How We Review

The overall score is the weighted sum of these eight values, rounded to one decimal. The same criterion meaning and service score appear on the home page, reviews hub and every deep review. A fact outside the published evidence is excluded from scoring rather than converted into a secret feature or an automatic zero.

An eight indicates strong public evidence, a clear fit for the criterion and only limited uncertainty. For anime fit, that means identifiable adult anime characters or meaningful style controls, not a generic toggle. For price clarity, it means a current amount, period and enough allowance detail to estimate entry cost, even if regional tax can still vary.

An eight is not perfection and does not promise identical results for every user. Public documentation can establish availability and intended function without proving response quality, output speed or consistency for every prompt.

A five indicates partial fit or evidence that is too broad to support a stronger judgement. A service may describe memory without showing how it is edited, or display a subscription price while leaving media currency unexplained. The feature can exist and still receive a middle score because the decision-relevant conditions remain unclear.

Scores below five usually reflect a weak category fit, a major access problem or limited attributable detail. JoyAI’s lower overall score reflects a Telegram-first product with a compact setup flow and less pre-purchase feature and pricing detail than the web-first services above it.

We use “reviewed,” “compared” and “checked” for public research. A first-party product, help, pricing or policy page is the strongest source for what the service says it supports. Public previews and supplied service-specific media can clarify presentation, but they do not prove output quality. Third-party material may help identify a subject; it does not override current first-party facts.

When pages conflict, the review says so. When a price is gated, the review does not copy an unattributed amount from a mirror. When two products share a similar name, capabilities stay with the identity that published them. These rules are especially important for fast-changing companion products.

Account-only checkout totals, regional taxes, private creator controls, memory duration, response quality, generation speed and the cost of every media action fall outside the public review surface. We also do not audit infrastructure or data retention beyond published documentation. Readers should inspect live terms, payment screens and deletion controls before sharing sensitive information or paying.

The lack of a hands-on account test is a deliberate boundary of this edition, not a hidden claim. If future work includes a recorded public demo or purchased plan, the relevant review will name the interface, date, account state and exact task observed.

The review set should be refreshed when a plan changes, a service launches or removes a core mode, public access rules shift or an ambiguous identity gains a stable first-party page. A scheduled refresh updates the checked month, price and access statements, comparison table, visible score explanation and structured data together.

Readers can report a changed public fact through the contact page. A correction requires an attributable source and triggers review of every page carrying that value, so the ranking and deep reviews do not drift apart.