Quick Answer
An AI knowledge base is the collection of documents an AI uses for the detail of your work: SOPs, templates, guidelines, scripts, decisions and past examples. It doesn't need a new system. It can live in your own SharePoint or Google Drive, as long as your AI has access and the documents are organised so it can find the right one.
What an AI Knowledge Base Is
When a new starter joins, the induction tells them who the company is. The SOPs, templates, guidelines and scripts show them how the work is actually done. That second set is the knowledge base.
For an AI, a useful knowledge base usually holds:
- SOPs: how your team does each recurring job.
- Templates: the current version of each document you produce.
- Guidelines: the rules for pricing, approvals, brand and quality.
- Scripts: what to say on sales calls, demos and support calls.
- Reference material: price lists, rate cards, policies and product details.
- Decisions: what was decided, when and why.
- Lessons: what went wrong or right, so the next job goes better.
- Past work: good examples of proposals, reports and handovers.
A knowledge base isn't the same as company context or a skill. Company context is short and applies to every task: who you are and how you work. A skill is the method for one job. The knowledge base is the detail both of them point to, opened only when a task needs it. Microsoft draws a similar line in Copilot Studio, where knowledge is data an agent can reference and skills are task-specific instructions (Microsoft: Skills overview for Copilot Studio agents).
How AI Tools Read a Knowledge Base
The AI needs its own access. An AI can only read documents it's connected to. Most AI tools offer connectors for the main storage services. Grok, for example, has connectors for SharePoint, OneDrive and Google Workspace (xAI: Connectors). Microsoft 365 Copilot Cowork can attach files and folders from OneDrive, SharePoint and Teams, and edit Office files where they're stored (Microsoft: Use Copilot Cowork).
Reading in place beats uploading copies. Some tools let you upload files into a workspace. Claude projects, for example, have their own knowledge base of uploaded documents (Anthropic: What are projects?). That's useful for one piece of work, but an uploaded copy doesn't change when the original does. Reading documents where they live means the AI sees the current version.
Access follows permissions. Connectors generally work with the signed-in person's own access. If someone can't open a folder, their AI usually can't either. Check what each team can reach before relying on a shared document.
Findability matters as much as access. An AI with access to a messy drive will still pick the wrong price list if three versions sit side by side. Clear folders, clear names and one current version do more for accuracy than any clever instruction.
Organising SharePoint or Google Drive for AI
You don't need new software. A simple structure in the storage you already use works well:
| Folder | What goes in it |
|---|---|
| Incoming | Meeting notes, transcripts and drafts waiting to be processed |
| SOPs | How each recurring job is done |
| Decisions | What was decided, the date and the reason |
| Lessons | What to repeat or avoid next time |
| Templates | The current version of each template |
| Reference | Price lists, policies and product details |
Give each folder an owner. Put dates in file names. Move superseded versions to an archive so the AI can't confuse them with the current one.
Worked Example: From Meeting Notes to a Better Proposal
Kestrel Joinery, a fictional cabinetry business, holds a debrief after each big job. The notes used to sit in someone's inbox.
Now they go into the Incoming folder. The AI processes them into three short documents:
- An SOP update: a new step in the site-measure checklist for checking wall levels.
- A decision: deposits on custom work are now 30%, agreed on 2 September.
- A lesson: allow two weeks for imported hardware, after a delayed kitchen.
The next time someone asks the AI for a proposal for a similar job, the proposal skill points to the Decisions and Lessons folders. The deposit is right and the timeline allows for imported hardware. Nobody had to remember to mention either.
Try It: Set Up a Knowledge Base in Your Storage
- Choose one home: SharePoint or Google Drive.
- Create the six folders above, or your own version of them.
- Move the current templates and price lists in, and archive old versions.
- Rename key files so the name says what they are and when they date from.
- Connect your AI tool to that storage and confirm it can open a test document.
- Point one skill to the specific documents it should use.
- Put one batch of meeting notes in Incoming and ask the AI to turn them into SOPs, decisions and lessons. Review what it writes.
Test It: Three Cases
- A normal request. Ask a question whose answer is in a known document. The AI should give the answer and name the file it came from.
- Incomplete input. Ask about something that isn't in the knowledge base. The AI should say it couldn't find it, not answer from general knowledge as if it were yours.
- An unrelated request. Ask for general industry news. The AI should make clear the answer doesn't come from your documents.
Where Contexta Fits
A skill tells your AI what to do. Your knowledge base shows it how, in detail. Your company context tells it who it's for. The Contexta harness brings them together, in every AI.
Contexta doesn't store or index your documents. Your knowledge base stays in your own SharePoint or Google Drive, and your AI reads it there with its own access. What Contexta adds is method: built-in knowledge base skills, built on best practice, for setting up a knowledge base, turning documents into SOPs, decisions and lessons, and finding what the AI needs for a task.
Company, team and personal spaces each have a home in your storage, with their own rules, so finance work is saved to finance's folders and sales work to sales'. Spaces route work; they don't change who can open which files. Skills point to the documents they need, and every change to a skill or rule is kept in version history. See How It Works for the setup.