AI Receptionist Benchmark: How VantaWeb Stacks Up Against Smith.ai, Goodcall, and Arini

Methodology

This benchmark combines two types of data: (1) publicly verifiable facts about each platform — pricing, features, integrations, compliance — sourced directly from each vendor's website with citations; and (2) standardized live call tests across 5 real-world scenarios, scored on 5 dimensions.

Each scenario is run against each platform's live demo or production number. Calls are scored on: Responsiveness (first-word latency), Comprehension (did the AI understand what the caller needed?), Empathy (appropriate tone), Completeness (all necessary information captured), and Accuracy (no hallucinated info). Score = average of 5 dimensions × 20 = % out of 100.

The 5 evaluation scenarios

Scoring rubric (5 dimensions, 1–5 each)

Platform Comparison

Every cell with public data is sourced. The small text below each value links to the page where it was found. Cells reading "Not publicly disclosed" were not findable on public pages as of 2026-05-24.

Scroll right to see all columns →

All public data sourced as of 2026-05-24. Source URLs linked inline. "Not publicly disclosed" = not found on vendor's public-facing pages as of that date.

Methodology limitations

We publish the caveats, not just the scores.

• Scenario-based scoring is partly subjective. When published, raw call recordings and per-call scoring notes will be linked from each score cell. A methodology page at /benchmark/methodology ships with the first test cycle.
• Pricing changes. We re-fetch all pricing pages monthly and update the table with the re-fetch date.
• VantaWeb (Anna) is rated by an independent peer reviewer to avoid self-serving scores. Until a third party is formalized, Hunter scores his own platform with the same rubric — and those scores will be labeled "self-scored" until independently verified.
• Arini does not publish pricing publicly. Their comparison cells reflect only what is verifiable without a sales conversation.
• Smith.ai's AI Receptionist product is separate from their Virtual Receptionists product. Comparison data here targets the AI Receptionist where distinguishable on their site.
• G2 and Capterra block automated scraping. Where vendors display badge ratings on their own sites, we cite the vendor page as source. Check G2 directly for current review counts.
• The "Latency claim" row is not apples-to-apples: VantaWeb's "< 300ms STT latency" measures a single pipeline component (speech-to-text), while Arini's "300ms streaming latency" is presented as an end-to-end figure. Only VantaWeb's separately-published sub-second first-word latency (see above) is directly comparable to a competitor's end-to-end claim.