I don’t know about you, but the idea of having customer service chatbots that are actually helpful sounds amazing! It’s a little easier said than done, as Air Canada could tell you after its bot gave out wrong information about tickets that they then had to pay for. A customer assistant chatbot that makes stuff up is not helpful, for anyone.

So, if this is something you’re in any way involved in, I have two suggestions.

1. Don’t let your chatbot wander the web – only give it access to specific, true sources of content about your products, services, news, and policies.
2. Make sure those content sources are as clear, thorough, and up to date as you can get them. Quality data is still your most valuable asset.

When you do have something set up though, something that can genuinely chat and helpfully respond to different ways people have of describing issues, then you have a tool your customers could truly value. They get their problem solved, it’s under their control, at a time convenient for them, in a place convenient for them.

Plus, you have a tool that can summarise and analyse customer service conversations to find information gaps, areas of confusion, and more – leading to better quality input data (provided you act on the feedback).

As with many other areas, these tools are going to see service call centres, especially outsourced ones, shrink drastically. The routine issues, pricing questions, and simple transactions (like basic renewals) can all be done via AI interfaces. The people that used to look after this, are no longer needed on the phones (or messaging apps). The people who ARE needed are the ones who can do more, understand deeper complexities, and, provide human connection and reassurance to people who need that kind of support. There won’t be as many of them, but they’re going to be important.

This is as true for internal customers as external, as various support helpdesks – IT, HR, Finance, etc – move to AI processes that can automate processes such as password resets, expense claims, and time-off requests. And even for those tickets that do need a human expert to step in, the tools can act as an assistant, summarising an issue’s history, providing reference information, and suggesting solutions or responses.

The end result is faster outcomes, happier end users, and reduced costs – an irresistible brew for any organisation, especially those with money as a key concern or motivation.

The challenge is when things go too far (as they too often do), and customers are left with no one capable of supporting them on those edge-case issues and situations, or those customers that need more than the bare minimum in the way of assistance and moral support are left out in the cold. Especially if it’s done too fast, with more focus on cost reduction than quality of service.

And, of course, the more specialised and focused the product or service is, the harder it can be for an AI tool to be effective in supporting it. That’s where AI-skilled customer support and service agents (human ones) are still needed, providing a trusted point of contact into expert advice and attention.

Using AI Tools in Customer Service and Support

As with all LLM prompts, the more context and direction you can provide in your prompt (and conversation), the better your outputs will be.

Complaint and Enquiry Emails

Use case: Draft responses to difficult or time-consuming customer emails.

Example prompt: I run a small [type of business]. A customer has sent this complaint: [paste email]. Please draft a professional, empathetic response that acknowledges the issue, explains what I can do to resolve it, and maintains goodwill. Use the [attached document/linked website] for product and service reference information if needed. Keep it under 150 words.

Make sure you review it before you send it! This is exactly the moment you don’t want an LLM getting creative.

Use case: Summarise and analyse your archive of support email conversations.

Example prompt: Here are [X] customer service conversations. In bullet points, list the five most common problems customers raised, and for each one note whether it was typically resolved well or poorly based on how the conversation ended.

Or, a bigger one: I’m going to share an archive of customer service conversations (emails/chat transcripts). Please analyse them and produce a structured report covering:
Quick wins: what are 3–5 specific, actionable changes that could improve customer satisfaction or reduce contact volume? Please present your findings clearly with the following headings, and where possible give brief example quotes from the conversations to illustrate each point.
Main themes and issues: what are customers most commonly contacting us about? Group into categories and show rough frequency.
Complaint and problem patterns: what are the most frequent complaints or pain points? Are any issues recurring or escalating?
What worked well: based on the conversations, which types of response or resolution approach led to the most positive outcomes? Look for patterns in tone, speed of resolution, solution offered, and any language that seemed to de-escalate or satisfy customers.
What needs improving: where do conversations go badly? Look for unresolved issues, repeated contacts about the same problem, frustrated or escalating customers, and any gaps in knowledge or process.
Here are the conversations: [paste or upload – make sure you remove all PII such as email addresses first]

SEO content for support pages

Support content and SEO work really well together, FAQ pages, help articles, and troubleshooting guides are what people search for. (For the record, FAQs etc are also great for email newsletter content).

Example prompt: Generate 10 frequently asked questions and answers for a [product/service] page on my website (the more context and info you can give here, the better). Each answer should be 50–80 words, written in plain English, and optimised for search. Include natural variations of the question that people might type into Google.

Note: if you want to take it a step further and think about visibility into LLM searches, add verifiable stats, expert quotes (especially if they’re ‘your’ stats and experts), and keep your information fresh.

Building a basic self-service chatbot

If you’re willing to go with an LLM (specifically ChatGPT) as your prototype chatbot, you can make a pretty decent start on automating your more mundane query responses. You will need the paid version though.

Use case: Creating a self-service response bot using a public custom GPT (Claude Projects and Gemini Gems aren’t shareable). Use something like the prompt below (make sure you test it!), then add a link to the Custom GPT (open in a new tab) for simple enquiries. Just make sure there’s an email address or other contact option as well.

Example prompt: You are a support assistant for [business name]. Answer questions using only the information below. If a question isn’t covered, say: ‘I’m afraid that question’s beyond my abilities. Could you please send us a message at [email]?’ Never make up information. Always be friendly and concise.
[Paste your product info, FAQs, pricing, policies etc. – again, make sure these are always up to date]

Another tool option is Microsoft Copilot Studio and using a bit of embed code to add it directly to your website.

Email template drafts

Use case: Write or improve a response template library for emails, chats, or social media DMs.

Example prompt: I need 8 standard responses for my most common customer service situation emails: order delays, refund requests, product questions I can’t answer immediately, positive feedback, wrong item received, cancellation requests, chasing a quote, and general enquiry acknowledgements. Keep each one under 60 words, warm in tone, and easy to personalise with [NAME] and [SPECIFIC DETAIL] placeholders.