AI voice agents
Asked as “AI voice agent platform”, and as “AI phone agent that answers customer calls”, on behalf of a mid-market B2B software company. 63 first choices recorded across the direct, paraphrase, budget and scale prompts, six models each. Added to the September 2026 Edition; its answers are read by the distilled judge (ai-indexes-judge-qwen3-14b), not the claude-opus-5 judge of the earlier categories: how the two compare.
Standing · first-choice share
27%
Contested · CloudTalk 19%
27Retell AI19CloudTalk13Vapi41others
27% of first choices, contested.
The standing
Share is the count of first choices across the direct, paraphrase, budget and scale prompts, over all six models, for a mid-market B2B software company. Ordered by share.
ProductFirst-choice shareNegative rateLabelsQuadrantSince September 2026
Show the two products at 0%, ordered by negative rate
The floor is 12 points of share, measured: how far the models move a leader on their own when the same questions are asked twice with nothing changed. A larger change is movement; a smaller one is noise, and both are shown. How the floor is measured
Bars are the share of first choices, 0 to 100Every product with at least 10 labels here. Every product name links to its product page.
Recommended versus criticized
Every product with at least 10 labels here, on both axes. The 30% line names a quadrant, not the verdict above: that one needs more than 40%.
Criticized challengerCriticized default
Negative label rate →
01
02
03
04
05
06
07
08Accepted challengerEndorsed leader
0%First-choice share → · lines at 30% share and 25% negative40%
Key
01Retell AI27%
Retell AI27%02CloudTalk19%
CloudTalk19%03Vapi13%
Vapi13%04Bland AI3%
Bland AI3%05Synthflow3%
Synthflow3%06Sierra2%
Sierra2%07PolyAI0%
PolyAI0%08ElevenLabs0%
ElevenLabs0%What they warned about
A high negative share on a product with few labels is a warning. A low share on a product with many labels is salience, not sentiment.
Vapi33%
13 of 39 labels negative · 9 of 6 models · 1 hard negative
“**Vapi** — Pricing Concerns - **55% price increase in December 2025**” Kimi K2, negative prompt
Retell AI16%
9 of 58 labels negative · 9 of 6 models
“some specific products (for example Synthflow, ElevenLabs, Retell AI, Deepgram) have recurring complaints about cost or credit limits” GPT-5.4 mini, negative prompt
ElevenLabs58%
7 of 12 labels negative · 7 of 6 models · 2 hard negative
“Avoid AI voice platforms that: Have been flagged by independent testing for weak anti-deepfake controls (e.g., ElevenLabs, Speechify, PlayHT, Lovo, Resemble AI, Descript).” GLM 4.7 FlashX, negative prompt
Synthflow22%
7 of 32 labels negative · 6 of 6 models · 2 hard negative
“**Why avoid:** If your call volumes exceed 5,000 calls per month, Synthflow may not scale well for you.” Mistral Small, negative prompt
What they cite
Citations exist only for the models that return a source list: fourteen of the six in this edition.
Sites the answers cite
78 of 84 answers in this category came back with a source list, from 14 of 6 models: citations where the model returns them, or the search results it consulted. 1140 links across 291 sites, every framing counted. Ranked by the number of answers carrying the site or page.
vendor site · CloudTalk53 answers · 130 citations · 12 models
vendor site · Retell AI50 answers · 103 citations · 12 models
25 answers · 28 citations · 11 models
vendor site · Brilo.ai21 answers · 34 citations · 11 models
vendor site · Synthflow21 answers · 30 citations · 9 models
vendor site · Lindy19 answers · 21 citations · 9 models
19 answers · 19 citations · 11 models
vendor site · Aloware18 answers · 20 citations · 11 models
vendor site · Bland17 answers · 28 citations · 11 models
vendor site · Murf17 answers · 19 citations · 9 models
14 answers · 17 citations · 8 models
vendor site · JustCall13 answers · 20 citations · 9 models
Pages the answers cite
The ten pages named in the most answers, by full address. A page here is one the models returned with a recommendation, not one the index endorses.
22 answers · 9 models
20 answers · 10 models
20 answers · 9 models
17 answers · 11 models
17 answers · 8 models
16 answers · 10 models
16 answers · 10 models
15 answers · 10 models
12 answers · 7 models
10 answers · 7 models
Search against answers
Each company's standing in the answers beside its site's footprint in Google search, one row a site: the products the models named on it with their shares, and the share they add up to; monthly searches on Google, and DataForSEO's estimate of AI search demand (modeled from search signals, directional, not a count of queries to any assistant), for the most-searched of the company's and its products' names (the name is in each row's hover text); estimated monthly organic visits to the site; and its best position in Google's top ten for “ai voice agent platform”, “best ai voice agent platform”. US estimates from DataForSEO and Google's Ads Transparency Center. A small company's site, or a mid-sized company's site for its flagship, is marked company; a product on a large parent's site (Google, Microsoft) has no site figures. A column with no figures for this category is left out, and an empty cell means none were seen, not none exist. Two measurements side by side: neither is read as the cause of the other.
