Published scorecard • 8 criteria • revised September 2026
How We Review
Every score on this site comes from the same eight criteria, applied to every service in the same order. We compare AI girlfriend builders using what anyone can check without paying: public pricing pages, feature pages, help-center articles, terms of service, privacy policies, app-store listings and the free previews each service shows before signup. We do not create accounts, buy plans or enter card details to produce a score, so we say "compared" and "reviewed", not "tested".
A criterion is scored from 1 to 10. The overall score is the weighted average, rounded to one decimal, and it is the same number on the home page, the ranking and the full review.
Builder depth — weight 20%
How much of the character the create flow lets you decide — face, body, hair, ethnicity, style, age range above 18, personality, relationship role, voice — and whether those choices carry into chat and images.
8 or above: a guided flow with ten or more distinct controls, a realistic and an anime style, and settings that can be edited after creation. 5 or below: a short form or a pick-from-gallery flow where most of the look is fixed by presets.
Image and video output — weight 15%
Whether the service generates pictures (and clips, where available) of the character on request, and how consistent the face and body stay from one picture to the next in its public samples.
8 or above: in-chat photo requests plus a separate generator, video or animated output, and public samples where the same character is recognizable across shots. 5 or below: images only through a separate tool, or samples where the face drifts between pictures.
Chat and memory — weight 15%
What the service publishes about memory, context length, roleplay scenarios and voice, and how its public demo conversations read.
8 or above: named long-term memory, editable backstory or scenario, voice messages, and demo chats that stay in character. 5 or below: no stated memory, generic replies in demos, text only.
Adult freedom and controls — weight 10%
Whether explicit chat and explicit images are allowed for adults, and whether the user can set how far things go.
8 or above: explicit chat and images permitted by the terms, with a clear on/off or intensity setting. 5 or below: explicit content blocked, filtered unpredictably, or allowed in chat but not in images.
Price clarity and value — weight 15%
Whether the full price list is visible before signup, how much a month of normal use costs, and how tokens or credits are spent.
8 or above: public plan prices, a stated token allowance, and a normal month of use that does not need top-ups. 5 or below: prices hidden until signup, or a base plan that runs out of credits quickly.
Free access — weight 10%
What a new user can do before paying: build a character, send messages, see a picture.
8 or above: a character can be built and chatted with for free, with at least a sample image. 5 or below: the builder or the first reply sits behind a paywall.
Privacy and billing discretion — weight 10%
What the privacy policy says about chat storage, training and deletion, and how the charge appears on a statement.
8 or above: a clear deletion route, no sale of chat data, and a neutral billing descriptor stated in public. 5 or below: vague policy wording, no stated deletion route, or nothing said about billing.
Access and platforms — weight 5%
Web, iOS, Android or installable web app, language support, and whether the service is reachable without a VPN.
8 or above: web plus an app or installable web app, several languages. 5 or below: one platform, English only, or regional blocks.
What we could not verify
Output quality inside a paid plan, reply speed under load, how long memory really holds, and whether a cancellation works on the first try all sit behind a login. Where a score depends on them, it rests on the published description and the public samples, and the review says so in plain words. When a service changes its plans, the dated status in each review is what we checked; the service's own page wins.
What does not move a score
Which service a button opens, how a brand styles its landing page and how often it advertises do not change a number. A service ranks first only when its weighted scorecard is the highest, and it still carries at least one stated limitation. If research ever stops supporting the current order, the ranking changes and the date changes with it.
Refresh cycle
We re-check prices, free-tier limits and the "checked" date on every review at least once a quarter, and sooner when a service announces a new plan. Readers new to the category can start with the guide library. Corrections go to contact@ai-girlfriend-generator.org; see the about page for what this publication is and is not, or go straight to the ranked comparison.
18+. Every character on this site is an AI-generated adult. No real people are depicted.