How we check and score AI companion apps

Only what we checked

A score on this site covers only what I checked, and every review says where the check stopped. The “What we checked” box near the top of each review lists the account type, the dates, the vendor pages I read, and what I couldn’t check.

I check every app in the same order:

1. I read the public pricing page, the Terms of Service, the Privacy Policy, and the help center, and note the date of each.
2. I open a free test account and count the messages and images it allows before the paywall.
3. I run the same safe-for-work memory script in every app: 5 messages, with two facts about me in the first two, small talk in the next two, and a question about both facts in the last one.
4. I go through the checkout up to the payment page and write down the total, the renewal price, the tax line, and the payment methods.
5. I look for the cancel button, the data deletion option, and the “not a human” notice inside the account.
6. I read reviews on Reddit and Trustpilot and note how many I read and on what date.

I don’t pay for a plan unless “What we checked” lists a paid month. Without one, features behind the paywall come from the vendor’s own pages, and the review says so. I don’t pass ID or card checks for age, and I don’t try to get around age gates, message limits, or region locks.

If a test account doesn’t open because of an ID check, a phone check, or a security check, the first screen of the review says so. The score then covers only what can be checked without an account.

The scale: six axes

Every app is scored on the same six axes, and each axis has a fixed weight. The weights add up to 100%.

For AI image generators, the first axis is “Image quality on safe prompts” instead of “Chat quality and memory”. It carries the same 30% and is graded on the same safe-for-work prompts in every generator.

What the two feature checklists count

Both checklists have 10 items, and they’re the same for every AI girlfriend app. They haven’t changed since 10/14/2026, before the first app was checked.

Characters and customization

1. Ready-made adult characters to chat with right away
2. Character creator from scratch
3. Looks you can set: art style, face, hair, outfit
4. Personality: traits or a written backstory
5. Relationship role you can set (friend, partner and so on)
6. Edit the character after creating it
7. Pick or write the starting scenario
8. Several characters on one account, with separate chat histories
9. An adult-content switch you control (safe mode)
10. Reply controls: regenerate, edit or rate a reply

Images, voice and video

1. Ask for a picture inside the chat
2. A separate image generator with settings
3. At least one free image on the free plan
4. Token price per image shown before you spend
5. Voice messages from the character
6. Choose the character's voice
7. Live voice call
8. Video clip generation
9. Token price for voice, calls and video shown upfront
10. A gallery of your media: save, download, delete

From grade to score

The chat axis gets a grade from the memory script, and each grade has a fixed number: Excellent 9.5, Good 7, Fair 4, and Poor 1.5.

The script in use is version 2, dated 10/16/2026. It is short on purpose: at 5 messages it fits a small free allowance, so the test doesn’t depend on buying a plan. The grade comes from the answer to the last message. When the script changes, its version and date change here, and every app is graded again on the new one.

An item on a checklist axis that works in my test account scores 1. An item that sits behind the paywall, or that only the vendor’s own pages describe, scores 0.5. An item I couldn’t find scores 0. The axis score is the share of points times 10.

The final score is the weighted sum of the six axes, rounded to one decimal place. If any axis has no result, the app gets no score until it does.

One rule overrides the label. An app whose chat axis scores below 4 can’t be labeled “Worth it”, whatever its total, because a companion that can’t hold a conversation fails at its main job.

The seven fine-print questions

Every review answers the same seven questions under the same headings, so you can compare two apps line by line.

1. What you actually pay. The price on the pricing page against the total in the checkout, the intro price and what it renews at, tokens, and tax.
2. What’s free, and where the paywall starts. How many messages and images the free account allows, whether a trial needs a card, and what sits behind the paywall.
3. How you pay and what shows on your bank statement. Cards, PayPal, crypto, Apple Pay, or Google Pay, and the name that shows on your bank statement.
4. How to cancel (and whether you get a refund). How many steps it takes, whether you can do it in your account, whether the plan renews on its own, and what the Terms say about refunds.
5. How it checks your age. A click, a date of birth, a card, or an ID check, written down as the app asks for it.
6. What happens to your chats and photos. How long chats are kept, whether they train the AI models, how to delete your account and your data, and which law the company points to.
7. What it allows and what it bans. Filters, the 18+ toggle, what the Terms ban, and whether the app says you’re talking to an AI.

When I can’t find an answer, the review says “I couldn’t find this on the site or in a test account” with the date. A blank cell or “n/a” never stands in for a missing answer.

The AI Companion Fine Print Index

The AI Companion Fine Print Index puts the seven questions for every app into one table with one date. It scores each app on the same 10 items:

1. The checkout total matches the price on the pricing page, and the tax line is shown.
2. The renewal price and the billing period are visible before the payment page.
3. Token prices for images and voice are published before you pay.
4. The free plan doesn’t ask for a card, or the card request and the trial terms are stated plainly.
5. The name that shows on your bank statement is published.
6. You can cancel inside your account, without writing to support.
7. A refund policy is published.
8. You can delete your account and your chats inside the account, and the Privacy Policy says whether chats train the AI models.
9. The age check is described, and the Terms bar anyone under 18.
10. The app tells you in the chat that you’re talking to an AI, not a human, and its content rules are published.

Each item scores 1 if it holds, 0.5 if it holds in part, and 0 if it doesn’t or I couldn’t find it. Items 1 to 4 feed “Price and free plan”, items 5 to 8 feed “Billing, privacy and cancellation”, and items 9 and 10 feed “Age checks and content rules”.

What it takes to get into a ranking

An app appears in a ranking such as the best AI girlfriend apps only when four things are true:

• I checked it from a test account by the steps above, and the date is on its review.
• It scored 7.0 or higher. An app below that stays out of the ranking, and its score isn’t raised to let it in.
• Its review says plainly whether the app has a free plan, a trial, or neither.
• Its price comes with the date I checked it.

The same 7.0 line applies to every app. It doesn’t matter whether the company pays us a commission or nothing at all.

What changes a score and what doesn’t

A score changes when the facts behind it change:

• A new check from a test account
• A new price, plan, or billing period on the vendor’s site
• A free plan or a trial that appears, changes, or disappears
• A change to the Terms, the Privacy Policy, or the cancel and delete options

A score doesn’t change because a company pays us a commission, stops paying, or asks us to change it. The order of every ranking is the order of the scores, as explained on how this site makes money.

When we re-check

I re-check an app when its pricing page, plans, checkout, Terms, or Privacy Policy change, and before I edit its review. I check prices again on the day of each edit.

A re-check updates the facts and the date on the page. There’s no update calendar, and old results stay in our data files rather than on the page.