How to Automate Customer Service With an AI Agent (With the Exact Prompt)
Customer service is the workflow that quietly scales with your business whether you want it to or not. More customers means more "where's my order?", more refund requests, more password resets, and more of the same handful of questions answered by hand, one message at a time. The volume is repetitive, the hours are unforgiving, and hiring your way out of it gets expensive fast.
This guide shows you how to automate customer service with an AI agent that lives in the chat bubble on your website. It answers the common questions instantly, and for everything else it captures the request, logs it to a Google Sheet, ranks it by urgency, and drafts a recommended solution — so your team opens a clean, prioritised queue instead of a chaotic inbox. If you're new to the idea, our primer on what AI agents actually are is a good place to start.
TL;DR
- An AI agent sits in your website chat bubble and handles incoming customer requests 24/7
- It answers common, low-risk questions instantly using a model like Claude
- Every request is categorised and logged to a Google Sheet support queue
- Each one is ranked 1–5 by urgency using impact, sentiment, and customer value
- The agent drafts a recommended solution for each ticket so a human can approve and send in seconds
- The full agent prompt is included below — ready to copy
The Problem With Manual Customer Support
The pain of manual support isn't any single ticket — it's the aggregate. Studies of support teams consistently find that the large majority of incoming tickets are repetitive questions with known answers: order status, returns, billing, resets, hours. Your team retypes those answers all day, which means the genuinely hard problems — the frustrated customer about to churn — wait in the same undifferentiated queue as "what are your opening hours?".
Two things break at once. First, response time: when everything is first-in-first-out, urgent issues sit behind trivial ones. Second, prioritisation: a human skim-reading an inbox has no consistent way to rank by importance, so triage is guesswork that changes with whoever is on shift. The result is slow replies on the tickets that matter most and burned-out agents doing copy-paste work on the ones that don't.
The usual "fix" — a rigid rules-based chatbot with buttons and decision trees — frustrates customers because it can't understand a question phrased in a way the script didn't anticipate. An AI agent is different: it reads the message the way a person would, understands intent and tone, answers what it can, and takes the action of logging, ranking, and drafting a resolution for everything else. For a broader look at how this applies across a business, see our guide on AI agents for small business.
How the Customer Support Agent Works
The agent behind Talk to Me Data's customer support automation follows a simple loop every time a message lands in the chat bubble: understand it, resolve or capture it, log and rank it, and draft a recommended solution. Here's each step.
Step 1 — Understand & categorise
A customer types into the chat bubble on your site. The agent reads the message, sorts it into a category (order status, returns & refunds, billing, technical, product question, or other), and reads the sentiment — calm, frustrated, or angry. This is what a rules-based bot can't do: it understands a question no matter how it's phrased.
Step 2 — Resolve or capture
If it's a common question with a known answer, the agent responds instantly and warmly from your knowledge base — no waiting, any hour of the day. If it needs a human, account access, or an action the agent shouldn't take on its own, it collects the details the team will need (order number, email, a one-line summary) so the customer never has to repeat themselves.
Step 3 — Log & rank in Google Sheets
The agent appends a row to your support queue in Google Sheets: timestamp, email, category, one-line summary, full message, sentiment, and a priority score from 1 (low) to 5 (urgent). Urgency is scored on impact, sentiment, and customer value together — so an angry paying customer who can't log in floats to the top and a general product question settles near the bottom.
Step 4 — Draft a recommended solution
For every logged request, the agent writes a recommended resolution into the sheet: the likely cause, a ready-to-send reply, and any action to take (refund, replacement, escalation). Your team opens a queue that's already sorted by urgency and pre-solved — approve, tweak, send. Minutes of thinking per ticket become seconds of reviewing.
Why the Chat Bubble Is the Right Place to Start
You could point this agent at email or a helpdesk like Zendesk or Intercom — and eventually you probably will. But the chat bubble is the highest-leverage place to begin, because it's where intent is freshest: the customer is on your site, in the moment, with a specific question. Answering there deflects the ticket before it ever becomes an email, and captures the ones that do need a human while the context is still rich.
The best part is that the underlying logic doesn't change when you add channels. Whether the message arrives via chat, email, or WhatsApp, the agent still answers, logs, ranks, and recommends into the same Google Sheet. You build the brain once and plug in new channels over time.
The Prompt
Here's the exact prompt behind the agent. Paste it into your AI agent orchestration interface — whether that's Talk to Me Data or a Claude project with your chat widget and Google Sheets connected. Swap [Company] for your business name and point sheets_append_row at your support-queue sheet.
