How We Test AI Companion Apps
By Alex Mercer, AI & tech writer18+ · UPDATED 06/26The short version
We treat each app like a product a real adult is about to trust with private chats and a credit card. The process has four stages: account setup with a disposable email, hands-on use across the free and paid tiers, structured scoring against a fixed rubric, and a privacy read where we go through the actual privacy policy and search for any reported breaches. Nothing is scored from a marketing page. If we cannot verify a claim by using the app, we say so. We re-test apps when they ship major changes, and we date every review so you can see how fresh it is. Our full app shortlists live on best AI companion apps and best AI girlfriend apps.
How Alex sets up and uses each app
The first rule is that we never test with a personal account. Alex creates a fresh, app-specific email address and a unique password, never reuses real names, photos, or payment details that tie back to an identity, and where a card is required, uses a privacy-protected virtual card. This is the same advice we give readers in the privacy guide: a throwaway email is the single easiest protection you have.
From there, Alex actually lives with the app. That means free-tier first, to see what is genuinely usable without paying, then the lowest paid plan that unlocks the core experience. We hold real conversations over several days, test how the companion remembers earlier details, push the personalization controls (personality, appearance, backstory, voice), and try the image and voice features the app advertises. We also test the boundaries of the content policy in general terms, to confirm what the app says it allows actually matches what it does. Mature features are noted only in clinical, general terms; we do not publish explicit content.
The seven things we score
Each app gets a 0 to 10 mark on seven criteria. The criteria are weighted, because for an app holding intimate data, privacy matters more than whether the avatar art is pretty. Here is the rubric and what we are actually looking for.
| Criterion | Weight | What we check |
|---|---|---|
| Chat realism & memory | 25% | Does it stay in character, hold a coherent thread, and recall facts from earlier sessions, or reset and contradict itself |
| Personalization | 15% | Depth of control over personality, appearance, backstory, and voice, and whether changes actually stick |
| Image & voice quality | 15% | Consistency of the character across images, generation speed and limits, and whether voice sounds natural |
| Content policy clarity | 10% | Whether the rules are written plainly, applied consistently, and not changed without notice |
| Pricing transparency | 15% | How clear the real cost is, including tokens, renewals, and what 'lifetime' actually covers |
| Privacy & data practices | 20% | What the privacy policy says about storage, sharing, training, and deletion, plus any known breaches |
The weighted average becomes the overall score. We do not grade on a curve, and a strong overall score never hides a serious privacy problem; in those cases we flag it in the verdict regardless of the number. You can see the criteria applied side by side in head-to-head pieces like Candy AI vs Muah AI and Character.AI vs Candy AI.
How we read the privacy and pricing fine print
Privacy gets the deepest dig because the downside is real. AI companion apps store your conversations, and some explicitly reserve the right to use chat data to train models or share it with third parties. Data breaches in this category have happened, and a breach here can expose deeply personal messages. So for every app, Alex reads the privacy policy and notes, in plain terms, what is stored, whether chats are used for training, who data is shared with, whether the company is based somewhere with meaningful data protection, and how account and data deletion actually work. We also search for any publicly reported breaches or security incidents and put them in the review. If a policy is vague or contradictory, that lowers the privacy score, because vagueness is itself a risk.
Pricing is the other place companies bury the truth. Token systems make spending hard to predict, free tiers are deliberately limited, and 'lifetime' deals often cover one feature set rather than everything forever. We document the actual cost of normal use, how tokens or credits are consumed, what auto-renews, and where the upsells sit. Our running summary of no-cost options is on free AI girlfriend apps, and the mechanics behind the category are explained in how AI companion apps work.
Editorial independence and our 18+ and wellbeing stance
We make money through affiliate links. When you sign up to a paid app through one of our links, we may earn a commission at no extra cost to you, and it never changes our rankings. The scoring rubric is fixed before any commercial relationship is considered, the privacy and pricing findings are written from hands-on testing, and a poorly performing app stays poorly rated whether or not it pays a commission. Paid apps we have tested are linked as /go/candy, /go/muah, /go/dreamgf, /go/soulgen, /go/nomi, and /go/kupid.
This site is for adults only, 18 and over. We will not endorse anything involving minors or non-consensual deepfakes of real people; reputable apps prohibit both, and both are illegal. We are also honest about the emotional side. These apps can become a time and money sink, and they are not a substitute for human relationships. We say that plainly in our reviews and recommend moderation, and we cover the tradeoffs in depth in AI girlfriend vs a real relationship and the safety overview at are AI girlfriend apps safe.
Limits of our testing
We want you to trust the scores, so we are clear about what they cannot tell you. We test as a single reviewer over a defined window, so we may not surface every edge case or a problem that only appears after months of use. Apps update constantly; a feature, price, or policy can change after we publish, which is why we date reviews and re-test on major releases. Privacy findings rely on what a company discloses and on public reporting, so an undisclosed practice can sit outside our view. And content moderation can vary by account and region. When something is outside what we could verify, we say so rather than guess. If you spot something we missed, tell us, and we will check it.
Candy AI is our top overall pick: the most polished experience. Start on the free tier, and set a spending limit before you buy tokens.
Affiliate link, 18+. We may earn a commission at no cost to you; it never changes our rankings (see how we test).
Frequently asked questions
How long do you spend testing each app?
At least a week of regular use per app, across both the free tier and at least one paid plan. Longer for apps with deep personalization or large feature sets. We re-test when an app ships a major update so the review stays current, and every review carries a date so you can see how fresh it is.
Do affiliate commissions change your rankings?
No. The scoring rubric and weights are set independently of any commercial relationship, and the findings come from hands-on testing. If you sign up through one of our links we may earn a commission at no extra cost to you, but a weak app stays weak in our ratings regardless of whether it pays out.
Why does privacy get more weight than image quality?
Because the worst-case outcome is different. A mediocre avatar is a minor annoyance. Intimate conversations being stored, used for training, shared, or exposed in a breach is a real harm. Privacy is 20% of the score and can be flagged in the verdict on its own when a policy or incident is serious.
Do you actually pay for the apps?
Yes. We use the free tier first, then pay for the lowest meaningful paid plan so we can judge what most users get for their money. We also document how tokens, renewals, and 'lifetime' deals really work, since that is where costs add up. See our notes on free options at /free-ai-girlfriend-apps/.
How do you protect your own privacy while testing?
We use a throwaway email, a unique password, no identifying personal details, and a privacy-protected virtual card where payment is required. We recommend readers do the same. Full guidance is in our /ai-companion-privacy-guide/.
Are these apps a replacement for a relationship or therapy?
No. They can be entertaining and, for some people, comforting, but they are not a substitute for human relationships or mental health care, and they can become a time and money sink. We flag this in our reviews and suggest moderation. We go deeper in /ai-girlfriend-vs-real-relationship/.
Signs up for every app, tests the free and paid tiers, and reads the privacy policies so you do not have to. He is upfront about the emotional and money traps. How we test →