What is an AI knowledge base? Definition and examples
What a knowledge base is, what AI changes, how it differs from a wiki or a database, and examples of what teams put in one.
By the Ophiry team · Updated on October 27, 2026
A knowledge base is the place where an organisation keeps what its people need to know, written as clear articles that are easy to search: procedures, policies, product answers, how-tos. An AI knowledge base does two more things: AI writes and updates the articles from the organisation's own documents, and readers ask questions in their own words and get an answer that cites its source. A person still checks what the AI writes before anyone reads it.
What is a knowledge base?
A knowledge base (KB) is a collection of articles that answer the questions a team, its customers or its partners keep asking. Each subject has one article, each article says what applies today, and a search finds it in seconds.
It is not a folder of files. A shared drive holds the documents as they were written: a 40-page handbook, three versions of the same policy, a spreadsheet of rates. A knowledge base holds the answer, once, in plain language. The documents remain the sources behind it.
Most knowledge bases serve one of three audiences:
- Employees (an internal knowledge base): HR policies, procedures, onboarding, IT help.
- Customers (a help centre): how to use a product, what a contract covers, what to do when something goes wrong.
- Clients of a firm (a client portal): the documents and answers that belong to each client, and to no one else.
What is an AI knowledge base?
An AI knowledge base, or AI-powered knowledge base, uses artificial intelligence at the two moments that cost the most effort in a traditional one.
- Writing it. Instead of people turning documents into articles by hand, the AI reads all the documents, notices where they disagree, proposes a plan and writes each article from its sources. People approve the plan and review the result.
- Finding an answer. Instead of typing keywords and opening pages, a reader asks a question in a sentence. The AI looks up the relevant passages and answers in a sentence, with a reference to the article it used. If the knowledge base does not contain the answer, it says so.
The answer comes from the organisation's own content, not from what the AI learned on the internet. That is the difference with asking a general chatbot: the answer can be checked against its source, and it only uses what the reader is allowed to see.
Knowledge base, wiki, database: the differences
The three are often confused because they all store information. They differ in who writes, how you search and what you get back.
Wiki
Corvella Leave Policy
Full-time employees receive 21 days of paid annual leave per calendar year.
Corvella Employee Handbook 2025
1 to 4 years: 21 · 5 to 9 years: 23 · 10 years or more: 25
Two pages come back, written at different times. You open both and work out which one applies.
Database
SELECT days FROM leave_bands WHERE years = '5-9';| days |
|---|
| 23 |
The exact figure, if someone entered it as data and you can write the query.
AI knowledge base
You receive 23 days of annual leave after five years of service (the 5–9 years entitlement is 23 days) [1].
Sources
1. Leave
An answer in one sentence, with the source to check it.
| Wiki | Database | Knowledge base | AI knowledge base | |
|---|---|---|---|---|
| What it holds | Pages anyone can create | Structured records (rows and columns) | Curated articles, one per subject | Curated articles, one per subject, tied to their source documents |
| Who writes | Any contributor | Applications and the people who enter data | A team of editors | AI drafts from the documents, people approve |
| How you search | Keywords, then you open pages | A query in a query language, or a form | Keywords and categories | A question in your own words |
| What you get | A list of pages, sometimes contradicting each other | Exact values, if someone entered them | The article that answers | An answer in a sentence, with its source |
| Typical weakness | Duplicate and outdated pages | Only answers what was structured as data | Slow to write and to keep up to date | Needs a person to check what the AI writes |
Knowledge base vs wiki
A wiki is open: anyone can add a page, so the same subject often ends up on several pages written at different times. A knowledge base is curated: each subject has one article and someone is responsible for it. A wiki suits a team documenting its own work as it goes; a knowledge base suits rules and answers that must be right.
Knowledge base vs database
A database stores facts as data: a leave entitlement is the number 23 in a table. It answers precisely, but only what someone structured, and you need to know how to ask. A knowledge base stores the explanation: who is entitled to 23 days, from when, how to request them. Many organisations need both; the knowledge base is the one people read.
Knowledge base vs RAG
RAG (retrieval-augmented generation) is a technique, not a product: before answering, the AI retrieves passages from a collection of documents and writes its answer from them. An AI knowledge base uses RAG to answer questions, and adds what RAG alone does not have: articles written and reviewed once, one version per subject, and control over who sees what.
How an AI knowledge base works
In Ophiry, building one follows the same steps whatever the subject.
You add the documents as they are: PDF, Word, PowerPoint, Excel, CSV, Markdown, web pages. They are read in the background.
The AI reads the whole set at once and maps it: which subjects come up, which documents disagree, which one replaces an older one. A person checks this map before anything is written. Then the AI proposes a plan, one article per subject, which a person approves or changes.
Structure
Editor in Chief
Tell the editor what to change about the plan.
Add an article on public holidays
Done: "Public holidays" is in the plan, written from the Employee Handbook.
Each article is written from the documents it covers, and every sentence carries a citation mark that opens its source. A person reviews the articles and publishes them.
Annual leave
A full-time employee receives 21 days of annual leave, 23 days from 5 to 9 years of service and 25 days after that.1 An absence of up to five days is requested in the HR portal at least two weeks in advance.2 Up to five unused days can be carried over, and must be taken before 31 March.3
Readers then ask their questions in their own words, and each answer cites the article it comes from. Each reader sees only the content shared with them.
Ophiry is a software that turns an organisation's documents into a clear knowledge base, written by AI and checked by people, that everyone can ask questions to and get answers with their sources. Each reader sees only what is meant for them.
Examples of AI knowledge bases
| Example | Built from | Readers ask, for example |
|---|---|---|
| HR & employee handbook | Handbook, leave and travel policies, benefits | "How many vacation days do I get in my first year?" |
| Compliance & procedures | Procedures, policies, regulations | "Who must approve a new supplier?" |
| Customer support | Product manuals, guides, terms | "The app says 'sync failed'. What should the customer try?" |
| Client portal for financial firms | Each client's own documents | "What are the management fees on my fund?" |
| Client portals for consulting firms | Each engagement's own documents | "What is in scope for phase 2?" |
The limits, honestly
- AI can misread a document. A table read the wrong way, or two rules merged that should stay apart. That is why a person checks the map before writing and reviews the articles before they are published, and why every sentence points to its source.
- It is only as good as its documents. If no document says it, the knowledge base cannot answer, and it should say so rather than guess.
- It still needs an owner. When a policy changes, someone adds the new document and approves the update. Without an owner, any knowledge base goes out of date.
How to create one
Start from the documents you already have, not from a blank page. The full method, step by step, is in How to create a knowledge base from your documents. You can try it on your own files with the free trial: no card is required.