How I test AI companions
I’ve tested the AI companion apps I cover across their features. My recommendations draw on that experience, with documentation used to verify changing details and published research attributed to its authors.
Comparable evidence
First-person recommendations reflect the named author's experience and preferences. I distinguish those judgments from documented product features and controlled test results. Extensive personal use does not imply that every product was tested with an identical protocol.
For test-based rankings, products receive the same core tests: onboarding, ordinary conversation, corrections, multi-session memory, initiative, controls, and paid-feature boundaries when they affect the verdict. I do not present a personal preference as a measured result.
Sources and claims
Material claims are mapped to dated sources or test records. Product pages establish what a company says; they do not establish that the feature works. Community reports provide useful leads but are labeled and corroborated before they support a conclusion.
Updates
A visible “Updated Month Year” label means the article received a substantive evidence or conclusion change. Quiet formatting and typo fixes do not reset the date. The update note describes what was rechecked or revised.