Document management for customer support AI assistants

Walk through how to manage knowledge sources for your newly-built customer support AI assistant.

Tags Streamline Icon:
Wrench Streamline Icon:
People Man Graduate Streamline Icon:
Paragraph Bullets Streamline Icon:
Included in the course:
Implementing AI in customer support

In this tutorial, we’ll walk through how to manage knowledge sources for your customer support AI assistant, including identifying gaps, generating new content for your documentation, and updating support articles.

This is an important part of a customer support AI assistant workflow as it’ll help ensure that your documentation is always up-to-date and comprehensive. You can think of this as a recursive feedback loop, making your customer support AI assistant more capable and smarter on your business the more you use it.

You’ll need:


  1. Identifying knowledge gaps
  2. Filling knowledge gaps and updating knowledge sources
Products Give Gift 1 Streamline Icon:

Try before you bite?
This one's on us.

Sign up for a free account to view our free courses

Thank you! Your submission has been received!
Oops! Something went wrong while submitting the form.

Step 1: Identifying knowledge gaps

There are a few ways you can identify knowledge gaps in your customer support documentation with a customer support AI assistant in ChatGPT — we’ll outline two of them:

  1. In context to agent questions: When a human agent asks a question to your customer support AI assistant, and there’s no supporting content in your knowledge sources.
  2. Explicitly via requests: Explicitly ask the GPT if there are any major knowledge gaps in your support documents based on generally accepted customer support norms.

We’ll deep dive into how to achieve both with your GPT.

#1. In context to agent questions

For this method, it's best to add a sentence or two to your custom Instructions to make sure that your GPT manages knowledge gaps clearly and effectively. Support agents won’t often be able to tell the difference between a hallucination, a norm from ChatGPT’s training data, or your actual documentation.

To make this clearer for agents working with your customer support AI assistant, we’ll add a new snippet to our custom GPT Instructions. Navigate back to the Configure tab of your CustomGPT to add this additional instruction.

Additional Instruction snippet:

If there isn't supporting data from the Knowledge sources, state this documentation gap upfront in your response, and provide a general account management best practice instead.

This does two things:

  1. Makes it clear when a question does not have supporting documentation and;
  2. Provides a best practice in its place.

Below is the full set of custom Instructions, with this new addition, for reference.

Sample full custom Instructions:

#Role: Support Assistant is tailored to assist customer support agents by providing answers or drafting sample responses based on uploaded documentation, including brand guidelines, FAQs, and past conversations. The support agents will then use these answers or draft responses to respond to customers.

#Tone: It balances a semi-formal tone, ensuring responses are professional yet approachable. Key focuses are on using a friendly tone, being concise, showing empathy, and maintaining brand consistency.

#Objective: The GPT is designed to offer the best initial response based on the available information and follow up with proactive questions if more details are needed.

#Additional Instruction: When providing answers to questions or drafting responses, always provide a reference to the file name and location within the file from the Knowledge source that provided the information by explicitly referencing the source file and information location as part of your response. You must always provide a clear reference to the source file in all responses. If there isn't supporting data from the Knowledge sources, state this documentation gap upfront in your response, and provide a general account management best practice instead.

Now, test your GPT with a question that you know is not in your knowledge source.

#2. Explicitly via requests

In addition to flagging knowledge gaps from support queries, you can also directly ask your GPT if there are any known knowledge gaps in the support documentation based on customer support best practices.

We recommend doing this via a prompt in a chat window with your GPT vs. developing additional custom instructions for this step.

Sample prompt:

Are there any known knowledge gaps in the uploaded files of your knowledge source, based on customer support best practices in the [insert industry] industry?

Step 2: Filling knowledge gaps and updating knowledge sources

Now that you’ve started to identify knowledge gaps, you can start to have your GPT generate new knowledge based on best industry practices to fill these gaps. You should think of these generations as boilerplate starting places for future codified policies.

Let’s work off the previous example and ask our GPT to draft sample guidelines for an escalation procedure.

Sample prompt:

Can you draft sample guidelines for Escalation Procedures for our customer support staff, based on what you know from our business, your knowledge sources, and best practice for [insert industry] companies?

Now that we have a draft of this new procedure, we should update our documentation accordingly. We can ask our GPT where the best place to add this additional knowledge is within our documentation.

Sample prompt:

Where should I add this new procedure in our existing documentation? If it doesn't make sense to add it into an existing source, we can start a new document.

Now, you can take this information and either update your existing documentation or create new documentation. With either option, make sure to upload (or re-upload) this documentation as a knowledge source to your GPT.

Get full access

✔️ All 100+ courses & tutorials in our catalog
✔️ New content added weekly
✔️ Private community access
✔️ No subscription, $150 paid once
✔️ Expense it using this template. Or get a team account.
✔️ 30-day refund policy. No questions asked
Join 3,107 learners from companies like Microsoft, Coca Cola, NBA, Adobe & Google

More tutorials like this

View all

If you scrolled this far, you must be a little interested...

Start learning ->

Join 3,107 professionals already learning