Short answer: yes, you can use ChatGPT for customer service — but not by pointing the chatgpt.com website at your customers. Out of the box, ChatGPT is a general assistant with no knowledge of your business, no connection to your systems, and no way to hand a frustrated customer to a human. Used properly, it works in three forms: a custom GPT for internal or low-stakes use, an API-powered bot wired into your website, or the AI features built into your existing helpdesk. Costs run from about $30 a month to a few thousand for a proper setup — and the businesses that get burnt are the ones that skip the guardrails.
This guide walks through what ChatGPT genuinely can and can’t do for customer service, the three realistic setup options with honest pricing, the pitfalls that catch Australian SMBs out, and when it’s time to graduate to a purpose-built chatbot.
What ChatGPT can and can’t do out of the box
Let’s be straight about the raw product first, because a lot of the confusion starts here.
What it can do well:
- Draft replies to customer emails and messages in seconds, which your staff then review and send
- Summarise long complaint threads so a team member can pick up context fast
- Rephrase awkward or blunt responses into something more diplomatic
- Answer genuinely general questions (“how do I reset my router?”) in plain language
What it can’t do without extra work:
- Know anything about your business. Ask the vanilla product about your returns policy and it will invent one — confidently, politely, and completely wrongly.
- Talk to your systems. It can’t check an order status, look up a booking, or create a support ticket on its own.
- Remember a customer between sessions, or escalate to a human when it’s out of its depth.
- Guarantee any particular answer twice. The same question can get two different replies.
None of these are reasons to avoid it. They’re reasons to choose the right setup option rather than the naive one — which is what the next section is for.
Three ways to actually set it up
Option 1: A custom GPT
OpenAI lets paying users build “custom GPTs” — ChatGPT with standing instructions and uploaded reference documents. Setup is genuinely quick: write instructions (“you answer questions for a Melbourne plumbing company — be friendly, never quote prices, always offer a booking”), upload your FAQ and policy documents, and you’re done in an afternoon.
The catch is access. Custom GPTs live inside ChatGPT itself, so whoever uses it needs a ChatGPT account. That makes them a poor fit for a public website widget, but a very good fit for internal use: a “policy bot” your own staff query, or a drafting assistant for your inbox. Think of it as a fast junior team member, not a front-of-house rep.
Cost: roughly US$20–$30/month per user for the ChatGPT subscription that unlocks GPT building (approximate — OpenAI’s pricing moves around). No build fee if you do it yourself.
Option 2: An API-powered bot on your website
This is what most people actually mean when they say “ChatGPT for customer service”. Instead of the ChatGPT app, you use the OpenAI API — the same underlying models, accessed programmatically — behind a chat widget on your own website. No ChatGPT accounts for customers, your branding, your rules.
This is also where the real work lives. The API gives you a brain; you still have to build everything around it: the chat interface, the instructions and knowledge it answers from, the rules for when it stops answering, and the handover to your team. Done well, this resolves the majority of repetitive enquiries around the clock. Done hastily, it’s a confident liar on your homepage.
Cost: API usage itself is cheap for SMB volumes — typically tens of dollars a month for a few thousand conversations, though it scales with traffic. The real cost is the build, which is where the price range widens enormously (see the table below).
Option 3: The AI inside your existing helpdesk
If you already run Zendesk, Intercom, Freshdesk or similar, they all ship AI features now — answer bots, suggested replies, auto-triage. The upside is integration: the AI already sees your tickets and help centre. The downside is pricing and lock-in: AI features usually sit on higher tiers or charge per resolution, and the bot only knows what’s in that vendor’s system.
Cost: very roughly $30–$150/month per agent on top of your existing plan, sometimes plus per-resolution fees. Check current vendor pricing — it changes often enough that any figure here is an estimate, not a quote.
