Chat quality
How natural, coherent and in-character the conversation feels: memory across a session, ability to hold a storyline, and how often the model breaks or repeats itself.
Editorial standards
How we rate AI companion apps
Every tool in our directory carries a 0–5 editorial score. This page explains the five dimensions we judge, the evidence limits behind each review, and—just as important—what our scores do not claim.
The five criteria
What each score is built from
Each tool gets a sub-score for five dimensions. The headline rating is an editorial summary of the scorecard; it is not a user-star average, and affiliate placement does not determine the result.
Media features
Image generation, video-style media, voice calls and selfies—both whether they exist and how convincing and consistent the output is.
Customization
How much control you have over a companion's appearance, personality, backstory and behaviour, from preset characters to full build-your-own creation.
Privacy posture
How clearly the product explains data collection, account controls, content settings and the boundaries users should keep around sensitive personal information.
Price & value
Total realistic cost—not just the headline subscription, but credits, tokens and media upgrades that drive up spend for active users—measured against what you actually get.
External signals
How we sanity-check our scores
Editorial judgement is calibrated against public reputation signals so a score is not just one team's opinion. We weigh these signals by how much data backs them.
Trustpilot
We read the TrustScore and, more importantly, the volume and substance of reviews. A 2.0 from 400+ reviews carries real weight; a 4.5 from 4 reviews barely moves our score. We also note when a company's profile has been removed for breaching platform guidelines.
App Store & Google Play
For apps with a large mobile install base, store ratings (often tens or hundreds of thousands of reviews) are the most reliable signal and can outweigh a small or skewed Trustpilot page—especially for mainstream apps where review-bombing over a policy change distorts one platform but not another.
Sample size matters
A handful of reviews is noise, not signal. When the only external data is a tiny sample, we lean on product research and positioning instead and keep the score conservative rather than over-reacting to a few loud voices.
Cross-checking the product
We confirm each tool's real positioning, pricing model and content posture (SFW vs. uncensored) against its own live site before scoring, and re-verify when a brand changes domain or direction.
Honesty about limits
What our scores are—and are not
They are editorial estimates
Scores are informed editorial assessments built from public product information and the reputation signals above. They are not the result of long-term hands-on lab testing of every feature, and they are not a statistical average of verified user reviews.
They change
Pricing, features, content policies and ownership in this space move fast. A score reflects our latest review date. Always verify current pricing, plan limits and cancellation terms on the tool's own site before paying.
Affiliate links don't buy a score
Some outbound links may earn us a commission, and some listings are featured. Neither removes a tool's cons, 18+ labels or pricing cautions, and neither inflates its rating. See our affiliate disclosure.
Tell us if we're wrong
If you've used a tool and our score doesn't match your experience, that feedback is valuable—real usage is exactly what calibrates these estimates over time.
Evidence labels and a repeatable trial protocol
Every review separates a dated official-source check from an editorial estimate. Official feature check means we read the cited public page on the displayed date. Documented means a linked official document supports that specific fact. Ambiguous means the public source leaves the billing selection or wording unresolved. Pricing source checked and Policy checked describe documents, not a completed purchase, cancellation or security audit. Hands-on tested is reserved for a documented trial; we do not apply that label to the current directory.
- Record the provider, date, platform, region and actual plan. Use an original fictional adult character and no private photos or real relationship records.
- Give the same three-message task to each candidate. Record specific successes, errors, refusals and unavailable features separately.
- For memory, provide three fictional facts, change topic and retest later without leading clues. For media, compare one permitted neutral prompt and record the credit cost.
- Inspect the current policy, cancellation route and account deletion instructions. Save the renewal price, token limits and any exceptions.
- Publish the sample prompts, conditions and observed results before using a tested label. Disclose gifted access or sponsorship. Do not change scores because of payout.
Current gaps: there is no standardized paid-account benchmark for all 26 products, no independently measured voice-quality ranking and no certification that provider chat data is confidential. Documented prices may be regional or promotional; unverified prices are not presented as current quotes. Annual monthly equivalents are not monthly bills. Partner selections are not verified discounts.
2026-09-28 editorial update: added explicit evidence limits, dated billing and policy checks for four products, individual Sponsored labels and buying checklists. Existing scores remain editorial estimates. Read the public editorial update log; no new hands-on results are claimed.
Billing quotes and platform guidance
Quote required means a public document or store listing was checked, but the selected account, tier, platform or regional charge remains unresolved. Such products are not assigned a comparable monthly amount in the directory filter. Platform guidance identifies instructions from a billing platform such as Apple, not an operated product account control.
The September 28 checks cover Candy, OurDream, DreamGF, Kupid, Nomi, Kindroid, Character.AI, CHAI, Replika, DarLink, Lovescape, Secrets and MyLovely. Thirteen products have dated evidence; the other 13 are not newly verified. Character.AI's privacy check covers its model-training explanation, not every retention clause. Source dates belong to individual products and do not renew merely because the site is rebuilt.
Conflicting documents are marked ambiguous, not resolved by guessing: DarLink's cancellation policies differ. Secrets' tier guide contradicts itself on free audio access, while its help and terms differ on post-cancellation account access. Lovescape's current tier selection also differs from older Premium help wording. Read the linked evidence and clarify any purchase-critical discrepancy with the provider. No paid-account, cancellation or independent quality test is claimed. See the change record.