Dialogue design
One agent, consistent across every channel
Businesses build a website bot, then a WhatsApp bot, then a voice agent. Three systems, three sets of answers, and a customer who gets different information depending on how they asked.
Quick answer
What is conversational AI?
Conversational AI is software that holds natural dialogue with people across channels — web chat, messaging apps, SMS and voice. Built properly it uses one knowledge base and one set of behavioural rules across all of them, so a customer receives consistent answers regardless of how they made contact.
The problem
What this actually fixes
The website bot and the WhatsApp bot give different answers to the same question.
A customer starts a conversation on chat and has to repeat everything on the phone.
Each channel was built separately, so improving one improves none of the others.
Nobody can say what the agents are actually telling customers.
Voice and chat need different behaviour and were built identically.
Conversation history is scattered across three tools.
Scope
What's included
Every engagement is scoped to what you actually need. This is the full deliverable list.
Deliverable 01
Shared knowledge base
One source of truth for what agents may say, so updating an answer updates it everywhere rather than in one channel.
Deliverable 02
Channel-appropriate behaviour
The same knowledge expressed differently — voice needs shorter turns and no lists; chat can be more detailed.
Deliverable 03
Context continuity
A conversation that starts in chat and continues on the phone carries what was already established.
Deliverable 04
Consistent guardrails
One set of rules about what agents may claim, applied everywhere rather than reimplemented per channel.
Deliverable 05
Unified CRM logging
Every conversation on the contact record regardless of channel, so the history is complete.
Deliverable 06
Escalation design
Handoff to a person that works consistently, with context, from any channel.
How it works
From first call to running system
- Step
Define the knowledge boundary
What agents may answer and what they must escalate — decided once, applied everywhere.
- Step
Design per channel
The same intents expressed appropriately for voice versus text, which are genuinely different mediums.
- Step
Build and connect
Agents built against the shared knowledge base and wired to one CRM.
- Step
Review transcripts
What agents actually said, reviewed across channels and corrected centrally.
Build the knowledge base first
The instinct is to build a bot. The better order is to decide what the business is willing to have said on its behalf, then build agents that say it.
That distinction sounds academic until the third channel, when someone asks whether the voice agent quotes the same prices as the website chat and nobody knows. By then there are three places to check and three places to update.
One knowledge base, one set of boundaries, agents that express them appropriately per channel. It is more work at the start and considerably less work every month afterwards.
Stack
What this connects to
- OpenAILanguage models for qualification, summarisation and reply drafting.
- VapiProgrammable voice agents with low-latency speech and function calling.
- Retell AIConversational voice agents for inbound answering and outbound calling.
- WhatsApp BusinessTwo-way messaging on the channel much of the world actually uses.
- GoHighLevelCRM, pipelines, funnels, calendars and workflows — the system of record.
Questions
Conversational AI — common questions
Why not just build each channel separately?
Because they drift. Three separately built agents mean three knowledge bases, three sets of rules and three places to update when your pricing changes. Within months they disagree, and the customer who checks two channels notices before you do.
Should voice and chat behave identically?
No — they should say consistent things differently. Voice needs short turns, no lists and tolerance for interruption. Chat can be more detailed and can offer links. Same knowledge, different expression. Building one and reusing it verbatim produces an agent that is wrong on at least one channel.
Can context carry between channels?
Where the contact can be identified, yes — a conversation that started in chat can be referenced on a subsequent call. It requires the CRM to be the shared record rather than each channel holding its own history, which is the main architectural reason to build this way.
How do you keep answers accurate?
By grounding agents in a maintained knowledge base rather than letting them improvise, and by reviewing transcripts. An agent that has been running for months without anyone reading what it said is a genuine risk, regardless of how well it was built initially.
Which channel should we start with?
Wherever your enquiries actually arrive. Businesses that get phone calls should start with voice; those whose enquiries come through the website or Instagram should start there. Starting with the more interesting technology rather than the busier channel is a common mistake.
Related services
A website chatbot that books, not one that deflects
An AI chatbot answers visitor questions from your own content, qualifies them against your criteria, captures contact details into the CRM, books appointments against live calendar availability, and hands conversations to a human with full context when the visitor needs one or asks for one.
AI voice agents that hold a real conversation
An AI voice agent is software that conducts a spoken phone conversation in natural language. It understands what a caller says, responds in real time, follows the rules it was given, and takes actions such as booking an appointment, creating a CRM record or transferring to a human when the conversation requires one.
Deflect the questions that never needed a person
AI customer support handles documented, repetitive questions — order status, delivery times, returns policy, account and billing basics, how-to questions covered by existing documentation. It escalates anything involving a decision, a complaint, an exception or information it cannot verify, passing the conversation context to a person.
WhatsApp automation on the channel people actually reply to
WhatsApp automation uses the WhatsApp Business API to send and receive messages programmatically. Outbound messages outside an active conversation must use templates pre-approved by Meta. Once a customer replies, a 24-hour window opens in which free-form messages are allowed, which is when automated conversations and AI replies can run.
Turn Instagram DMs into booked appointments
Instagram DM automation uses Meta's messaging API to respond automatically to direct messages, comment triggers and story replies. Messages can be answered instantly, qualified conversationally, and converted into CRM contacts and booked appointments, subject to Meta's 24-hour messaging window rules.
Ready to put your growth on autopilot?
GoHighLevel, AI agents, CRM automation and workflows — engineered around how your business actually operates.
Free · 30 minutes · no obligation