Cost comparison at a glance
| Option | Typical cost | Setup effort | Best for | Watch out for |
|---|---|---|---|---|
| Custom GPT (internal use) | ~US$20–$30/user/month | An afternoon DIY | Staff drafting replies, internal policy questions | Not public-facing; needs ChatGPT accounts |
| Helpdesk AI add-on | ~$30–$150/agent/month, sometimes per-resolution | Toggle on, tune help centre | Teams already on Zendesk/Intercom/Freshdesk | Vendor lock-in; per-resolution fees add up |
| API bot, professionally configured | $300–$1,000 one-off + modest monthly API/hosting | 2–7 days with help | SMBs wanting a website bot live this week | Vendor platform limits underneath |
| API bot, custom built | From $6,500 fixed + API usage | 3–5 weeks | Real volume, integrations, proper handover | Bigger upfront — but you own it |
The honest pattern: the monthly fees are rarely the problem. The setup quality is. A $30/month tool configured badly costs you more in lost customers than a $6,500 build that works.
The pitfalls that catch businesses out
Most ChatGPT customer service failures come from the same three mistakes.
Hallucinated policies. This is the big one. A general model asked about your warranty, refund terms or pricing will often answer with something plausible rather than something true. One invented “yes, we’ll refund that” screenshot can cost you real money. The fix isn’t hoping the model behaves — it’s grounding it in your actual documents and instructing it to say “I don’t know” when the answer isn’t there. Grounded, cited answers are the whole point of a properly built knowledge system, and they’re non-negotiable for anything customer-facing.
No human handover. ChatGPT has no idea when it’s failing. Without an explicit escape hatch, customers get trapped in polite loops while getting angrier — and you never find out. Every deployment needs a “talk to a person” path that’s always visible, triggers on signs of frustration or repeated failure, and passes the full transcript to your team so the customer doesn’t have to repeat themselves.
Privacy and data handling. Customer conversations contain personal information, and Australian businesses have obligations under the Privacy Act about how that’s collected, stored and disclosed. Things to check before you go live: where conversation data is processed and stored, whether it’s used to train models (API access can be configured not to; consumer ChatGPT is a different story — don’t paste customer details into the free product), and what your privacy policy says about AI-handled enquiries. This isn’t a reason to avoid AI support; it’s a 30-minute checklist you do once.
There’s a fourth, quieter pitfall: set-and-forget. Your prices, policies and stock change. A bot built in January is confidently quoting October’s catalogue by March unless someone owns keeping its knowledge current.
When to graduate to a proper custom build
The lighter options are genuinely fine for a while. Here’s how you know you’ve outgrown them:
- Volume. You’re handling hundreds of conversations a month and per-seat or per-resolution fees are stacking up to real money.
- Integrations. Customers keep asking things that require live data — order status, booking availability, account details — and copy-paste answers aren’t cutting it.
- Channels. You need the same bot on your website, WhatsApp and Messenger, sharing one brain.
- Compliance or brand risk. Wrong answers are starting to cost you, and “mostly right” is no longer acceptable.
At that point a custom AI chatbot build stops being a luxury. Ours is a fixed fee from $6,500, delivered in 3–5 weeks: conversation design, the right engine for your case (not whichever we happen to prefer), deployment to your channels, human handover with full transcript, analytics, and handover documentation so your team can run it. No hourly billing, no per-seat licence, and if you’ve already done a lighter setup, the intents and FAQs carry over — you don’t pay twice for the thinking.
Not there yet? A professionally configured chatbot setup from $300 gets an off-the-shelf or API-powered bot live in 2–7 days with the guardrails done properly — grounding, fallback answers and the human escape hatch — which is where most SMBs should start.
Which option are you?
- Just want to speed up your inbox? Start with a custom GPT for internal drafting. An afternoon, ~US$20–$30/month, zero risk to customers.
- Want a website bot live this month, on a budget? Get one configured properly — from $300, done in days — with grounding and handover built in from day one.
- Real volume, multiple channels, or systems to integrate? That’s custom-build territory — fixed fee from $6,500, live in 3–5 weeks.
If you’re genuinely unsure which row you’re in, that’s exactly what our free 20-minute scoping call is for — we’ll look at your enquiry volume and systems, tell you which option fits, and if the honest answer is “you don’t need this yet”, you’ll hear that too.