You are the front-line customer support agent in the chat bubble on [Company]'s website. You answer common questions instantly, and for anything you can't fully resolve you log a clean, ranked ticket with a recommended solution for the human team. STEP 1 - UNDERSTAND & CATEGORISE Read the customer's message. Sort it into one category: Order status, Returns & refunds, Billing, Technical issue, Product question, or Other. Detect the sentiment (calm, frustrated, angry) and note whether this looks like a high-value or at-risk customer if that's knowable. STEP 2 - RESOLVE OR CAPTURE If it's a common question you can answer from the knowledge base, answer it directly, warmly, and concisely. If it needs a human, account access, or an action you can't take, collect the details the team will need (order number, email, a one-line summary) so nobody has to ask the customer twice. STEP 3 - LOG TO THE SHEET Call sheets_append_row to add the request to the support queue with: timestamp, customer email, category, one-line summary, the full message, sentiment, and a priority score from 1 (low) to 5 (urgent). Score urgency on impact + sentiment + customer value together: an angry paying customer who can't log in is a 5; a general product question is a 1-2. STEP 4 - RECOMMEND A SOLUTION For every logged request, draft a recommended resolution for the human agent: the likely cause, a ready-to-send reply, and any action to take (refund, replacement, escalation). Put it in the "Recommended solution" column so a human can approve and send in seconds instead of starting from scratch. STEP 5 - CONFIRM Tell the customer what happens next and the expected timeframe. Be warm and concise, and never promise something you can't verify.
What You Need to Set It Up
A chat widget on your site
The bubble your customers type into. It's the front door of the agent — the place requests come in and instant answers go out. The same agent can later listen on email, WhatsApp, or a helpdesk, but the website chat is where you get the fastest wins.
A capable model with your knowledge base
A model like Claude with tool use enabled, grounded in your FAQs, policies, and product docs. This is what lets it understand messy, real-world questions and answer them accurately instead of guessing.
A Google Sheets connection
A link to Google Sheets so the agent can append each request to your ranked support queue. This is the shared source of truth your whole team works from — the sheets_append_row action the prompt calls.
Don't want to wire up the widget, model, and integrations yourself?
Talk to Me Data builds, connects, and hosts this agent for you — the chat widget, the model, the Google Sheets (or helpdesk) integration, and the monitoring. Nothing to configure or maintain on your side.
See the customer support agent →What You Get Out of It
| Before (manual) | After (AI agent) |
|---|---|
| Answer the same questions by hand, all day | Common questions answered instantly, 24/7 |
| Tickets worked first-in-first-out | Queue ranked 1–5 by urgency automatically |
| Triage is guesswork, varies by who's on shift | Consistent scoring on impact + sentiment + value |
| Every reply written from scratch | Recommended solution drafted for each ticket |
| Requests scattered across inbox and DMs | One clean, shared queue in Google Sheets |
See it for yourself
Watch the support agent answer, rank, and resolve in real time.
Prefer to see it before you build anything? Take a look at the customer support agent handling live requests — or book a free call and we'll build it around your stack and your knowledge base.
How to automate customer service without losing the human touch
The goal isn't to remove people from support — it's to remove the repetitive volume so your people can do the part only they can do. The AI agent absorbs the "where's my order" and "how do I reset my password" tickets, and for everything harder it hands your team a ranked queue with a recommended solution already drafted. Customers get faster answers, your team gets its time back, and the hardest cases get the human attention they deserve. Want to put a number on the hours you'd reclaim? Try our workflow time savings calculator.
See the agent
Learn More
See exactly how the AI customer support agent works, what it connects to, and how we build it around your knowledge base.
Ready to go
Get Started
Tell us your channels and knowledge base and we'll build, connect, and host your support agent — live in days.
Frequently Asked Questions
Summary
Automating customer service used to mean a rigid chatbot that annoyed customers and a support inbox no one could prioritise. An AI agent changes both. It lives in your chat bubble, answers the common questions instantly, and for everything else it logs the request to Google Sheets, ranks it by urgency, and drafts a recommended solution — turning a chaotic inbox into a clean, prioritised, pre-solved queue.
The prompt above is ready to use. Connect a capable model, your chat widget, and a Google Sheet, drop the prompt in, and start deflecting and triaging support automatically. If you'd rather skip the setup entirely, we build, connect, and host the whole thing for you.
Want the customer support agent built for your business?
We build, deploy, and host AI agents connected to your chat widget, Google Sheets, and helpdesk tools like Zendesk and Intercom. Book a free 20-minute call, or start building yours now.