Voice AI

With companies like Nedzo that have ultra realistic voices, I think the HL voice ai has a long way to go. Would be good if the voices could sound less robotic.

This would be great for 2 reasons. 1) When one of your contacts is calling using the AI phone number they would be recognized " Hi John" Thanks for calling, how can I help you? 2) When doing voicemail drops to sell AI voice the caller could call back, be recognized and have a demo of what their missing on their call back.

Our team manages multiple clients within similar industries, where approximately 90% of the knowledge base content is identical, with only minor variations between accounts. Currently, there is no efficient way to reuse this content. Each time we onboard a new client, we have to manually recreate or copy and paste the knowledge base from scratch, which is time-consuming and increases the risk of inconsistencies or errors. Proposed Solution: Introduce the ability to clone or duplicate an existing knowledge base into another subaccount. Ideally, this would allow us to: Select an existing knowledge base as a template Clone it directly into a different subaccount Make minor edits as needed without rebuilding from scratch Impact: Significantly reduces onboarding time for new clients Ensures consistency across similar accounts Minimizes manual work and human error Improves scalability for agencies managing multiple clients

As we work towards offering AI receptionists to clients, it would be super helpful if the AI could train based on previous call transcripts. For example if we have someone call in and ask to speak with Patty, the AI may not realize they're asking for Patricia. Or maybe industry jargon is used on calls, it would be helpful if the AI can learn what all those terms are so it can respond properly. Some of this stuff only turns up in client calls, and in an ideal scenario, after the call the AI would ask the sub account owner or admin to clarify or add context to what was discussed on the call.

I would like to request HL allow higher concurrency limits (100) and allowance for Quiet Time calls for customers who sign up for coaching calls as part of their service and want early morning calls. Our target audience are High D, achievers, who want early morning calls (most of them will) or late at night. Not allowing those early and late calls that customers explicitly ask for and schedule eliminates HL as our provider, and concurrency limits throttles the ability to scale.

Since you have clients from all over the world, is is important to enable them to use the Voice AI angents in their country language. Otherwise you limit the access to this valuable feature for most of your clients.

The AI Voice Agent could experience spam phone calls. Could there be a way to block incoming numbers - I see this as being a problem with my GHL (business number). This could disrupt Automation and Workflow with bad information and fake contacts.

OUTBOUND VOICE AI SHOULD BE ABLE TO CALL DIFFERENT LISTS OF CONTACTS TO ASK THE SAME QUESTIONS AND COLLECT ALL THEIR RESPONSES. i, e, call all mechanics to get the their individual prices and select the ones in a particular price range if you want (or to use that data in a calculation like: take the average, median, least or max).

Voice AI works well for general conversation, but it has noticeable delays (typically 3–6 seconds of silence) when it needs to search the knowledge base before responding. This creates an unnatural pause during interactions. ### Proposed Improvements Pre-Response Acknowledgment Messages Add support for configurable spoken messages that the Voice AI can say immediately while it begins searching the knowledge base — similar to the existing "Custom Actions" feature (e.g., “Just a second while I check that”). Examples: - “Just a moment while I look that up…” - “Let me check our documentation for you…” Knowledge Base Prioritization Allow administrators to set the search priority/order of different knowledge bases. For example: - Search rich text documentation first (primary source) - Use the website or other sources only as fallback Currently, the AI seems to evaluate all knowledge bases every time, which is inefficient when some sources are more important or reliable than others. Improved Knowledge Base Search Speed Significantly optimize retrieval performance. Even with a modest knowledge base (one 20-page website + a couple of rich text documents), response times should be much faster than the current 6 seconds. ### Why This Matters These enhancements would make Voice AI feel much more natural and responsive in patron-facing situations, reduce awkward silences, and help libraries optimize the use of their documentation sources more efficiently.

GHL currently offers Deepgram and Azure in Custom transcription mode. Both are solid but fall short for multilingual deployments and domain-specific vocabulary. 🔹 ElevenLabs's Scribe covers 90+ languages at under 150ms latency, vs Deepgram's English-first focus 🔹 Keyterm prompting boosts accuracy for business names and niche terminology out of the box 🔹 Mid-conversation language switching, something neither Deepgram nor Azure handle well natively Adding Scribe as a third option in the STT Provider dropdown would be a direct, low-lift improvement that opens GHL Voice AI to non-English markets agencies are actively trying to serve.