Skip to content
AI

Knowledge base

A knowledge base is the collection of documents, web addresses and frequently asked questions an AI employee consults when it answers a customer.

A knowledge base is where the written information you give an AI employee is kept. Return terms, delivery times, size charts, campaign rules and frequently asked questions sit there. When the employee answers a customer, it looks at these sources first. So the reply comes from the rules your store wrote, not from the employee's general knowledge.

A knowledge base has a limit too: it knows only what you wrote. An old PDF, two documents that contradict each other or a missing returns clause are the most common causes of a wrong answer. So filling it once and walking away is not enough; you test it, fill the gaps and update it when rules change. The chance of error is never zero, but it shrinks with a well-kept knowledge base.

Why It Matters

  • Answers rest on your store's own rules, not on guesses.
  • When a return or delivery rule changes, you update one place, and every conversation uses the same information.
  • You can see which questions the employee is weak on, which shows what to add.

How It Works

  1. Gather the sourcesBring together return terms, delivery information, product descriptions and frequently asked questions. They can be PDFs, documents, web addresses or plain text.
  2. Link it to the employeeConnect the information to the right AI employee. A sales employee and a support employee do not have to hold the same knowledge.
  3. Test with sample questionsAsk the kind of questions a customer would type and read the answers. Add what is wrong or missing to the information.
  4. Keep it currentWhen rules change, change the source too.

Example

Liva Butik uploads its return terms as a PDF. Before buying the Linen shirt (₺1,490), Elif Arslan asks about the return period, and the answer comes straight from that document. Mert K. asks whether order #1042 has shipped. That is not in the PDF, because it changes every day; the employee takes it from the connected store record. A week later Burak Demir changes the return period and updates the document, and the employee starts answering with the new period. What goes in the knowledge base is the rule; what is read from the store is the current state.

Knowledge base and connected store data
Knowledge baseConnected store
ContentReturn terms, size chart, FAQsStock, price, order status
How often it changesRarelyConstantly
Who prepares itYou add documents and textStore data comes through the connection
Sample questionCan the Linen shirt be returned?Is the Linen shirt in size M?

Not to Be Confused With

A knowledge base is not a copy of your store database. Constantly changing facts such as stock, price and order status are not looked up in the knowledge base but in the connected store. The knowledge base holds written rules and explanations.

In Creato

In Creato the knowledge base is linked to AI Employees, and an employee answers only from these sources and approved rules.

  • PDF, DOCX, web address and text sources, along with frequently asked questions, are linked to AI Employees.
  • You link the information to the right AI Employee and test it with sample questions.
  • At setup your website is scanned; working hours, returns, delivery and brand tone are extracted, and none of it is used until you approve it.
  • PDFs with selectable text are understood better than scanned images.
Explore AI Employees

Frequently Asked Questions

Does a knowledge base make the AI answer without errors?

No, it does not bring the chance of error to zero. A knowledge base ties the answer to a source, but the source can be missing, old or contradictory. Test the employee with real customer questions before going live and fix the source of any wrong answer. A rule for handing over to the team on uncertain topics also lowers the risk.

Which file types should I add to the knowledge base?

PDFs, DOCX files, web addresses and plain text all work. If you use a PDF, prefer one whose text can be selected; a scanned image PDF is harder to read. Keep each document to one topic, such as returns or delivery.

Should I write price and stock in the knowledge base?

No. Price and stock change daily, so a value written in the knowledge base is out of date by tomorrow. They are checked from the connected store instead. If there is no connection or no data, the employee does not answer stock as if it were verified.

Related Terms