How We Review AI Companions
Where our information comes from, how facts are verified, and how ratings are calculated.
- Published
- Updated
Our aim is to make AI companion apps comparable using information that can be checked. This page explains every step, including what we do not do.
1. Sources
We prioritise sources in this order: the platform's official website, its pricing page, its feature, help and documentation pages, its terms and privacy pages, direct product testing, and finally reputable independent sources when first-party sources cannot answer a question. We do not treat search snippets as evidence, and we do not copy competing affiliate sites.
2. Verification status
Every important product fact is stored with its value, source URL, the date it was checked, and a status. Only facts with the status 'verified' appear on public pages.
- Verified: confirmed against a first-party source on the date shown.
- Needs review: seen but ambiguous, promotional or incomplete. Not shown publicly.
- Unknown: not found. Not shown publicly, and never guessed.
3. Hands-on testing
Testing is recorded as its own dataset: who tested, when, notes and evidence such as screenshots or logs. We only use testing language ('we tested…') on a page when that record exists. Until then, reviews are labelled as based on published information.
4. Ratings
An overall rating is the weighted average of the dimensions below, each scored from 0 to 10 using a written rubric. An overall rating is published only when every dimension has a verified score. Until then no rating is shown, and platforms are listed alphabetically rather than ranked.
5. Freshness
AI companion products change often. Each platform has a 'last verified' date that changes only when facts are re-checked, never when the site is rebuilt. Data older than 90 days is flagged for re-verification.
6. Awards
Awards such as Editor's Choice or Best Value are assigned by editors from rated data, with a written rationale. Awards are not assigned while the underlying ratings do not exist.
7. Independence
Affiliate commissions do not influence any of the above. See our editorial policy and affiliate disclosure.
Appendix: rating dimensions and weights
Each dimension is scored from 0 to 10 against a written rubric. The overall rating is the weighted average, and it is published only when every dimension has a verified score.
| Dimension | What it measures | Weight |
|---|---|---|
| Conversation quality | Coherence, responsiveness and consistency of conversation over a long session. | 18% |
| Realism | How natural the character feels across text, voice and media. | 12% |
| Memory | How reliably details from earlier conversations are recalled. | 12% |
| Customisation | Depth and control of character creation and personality settings. | 10% |
| Image generation | Quality, consistency and control of generated images. | 10% |
| Voice | Quality and flexibility of voice messages and calls. | 8% |
| Video | Quality and availability of generated video. | 6% |
| Ease of use | Onboarding, interface clarity and day-to-day usability. | 8% |
| Privacy and transparency | Clarity of policies, data handling disclosures and billing practices. | 6% |
| Value for money | What you get relative to what it costs, including usage limits. | 10% |