Will Customers Hate Talking to an AI Agent? How to Keep Your Brand Voice

An AI support agent can keep your brand voice, as long as the docs and playbooks it works from sound like you. The worry usually comes before that, though: will customers hate talking to it?

What do customers actually mind about AI support?#

Think about the last bad bot you dealt with. It probably answered a question you didn't ask, sent you round the same three options twice, or wouldn't let you talk to anyone.

That's what people mind: the dodging, the looping, the bot that calls itself Sarah and is "just checking that for you", and the chat with no way out. Plenty of customers say they'd rather companies didn't use AI at all: 64% in a Gartner survey of 5,728 people. We think what sours a chat is a bot that was useless and wouldn't admit it.

Where does an AI agent's voice come from?#

From what it reads and the rules you give it. Hay.chat answers customers from your own documents and follows plain-language playbooks you write. So the voice you get back is mostly the voice you put in.

If your returns policy reads like a legal notice, the agent's answer about returns will too. If your help docs are friendly and short, the answers will be friendly and short. Before you blame the AI for sounding stiff, read the docs it's working from.

Write your playbooks the way you talk#

A playbook is a set of instructions for one kind of conversation, like a late delivery or a damaged item. Write it the way you'd brief a new hire on their first day.

Compare these two lines for a late order:

• Inform the customer that the order is delayed and provide the updated delivery estimate.
• Say sorry it's late, give them the new delivery date, and if it's more than a week late tell them a person from the team will follow up.

The second one sounds like a person, and it tells the agent what to do. Playbooks keep a version history you can roll back, so you can try a new wording for a week and undo it if it doesn't land.

Say it's AI, then sound like you#

Replies carry an AI label by default, and the agent is told never to pretend to be a person. On those defaults, nobody should find out halfway through that they've been talking to a machine.

The label does nothing for your voice, though. That comes from everything after it: how you say sorry, how you explain a delay, whether you sign off with a joke or keep it plain.

Hand over before it gets awkward#

An AI that won't let go can ruin a conversation that was going fine. Hay.chat hands the conversation to your team in six cases:

• the customer asks for a person
• a tool fails and needs someone to finish the job by hand
• confidence is low
• an action it said it did didn't go through
• a playbook step says to hand off
• an approval runs out of time

The first one matters most for your brand. When someone asks for a person, the chat goes to your team without them having to argue. In Gartner's 2026 survey of 3,566 customers, 87% said a route to a human is essential, and 50% said GenAI makes their interactions easier. If they're already cross by then, here's how to handle angry customers once they reach your team.

Read the first few weeks of replies#

Reply approval is optional and off by default. Turning it on for the first few weeks means someone on your team reads each reply before it goes out, which is the fastest way to spot where the tone is off.

While you're reading, the conversation page shows a confidence score on every AI message and the source behind each reply. If an answer sounds wrong, you can see which doc or playbook it came from and fix it there, not in the reply. And it answers in the customer's own language, detected per message, so check a few replies in your other markets too.

Upload your help docs, write one playbook the way you'd say it, then read what the agent sends back. The 30-day free trial doesn't need a credit card.

Start your free trial