paywall. Our guide to comparing apps in one week covers where the rest of the week goes.
The app you set up second is not better. You are just better at setting up apps than you were an hour ago.
What to keep out of the brief
The persona fields are not scratch paper. What you type there is stored on a server and, in most designs, sent along with every message you ever send, which makes it the single most reused piece of personal data in the product. The data minimisation principle is the regulator's version of a rule you can apply to yourself here: hand over what the purpose needs, and nothing on the off-chance it is useful.
So keep your employer, your address, your full name and other people's names out of it. Your memory-test facts can be specific and unusual without being identifying — an invented label tests recall as well as a real name does, and recall is all you are testing. A useful filter: would you mind this text appearing in a support ticket a stranger reads while fixing your account. Our privacy checklist covers the clauses worth finding before you type any of it.
Where the brief works against you
Here is the honest cost, and it is not small. An identical brief flattens the thing some apps are genuinely best at: their own default characters, written by people who know the model and have tested a hundred openings. Judge every app only through your own persona and you may quietly penalise the one that does that best.
The cheap fix is two sessions per app rather than one: one on the app's default character, untouched, and one on your brief. It costs an evening and it separates two questions worth keeping apart. How good is this app when it is doing the writing, and how good is it when I am. The first tells you what the app can do at its best; the second is what matters if you plan to keep a character for months.
Two more caveats. The brief is not a contract — at the end of the week, rewrite it in both apps with what you learned and see what changes. And treat setup itself as a column: whether you can edit the description after creation, whether persona editing sits behind a tier, whether the app holds the brief or drifts away from it by Thursday. Those are product differences, and they show in the first hour if you watch for them. Our piece on the columns that matter has the rest of the table.
If you would rather run one brief past one app before building a table at all, start with the app we currently recommend and keep the notes for whichever app you try second.
Frequently asked questions
How long should AI girlfriend app setup take?
Ten minutes of writing before you open anything, then two or three minutes per app to transfer it. The long version, invented live on the sign-up screen, is not more thorough — only less repeatable.
Should I use the same character in every app I try?
For the comparison, yes: the same name, traits, speaking style and test facts, so the differences you notice belong to the app rather than to your typing. Run one extra session on each app's own default character as well, since that is where some apps do their best work.
Can I change the personality settings later?
In most apps you can edit the description or move the sliders at any time, though some lock parts of the character after creation or put persona editing behind a paid tier. Check which kind you are dealing with before the week starts, because it decides whether a second run is possible at all.
Disclosure. DearHeart AI 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.