Skip to main content

Command Palette

Search for a command to run...

How to Train Your Instagram AI Agent With Knowledge in ZazzyAgent

Updated
View as Markdown

An AI Agent knows how to understand a conversation.

But it doesn't automatically know your business.

You need to give it the information it should use when answering customers.

That's what Knowledge is for.

System Prompt vs Knowledge

These two areas have different jobs.

System Prompt

Tells the AI:

Who are you?

What is your job?

How should you behave?

What rules must you follow?

Knowledge

Tells the AI:

What does the business actually offer?

What are the prices?

What are the policies?

What are the FAQs?

Think:

System Prompt = behaviour

Knowledge = information

What should you put in Knowledge?

For an Instagram AI Agent, useful information includes:

  • Company information

  • Services

  • Products

  • Pricing

  • FAQs

  • Business hours

  • Locations

  • Delivery information

  • Policies

  • Booking information

  • Contact information

Example

Suppose a customer asks:

Where are you located?

Your System Prompt should not contain a huge list of addresses.

Your Knowledge should contain:

Our Mumbai office is located at...

The AI can then use that information when answering.

Add FAQs

FAQs are one of the easiest ways to train your agent.

For example:

Question

What are your business hours?

Answer

Monday to Saturday, 10 AM to 7 PM.

Another:

Question

Do you offer home service?

Answer

Yes, home service is available in selected areas.

Add your website

Where supported, you can use website/URL-based knowledge.

This is useful when your business information already exists online.

For example:

  • Services page

  • FAQ page

  • Pricing page

  • Booking page

  • Product pages

The current AI knowledge system supports sources such as FAQs, URLs and files, with additional data-source options depending on the setup.

Add documents

You can also train the agent using business documents where supported.

Examples:

  • PDF brochures

  • Product information

  • Policies

  • Service documents

  • Internal FAQs

The important thing is that the document should contain information the AI can actually use.

Keep Knowledge current

If your pricing changes, update the Knowledge.

If your business hours change, update them.

If your cancellation policy changes, update it.

Otherwise the AI may continue using outdated information.

Don't put everything into one huge document

Organize information logically.

For example:

Business FAQs

Pricing

Products

Shipping Policy

Refund Policy

Booking Information

This makes maintenance easier.

Knowledge should answer real customer questions

Don't upload random company information simply because you have it.

Ask:

What will Instagram customers actually ask?

Then build Knowledge around those questions.

Example Instagram FAQ Knowledge

Business Name:
ABC Fitness

Location:
Andheri West, Mumbai

Opening Hours:
Monday-Saturday, 6 AM-10 PM

Services:
Personal Training
Group Classes
Nutrition Coaching

Trial:
One complimentary trial session.

Booking:
Customers can request a trial through Instagram.

Now the AI has useful information for real conversations.

Add boundaries

Knowledge tells the AI what is true.

Your System Prompt should tell it what to do when information is missing.

For example:

Never guess pricing or availability.

If the required information isn't in Knowledge, tell the customer that you don't have the information and offer human assistance.

This is much safer than allowing the AI to invent an answer.

Example

Customer:

Do you have a Sunday appointment?

Knowledge doesn't contain Sunday availability.

The AI should not say:

Yes, 11 AM is available.

It should say something like:

I don't have the current Sunday availability here. I can connect you with the team to confirm it.

Knowledge and lead generation

Knowledge doesn't only answer FAQs.

It can help an AI Agent have better sales conversations.

Customer:

I need WhatsApp automation for my business.

The AI can use Knowledge to explain:

  • Available services

  • Relevant features

  • Typical use cases

  • Next steps

It can then qualify the lead using its configured actions.

Knowledge and booking

A booking agent might use Knowledge for:

  • Services

  • Duration

  • Location

  • Preparation instructions

  • Booking policy

  • Cancellation policy

Then the agent can collect the required booking information.

Test your Knowledge

After training your agent, ask real questions.

For example:

What services do you provide?

How much does it cost?

Where are you located?

What are your opening hours?

Can I book tomorrow?

What's your cancellation policy?

Check whether the answers are accurate.

Check unsupported questions

Also ask something that your Knowledge doesn't contain.

For example:

What is your CEO's favourite food?

The AI should not confidently make up an answer.

Common Knowledge mistakes

Outdated information

Update the source.

Conflicting information

Remove or correct contradictory content.

Missing pricing

Add the actual pricing information if you want the AI to answer pricing questions.

Expecting the AI to know everything

The AI only knows the information made available to it and what it can otherwise infer appropriately.

A useful rule

When an Instagram AI Agent gives bad answers, don't immediately rewrite the System Prompt.

First ask:

Does the agent actually have the correct information?

If not, the problem may be Knowledge rather than behaviour.

Instagram AI Agent
       ↓
System Prompt
   "How to behave"
       +
Knowledge
   "What to know"
       ↓
Customer Question
       ↓
Accurate Answer

A well-written System Prompt without good Knowledge creates a polite AI that may still give poor answers.

Good Knowledge turns it into a useful business assistant.