How we compare AI receptionists
Our goal is to make purchase decisions more measurable while avoiding fake first-hand claims.
Evidence levels
- Vendor-verified fact: current price, included usage, integrations or documented feature from a primary source.
- Independent judgment: our view of which business model or use case appears to fit the public facts.
- Hands-on finding: reserved for behavior we personally test with a repeatable scenario. We do not use this label yet where we have not performed the test.
What does not affect the score
A higher commission rate does not earn a product extra points. Affiliate relationships are disclosed separately.
What we record
For material pricing and feature claims we store the source URL, source type, date checked and a short claim note in our internal source ledger. If a vendor changes price, we update the article rather than pretending old pricing is current.
Future hands-on test protocol
- Noise and interruption handling
- Names, addresses and spelling
- Appointment edge cases
- Human-transfer behavior
- Unsupported questions and safe refusal
- Transcript/summary accuracy
- Cost per successfully handled scenario
Tool-first decision model
We prioritize pages that help a buyer make a decision: normalized cost calculations, feature-fit diagnostics and sourced data. General articles support those tools rather than being the product themselves.
Reusable data architecture
Public vendor facts are stored once in a structured data layer with verification dates and source URLs. The same data can feed calculators, comparisons, FAQs and future exports. We do not let affiliate commission influence vendor ranking inside buyer tools.
Change control
Pricing pages change often. Our update process is conservative: detect source changes first, review the changed vendor material, update structured data, validate calculations, then publish. We do not silently scrape a changed number straight into a recommendation.