Five subscores
| No. | Subscore | Weight | What it measures | Evidence base |
|---|---|---|---|---|
| 01 | Voice | 20% | Call quality and latency, voice realism, availability of voice calls and a real phone line, hands-free session handling. | Hands-on testing, vendor documentation, community reports |
| 02 | Memory | 20% | Long-term recall across sessions, context handling, whether the companion remembers declared facts without re-prompting. | Hands-on memory probes, vendor docs, independent reviews |
| 03 | Value | 20% | What the free tier actually allows, price-to-feature ratio, credit/message economics, billing clarity. | Vendor pricing pages verified at research time |
| 04 | Privacy | 20% | Chat data retention policy, training-use disclosures, deletion rights, anonymous payment options. | Privacy policy documents, vendor statements |
| 05 | Policy | 20% | Content-policy clarity, refund terms, cancellation friction, charge appearance on statements. | Terms of service, payment processor records, user reports |
Scores are rendered as count-up rings on profile pages with the subscore label attached — voice, memory, value, privacy, policy — so a high overall number can never hide a weak subscore.
Verification policy
Every factual claim on this site traces to a source, and every source carries a verification state. Pricing figures are only marked Verified Oct 2026 when they were confirmed on the vendor's own pricing page at research time. Figures we could only find in third-party reviews are marked Unverified and never enter a score until confirmed.
Each platform's research lives in a structured fact sheet with per-source notes — including dates and how each source was checked. When a fact sheet is older than its freshness window, its profile is re-verified before scores update.
Where two sources conflict, the vendor's own documentation wins for factual capability claims; hands-on testing wins for quality claims; and conflicts are disclosed on the profile rather than silently resolved.
On “no clear winner”
Comparison pages only declare a winner when one platform leads on the subscores that matter for the stated use case. When the data is close — or when the right pick genuinely depends on what you weight — the verdict says exactly that. A comparison that always crowns a winner is a sales page, not a comparison.
Ownership disclosure
Companion Index is operated by the makers of FlirtAlly, an AI companion platform. FlirtAlly is profiled in this directory and scored by the same rubric as everyone else. It receives no ranking boost, no exemption from verification, and no star rating of any kind on this site. Where FlirtAlly appears in a comparison, the disclosure is repeated on the page.
No platform can pay for placement, a tier upgrade, or a verdict. If that ever changes, this page changes first.
Update cadence
The index updates monthly. Profiles carry a LAST VERIFIED stamp; pricing older than one cycle drops to UNVERIFIED until re-checked. Methodology changes are versioned — the current rubric is v1.0, published with the first profile batch in October 2026.
On this page