Testing methodology

    How we test AI companion apps

    We do not claim to personally test every listed product. A “Hands-on tested” label means a specific test has a named scope and retained evidence. Current prices, limits, and policies are checked separately, and community reports remain clearly attributed.

    Published and last updated August 28, 2026

    1. 1

      Define the reader's question

      We identify the search intent and the decisions a useful comparison must answer. A video ranking, privacy guide, and roleplay comparison do not use the same criteria.

    2. 2

      Assign an evidence status

      Before making a claim, we record whether it comes from hands-on testing, an official fact check, community information, or limited evidence.

    3. 3

      Test a documented scope

      Hands-on work records the tester, date, features tried, notes, and retained screenshots or media. The label applies only to that recorded scope—not every feature the product advertises.

    4. 4

      Verify changing commercial facts

      Prices, free access, credits, limits, billing periods, and policy claims are checked separately against current first-party sources whenever possible.

    5. 5

      Compare fit and trade-offs

      We evaluate the criteria that matter for the query, explain who each option suits, and include reasons a product may not be ideal.

    6. 6

      Publish, monitor, and correct

      We publish a fixed editorial result, retain its supporting records, review aging evidence, and correct material errors when verified.

    What a hands-on record contains

    • The tester and test date.
    • The features and workflow actually tried.
    • Notes about results, limits, failures, and uncertainty.
    • Retained screenshots, clips, or other supporting media where appropriate.
    • Separate first-party sources for changing prices and commercial terms.

    Some retained evidence may not be displayed publicly because it contains account information, licensed material, or explicit output. Its existence does not expand the published test scope.

    What we may test by category

    • Chat quality, roleplay controls, continuity, and memory.
    • Character creation and customization options.
    • Image or video workflow, output consistency, controls, retries, and generation cost.
    • Voice features, usability, queues, and device experience.
    • Free access, paywalls, subscription inclusions, credits, and refill costs.
    • Published privacy, deletion, billing, and account-management information.

    Not every review covers every criterion. Each page should state what was tested and what remains based on official or community information.

    Freshness and uncertainty

    Evidence older than 180 days is treated internally as needing a freshness review. A newer fact-check date does not imply that every hands-on observation was repeated. When a reliable current fact is unavailable, we use a qualified label such as “See current price,” “not published,” or “evidence limited” rather than guessing.

    Testing shows what happened in a defined sample. It cannot guarantee identical output, moderation behavior, queue time, billing treatment, or support quality for every account and region.