HubSpot’s new Customer Agent represents one of the most significant platform updates in recent years. It replaces traditional scripted chatbots with a powerful AI assistant that answers questions, understands context, and guides visitors to the correct next step. However, like any AI model, the quality of its output hinges entirely on the quality of its setup.
Through our own implementation at Growth London and by supporting clients with early adoption, we have developed a structured, reliable approach that ensures the Customer Agent is accurate, helpful, and aligned with your brand. This guide walks you through exactly how to set it up correctly.
The Customer Agent is HubSpot’s AI-powered assistant that appears through your website’s chat widget, support inbox, or knowledge base. Its purpose is to:
It is not a chatbot in the traditional sense. It does not rely on scripts or decision trees. Instead, it uses a combination of:
When configured correctly, the Customer Agent operates as a knowledgeable, on-brand front line assistant for your business.
A Customer Agent is only as good as the foundation you give it. Before activating it, you need:
This should include your services, ICPs, case studies, FAQs, pain points, and fallback responses. Think of this as a single document that fully describes your business and what you do.
🧠 Top tip - you can use AI to brainstorm the outline and contents for this document
If your website is outdated, inconsistent, or unclear, the Customer Agent will inherit those weaknesses.
🧠 Top tip - you can carefully review and select exactly which website pages, landing pages and blog posts the agent has access to and which to exclude.
Tone, voice, and messaging need to be defined so that the agent behaves consistently. Take your time to set this up properly once and for all - it will pay dividends in the long run.
These are essential to prevent hallucinations and ensure factual accuracy.
For example, the agent should share your HubSpot Meetings booking link if the user asks about specific topics, such as pricing or talking to sales, etc.
The agent should have a clear view of who you serve, what you offer, and who is not a fit.
This preparation ensures your Customer Agent starts from a strong, on-brand foundation.
Understanding how the model processes information helps you structure your inputs correctly.
If your uploaded content is well structured, the agent becomes far more accurate. If your inputs are messy or conflicting, the agent becomes inconsistent.
This is why we believe a comprehensive, well-structured "Master Knowledge Document" is central to your success.
Below is a practical walkthrough of how to configure the Customer Agent correctly in HubSpot.
Go to Settings > Tools > Customer Agents > Create Customer Agent.
Give it a simple, trustworthy name such as “Growth London Assistant”.
This guides its behaviour. For example:
“Assist visitors by providing accurate information about Growth London’s HubSpot CRM and CMS services. Direct users to relevant content, answer questions using approved knowledge sources, and escalate to a human when necessary.”
Keep it clear and factual.
Upload your Master Knowledge Document. This should be your single source of truth for:
If possible, avoid uploading multiple fragmented documents. Consistency matters.
Define:
This ensures the agent communicates the way your brand does.
Tell the agent:
This drastically reduces hallucinations.
Set up your escalation rules. For example, escalate when users ask about:
Route users to:
We recommend using your HubSpot Meetings link for consistency.
HubSpot allows you to choose whether the Customer Agent can read your site.
We recommend:
Ideally, your Master Knowledge Document remains the primary source.
Test using real-world queries such as:
You should also test leading questions designed to tempt hallucination.
HubSpot’s analytics dashboard shows:
Use these insights to update your knowledge sources and improve performance over time.
One of the biggest differentiators between a good Customer Agent and a great one is the quality of the source.
Keep the information structured
Use clear sections like:
If you say “HubSpot CRM implementation” in one place and “HubSpot onboarding” in another, the AI may treat them as separate concepts.
Avoid:
Aim for clean, direct, factual prose.
Avoid uploading:
The AI will consume anything you give it.
This is vital for reducing hallucination and improving reliability.
Quality inputs lead to quality AI outputs.
Your fallback library is one of your strongest assets.
A few examples:
Fallbacks protect your brand, maintain accuracy, and create a frictionless user experience.
Test in three dimensions.
Include general, service-specific, ICP specific, and technical questions.
Check whether it escalates correctly for:
A Customer Agent should answer confidently when it can and escalate when it should not.
After supporting multiple Customer Agent setups, these are the pitfalls we see most often:
Avoid these, and your Customer Agent will perform well from day one.
Once your Customer Agent is live, review performance weekly at first, then monthly.
Look for:
Continuous refinement keeps the agent accurate and aligned with your brand. Get your colleagues involved, ask them to test the agent and look for gaps or areas for improvement.
HubSpot’s Customer Agent is a powerful tool, but only when implemented with care. The difference between a generic AI chatbot and a high-performing Customer Agent comes down to:
At Growth London, we treat AI assistants the same way we treat CRM architecture. They must be structured, intentional, and aligned with how your business works.
If you would like help configuring your Customer Agent, preparing your knowledge sources, or reviewing your setup, our team would be happy to support you.