| Company and products | Share | Own site cited | Name searches, Google | AI search demand, est. | Organic visits | Paid search a month, est. | Google ads, last 30 days | Best Google position |
|---|---|---|---|---|---|---|---|---|
| Retell AI retellai.com | 27% | 155 | 22,200 | 177 | 105,030 | 14 | 3 | |
| CloudTalk cloudtalk.io CloudTalk 19%, CloudTalk AI Voice Agent 2% | 21% | 358 | 2,900 | 62 | 95,515 | $1,909 | 119 | 2 |
| Vapi vapi.ai Vapi 13%, Vapi.ai 3% | 16% | 102 | 14,800 | 468 | 30,606 | 13 | ||
| Aloware aloware.com | 8% | 47 | 2,900 | 13 | 27,544 | $442 | 24 | |
| Synthflow synthflow.ai Synthflow 3%, Synthflow AI 2% | 5% | 77 | 2,400 | 24 | 15,596 | 0 | ||
| Bland AI bland.ai | 3% | 102 | 5,400 | 115 | 11,867 | $880 | 21 | |
| Brilo.ai brilo.ai | 3% | 12 | ||||||
| Goodcall goodcall.com | 3% | 31 | 1,900 | 9 | 1,057,553 | 1 | ||
| Regal regal.ai | 3% | 7 | ||||||
| Talkdesk talkdesk.com | 2% | 187 | 8,100 | 245 | 72,323 | $50 | 22 |
Google's top ten for the two searches hold 15 sites; 3 of them are among the sites the models cited here (cloudtalk.io, lumay.ai, retellai.com). The first result for “ai voice agent platform” is callrail.com.
How they answered
Six framings of the same buying question, each sent to every model in a fresh session with search on. One row per model, so a row shows whether it held its answer under rewording, what it named when cost was the constraint, and what it argued against. Computed from the raw judge labels.ShowHide
How they answered
Six framings of the same buying question, each sent to every model in a fresh session with search on. One row per model, so a row shows whether it held its answer under rewording, what it named when cost was the constraint, and what it argued against. Computed from the raw judge labels.
ShowHideModelDirect“What is the best AI voice agent platform for a mid-market B2B company?”Paraphrase“Which AI phone agent that answers customer calls would you recommend to a mid-sized B2B company?”Comparative“What are the top AI voice agent platforms and how do they differ?”Budget-constrained“What is the best AI voice agent platform for a company with a limited budget?”Scale-constrained“We are a 500 person company evaluating an AI voice agent platform. What should we look at?”Negative“Which AI voice agent platforms should I avoid or be cautious about?”
Claude Opus 5no first choiceno first choiceno first choiceno first choiceno first choicenothing named
Claude Opus 4.8no first choiceno first choiceno first choiceno first choiceno first choicenothing named
GPT-6 Astrano first choiceno first choiceno first choiceno first choiceno first choicenothing named
GPT-5.6 Solno first choiceno first choiceno first choiceno first choiceno first choicenothing named
Gemini 3.1 Prono first choiceno first choiceno first choiceno first choiceno first choicenothing named
Perplexity Sonar Prono first choiceno first choiceno first choiceno first choiceno first choicenothing named
Bold is the first choiceAlternatives are counted; the count opens them.What the answer argued against
The record
One row per call: the version string exactly as returned, whether the model searched, sources cited, and latency. Full answer text is in the free responses file. Download the record
Zero rows: every prompt, every model, every answer.
| Prompt | Model | Version string | Time (UTC) | Searched | Sources | Latency |
|---|
Normalization in this category
Every judgment call made between the raw labels and the numbers above, listed so it is visible and reversible.
ShowHide
Category-scoped readings
Bland read as Bland AI
Deepgram read as Deepgram Voice Agent API
Retell read as Retell AI
Salesforce read as Salesforce Agentforce Voice
Salesforce Agentforce read as Salesforce Agentforce Voice
Unresolved, counted raw
Astra Voice 2.0
Bland.AI
Bland.ai
CallSphere
CloudTalk AI Voice Agent
GetVoIP
Google Cloud
Google’s contact-center offerings
HubSpot Breeze AI
Inworld AI
JustCall's AI Voice Agent
Kore
LOVO AI (Genny)
Leaping AI
OpenAI
Presence
Quo (formerly OpenPhone) Sona
Rezora IO
Sobot Voicebot
SquadStack AI
Talkdesk AI
Tone
Twilio
Vocode
Watson
Zendesk Voice AI Agents
Discontinued, still offered
No shut-down product was recommended here.