How to Scale Support and Growth with the Customer Agent in HubSpot

What the Customer Agent actually does

The Customer Agent is an AI agent built into HubSpot that lives on your website and support channels. Any visitor can start a conversation with it — and that's an important point most people miss: it isn't only for existing customers. A prospect sizing you up chats with it exactly the way a customer with a question does. Around the clock, it answers using your own content, points people to the right resource, performs actions like booking a meeting or looking something up, and hands the conversation to a human when it should.

That dual nature is what makes it powerful. It's a support tool that deflects the repetitive questions eating your team's day, and at the same time a 24/7 front door that can capture and qualify leads you'd otherwise never talk to. But it only earns that value if it's set up with care, because it's speaking for your brand in public.

The content behind it is the whole game

Before anything else, understand this: the agent is only as credible as the knowledge you give it. It answers from your existing content — your knowledge base, help articles, FAQs, and website pages. Point it at thin, stale, or missing content and it will confidently tell your customers and prospects the wrong thing, in public, which is far worse than having no agent at all.

So the real preparation isn't clicking through a wizard — it's making sure your content is current, accurate, and actually covers the questions people ask. If you have gaps, the fastest way to close them is usually to get your team talking: record a short call where they walk through the questions they answer every day, then turn that into articles they review and publish. Their knowledge becomes the agent's foundation.

Build the agent and choose how it sounds

Once your content is ready, creating the agent is straightforward. You give it a name and an avatar, choose a personality — Friendly, Professional, Casual, Empathetic, or Witty — or have it speak in your established brand voice, and point it at the content sources it should learn from. You can add specific pages or crawl a domain, using path filters to include the sections that help (like your support pages) and exclude the ones that don't (like your blog).

Because it represents you to the public, the voice decision matters. Using your brand voice keeps the agent consistent with everything else your company says.

Set guidelines and guardrails

Guidelines tell the agent how to communicate: its tone, how it structures answers, reusable scripted responses for common situations, and — critically — its guardrails, the topics it should avoid and how to handle them if a visitor raises them. This is where you set boundaries for an AI that talks to strangers on your behalf: the claims it must never make, the sensitive subjects it should deflect, the way it should treat a frustrated customer. Write these clearly and test them, because guidelines shape the agent's behavior but don't guarantee word-for-word wording.

Decide what it's allowed to do

The agent can go beyond answering and actually perform tasks — booking a meeting, looking up a status, or helping start a routine account task — by connecting to your systems. Each action gets a protection level so the agent verifies who it's talking to before acting on their behalf. For anything sensitive, like account access, the agent should only ever send a secure link or instructions, never make the change inside the conversation. Grant it the minimum it needs and no more.

If you want it working as a lead-generation tool, you can also configure it to qualify prospects who chat in and route the good ones to your sales team.

Make the handoff reliable

An agent that never brings in a human is as dangerous as one that guesses. Decide when a person should take over — an explicit request, a sensitive topic, or simply low confidence — and make sure that handoff routes to a real, available human rather than a dead end. When the agent doesn't know an answer, it should ask the visitor to rephrase or transfer them, never invent a response.

Test thoroughly before anyone sees it

This is the step that protects you, and the good news is that testing is free — it doesn't consume any credits. Preview the agent on the channels you'll use, ask it the questions your customers and prospects actually ask, and deliberately push it on your edge cases and off-limits topics. HubSpot shows you why it responded the way it did, which sources it cited, and where it's weak, so you can coach it until its answers are accurate, on-brand, correctly sourced, and escalating when they should. Only once it's consistently right do you deploy it to live channels.

Budget for credits before you scale

The Customer Agent runs on HubSpot credits, consumed when it resolves a conversation — roughly 50 credits per resolution — and your plan includes a monthly allowance that doesn't roll over. It's worth knowing the resolution rule: a conversation that isn't handed to a human within 72 hours can count as resolved, so a talkative agent that rarely escalates can quietly add up. Estimate your volume up front, and take advantage of the 14-day free-access trial to prove out quality before any credits are spent.

What good configuration gives you

Set up with care, the Customer Agent handles the repetitive questions your team never has time for, answers customers and prospects instantly in your voice, performs real tasks, captures leads, and knows when to bring in a human — with guardrails that protect your brand and your budget. It's not a set-and-forget tool: it rewards keeping your content fresh and reviewing its performance over time. But a well-prepared, well-governed agent scales your support and your lead generation at the same time.

If you want to learn more about how to configure the Customer Agent in HubSpot, contact The Gist.