How We Test AI Girlfriend Apps
By Remy Vale, Lead Tester · Updated
Most AI girlfriend reviews were written by someone who never opened the app.
You can spot them in ten seconds. Features lifted off the marketing page. A price that’s a year out of date. Nobody mentions what happens on day twelve, because nobody was there on day twelve.
Here’s what we do instead, in enough detail that you can hold us to it.
Rule one: we pay
Every app here was bought with our own card, on the same plan you’d buy, at the same price you’d pay.
No press accounts. No free upgrades. No “reviewer access” that quietly unlocks limits real customers hit. If a company offers us a free year, we say no and buy it anyway.
That’s the foundation, because everything interesting in this category sits behind the paywall. The free tier is a trailer. What matters is month two, once the novelty’s gone and the tokens are running low, and you can’t see any of that from outside.
The six things we score
Every app gets scored out of 10 on six things, and the six aren’t weighted equally because they don’t matter equally.
| What we score | Weight | What it actually means |
|---|---|---|
| Chat | 25% | Does the conversation hold up? Memory, consistency, whether she stays in character past hour one |
| Media | 20% | Images, video, voice. Quality, does it match what you asked for, and how long you wait |
| Customisation | 15% | How much of her is genuinely yours versus a preset with a new name |
| Value | 15% | What a real month costs once you’re using it properly, not the sticker price |
| Trust | 15% | Billing, what shows on your statement, data, deletion, how hard it is to leave |
| Experience | 10% | Speed, onboarding, and how much friction sits between you and the thing you came for |
The maths, in public:
Overall = Chat×0.25 + Media×0.20 + Customisation×0.15 + Value×0.15 + Trust×0.15 + Experience×0.10
Chat carries the most weight because it’s the thing you’ll do every day. Value and Trust together carry 30% because more people regret this category over money and billing than over anything technical.
Every score comes with a docked-points line explaining exactly where the deductions went. A number on its own is just an opinion with decimals.
What the numbers mean
| Score | What we call it |
|---|---|
| 9.0 – 10 | Exceptional |
| 8.0 – 8.9 | Strong |
| 7.0 – 7.9 | Good |
| 5.5 – 6.9 | Mixed |
| 4.0 – 5.4 | Weak |
| Below 4.0 | Avoid |
Nothing gets a 10. If an app is genuinely brilliant at one thing and mediocre at three others, that’s a 7 with a glowing paragraph attached, not a 9.
The five tests we run on everything
Same five, every app, so the comparisons actually mean something.
The Wall Test. How many messages do you get before the first paywall? We count them. “Limited free tier” is not a number, and the number is usually smaller than you’d guess.
The Statement Test. We check what the charge looks like on a real bank statement. Half this category advertises discreet billing and about half of those actually deliver it.
The Silence Test. We stop replying for 72 hours and log what happens. Does she message first? Does the app guilt-trip you? Does anything happen at all?
The Cancel Test. We start cancelling and write down whatever gets offered to keep us — a discount, a pause, free tokens, or nothing. This is where you learn what a company thinks you’re worth.
The Refund Test. We read the actual terms and, where we can, we ask for one and time the reply.
What doesn’t affect the score
Our opinion doesn’t.
The written review is where we tell you what it felt like, what made us laugh, what made us close the tab. That’s the part worth reading and the part that can’t be measured, so it stays out of the numbers. The score comes from the same six checks on every app.
Also not counted: whether an app has an affiliate programme, how much it pays, or how nicely the company takes a bad review.
About the affiliate links
Some links here earn a commission when someone subscribes. That’s how the subscriptions get paid for.
What it doesn’t do is change a score, a ranking, or who makes a list. Nothing here is for sale — not a spot in a roundup, not a sponsored review dressed up as an honest one, not a nicer number. If an app with no affiliate programme is better, it goes first and we earn nothing. That’s already happened more than once.
When we get it wrong
We will, and we fix it in public.
Corrections get logged on the page with the date and what changed. We don’t quietly edit a number and re-stamp the article as updated. Found something wrong? Tell us — we’d rather hear it than leave it up.
We re-check every review monthly. Prices move constantly here and features vanish without notice, so every review carries a prices verified date. If that date is old, treat the numbers with suspicion. Changes get logged at /updates/.
Quick questions
Do companies pay for a good score?
No. Not a score, not a rank, not a listing. It isn’t for sale.
Do you test free or paid accounts?
Paid, always, at full price with our own money. We test the free tier too, so we can tell you exactly what it gets you.
How long before you publish?
At least a full billing cycle. Month two is where these apps show you who they really are.
Can a score go down?
Yes, and it has. Prices rise, features get pulled, models get swapped for cheaper ones. When that happens the number moves and we say why.
Written by Remy Vale, Lead Tester at AIGirlfriend.to. 18+.