Aircall Outbound AI Voice Campaigns and Gemini 3.8 Live: What September’s Voice AI Launches Mean for SMBs

01What Aircall shipped

On 22 September 2026 Aircall published its first monthly AI Voice Agents roundup, and the headline is Outbound Campaigns: build a campaign, upload contacts, choose an AI Voice Agent and launch. The agent works through the list handling conversations end to end, from answering questions and qualifying leads to booking meetings and pulling in a human when needed, with connection rates and outcomes tracked as the campaign runs. Aircall says it works for 10 contacts or 5,000, is available on every plan, and uses your existing AI Voice Agent minutes.

The rest of the release fills in what outbound needs to work:

• Templates for common inbound use cases such as FAQs and lead qualification (English-speaking markets first), with a Shopify-integrated order management agent announced as coming soon.
• Deeper Salesforce integration: agents can use Salesforce Knowledge articles, look up the caller as the call begins, and create or update cases and opportunities mid-call.
• 13 new languages, taking the total to 38, including Hindi, Tamil, Mandarin, Cantonese, Japanese, Korean, Vietnamese and Malay, plus a filterable voice library.
• Shared knowledge from Zendesk, Salesforce, HubSpot or Document360, and Agent Memory so a transferred caller does not repeat themselves; Aircall reports 69% less handling time for the receiving agent in its pilot.

02What Google shipped: Gemini 3.8 Live

A week earlier, on 15 September, Google released Gemini 3.8 Live and Gemini 3.8 Live Extended Thinking, which it calls its most advanced live dialogue models. The design goal is a voice agent that reasons and talks at the same time: it acknowledges a request (“let me check that”), runs tool calls and API requests in the background, and narrates progress without dead air. Gemini 3.8 Live detects and switches between 97 languages mid-conversation.

On benchmarks, Google reports Extended Thinking at 82.6 on Artificial Analysis’ Speech to Speech Quality Index (the top overall spot) and 68.6% on the τ-Voice agentic benchmark, while the lighter Gemini 3.8 Live placed second in the Speech Agent Arena. Both models are available to developers through the Gemini API and AI Studio, with platforms such as LiveKit, Pipecat and LangChain supporting the Live API, and all generated audio carries Google’s SynthID watermark. Treat vendor-reported benchmarks with caution until independent replication appears, but the direction is clear.

03What this means for small business phone lines

1. Outbound follow-up is now table stakes. For many SMBs, slow follow-up leaks as much revenue as missed inbound calls: quote requests never chased, no-shows never rebooked, renewals left to chance. When a mainstream phone system ships AI outbound campaigns on every plan, prospects start to expect a call-back within minutes. If your AI setup only answers the phone, it is half a system.

2. Multilingual is no longer a premium feature. Between Aircall’s 38 languages, Gemini’s 97 and ElevenLabs’ 70+, serving Spanish, Hindi or Arabic-speaking callers no longer needs a separate team. That matters most for clinics, trades and hospitality in mixed-language cities; see our multilingual AI receptionist guide.

3. CRM write-back is the real differentiator. Aircall leading with Salesforce read/write is the tell. An agent that talks well but does not update your CRM just creates admin. Whatever you deploy, insist on the caller lookup at the start and the record update at the end.

4. Do not lock in to one model. Google shipped four Gemini 3.8 models in two weeks. The best-sounding, cheapest model in January will not be the one you launched on in September. Platforms that let you swap the underlying model, such as those in our Vapi vs ElevenLabs and Retell vs ElevenLabs comparisons, protect you from that churn.

04Should you switch to Aircall, or build on Gemini Live?

If you already run Aircall as your phone system, try Outbound Campaigns on a low-risk list first (appointment reminders or lapsed-customer check-ins) before pointing it at new leads, and check your minute allowance since campaigns draw from the same pool.

If you are building a custom voice agent, Gemini 3.8 Live is worth testing as the model layer, especially for multilingual or tool-heavy calls, but benchmark it on your own scripts against your current setup before switching production traffic.

If you just want it working, that is what our AI voice agent service is for: inbound answering plus outbound follow-up, connected to your CRM, on whichever model is best this quarter. Book a free 15-minute call.