CARES+ is the seven-dimension scoring framework CyberPals uses to evaluate every AI plush companion we test. It exists because the existing review playbooks — tech-magazine spec sheets, parenting-blog vibes-checks, and AI-toy unboxings on YouTube — all miss what actually makes or breaks an AI pet over a six-month relationship. CARES+ is built to surface that.
The seven dimensions
| Dimension | Full Name | What we test | Weight |
|---|---|---|---|
| C1 | Emotional Companionship | Facial expression range, body movement, ritual moments, response to touch and voice tone, “does it feel alive” over a 30-day window. | 9% |
| C2 | Conversational Depth | LLM behind the product, memory persistence, ability to learn user-specific facts, multi-turn dialog quality, language coverage. | 9% |
| A | App & Cloud Reliability | App stability across iOS / Android, cloud uptime over 30 days, OTA update cadence, what happens when the cloud is offline. | 14% |
| R | Real-World Build | Materials, fit and finish, battery life under realistic load, charging behavior, cleanability, repairability, safety certifications. | 12% |
| E | Emotional Fit | Persona-to-product matching: who is this actually for? Does it fit the buyer’s lifestyle, age, expectations, and emotional needs? | 16% |
| S | Support Lifecycle | Warranty terms, brand survival risk, firmware update commitment, privacy policy, data handling, refund and return reality. | 12% |
| K / B | Kid Safety or Bond & Bring-Your-Own | For S1 (kids): COPPA compliance, content filtering, parental controls, age appropriateness. For S0 (collectibles): pairing logic, cross-device bonding, social and collectible value, future-proofing. | 14% |
Total weight is 100%. The K / B slot swaps depending on whether the product is targeted at children or adult collectors. For S2 (adult wellness and elderly care), we use a modified rubric where R and S carry slightly more weight to reflect long-term reliability needs.
How a review is built
- Step 1 — Sourcing the unit. We buy our own units whenever possible, on the same channels a real customer would use (Amazon, Tmall, brand-direct). When manufacturers send a sample, we disclose it at the top of the article.
- Step 2 — Daily-use period. Minimum 14 days of real-world use across multiple environments — desk, living room, travel — before a single word of review is drafted. For S2 elderly-care products, the minimum is 30 days.
- Step 3 — CARES+ scoring. Each dimension gets an evidence-backed 0–5 score with a written rationale. Scores cannot be assigned without at least one specific observation.
- Step 4 — Privacy probe. For any product with a microphone, camera, or cloud connection, our engineering partner runs a network capture to identify what data is leaving the device, where it’s going, and whether it’s encrypted in transit. Findings go into the S dimension.
- Step 5 — Persona fit. We map the product against our four reference Personas (Maya 27 collector / Marcus 34 nostalgic adult / Sarah 38 parent / Linda 52 elder-care daughter) and call out which Persona it’s actually for.
- Step 6 — Edit and publish. Editor-in-chief reviews the draft against the rubric, brand-specific claims are double-checked against primary sources, and the article ships with full citations.
How we avoid bias
- Equal scoring across brand size. A scrappy startup like Pophie and an OEM giant like Hisense Moii get the same CARES+ rubric. Brand fame does not move a score.
- Equal scoring across commission status. Brands we plan to partner with via affiliate (Ropet, Enabot at T+30) and brands we don’t are scored identically. Commission status is disclosed but never weighted.
- Mandatory devil’s advocate. For every recommended pick, we are required to write the strongest case against the product. If we can’t write a credible counter-case, the recommendation is held until we can.
- Three-bias guardrail. Every review is reviewed for survivorship bias (only looking at units that survived), confirmation bias (cherry-picking evidence that fits a thesis), and hindsight bias (rewriting expectations after seeing outcomes).
- External fact-check. Manufacturer specs are verified against at least two independent sources before publication: brand site, retailer listing, third-party teardown, or regulatory filing.
When we update
AI products move fast. Firmware changes can rewrite a product’s personality overnight. We re-test every reviewed product at least once per quarter and rerun the CARES+ scoring whenever:
- A major firmware or app update ships
- The price changes by more than 15%
- A privacy policy or data-handling change is announced
- A new product in the same brand line is released that affects positioning
- A reader or external researcher flags a finding that contradicts our score
Updated reviews carry an “Updated on” timestamp and a short changelog at the top so you can see exactly what moved.
What CARES+ isn’t
CARES+ is not a benchmark, not a certification, and not a substitute for trying a product yourself. It’s a structured way for us to be honest with our readers and consistent with ourselves over time. If you disagree with a score, write to us at hello@cyberpals.tech with your reasoning — the framework is built to be argued with.
Read our reviews
The flagship application of CARES+ is our pillar review: Best AI Plush Toys 2026 — We Tested 7 Chinese AI Pals.