If every new AI conversation starts with pasting the same product description and policy notes, you have a reusable-context problem. An AI knowledge base gives that material a more deliberate home.
Studio Global's Infobase lets you organize supported source files into a collection and select that collection for AI work. Think of it as maintained reference material, not an automatic guarantee that every answer will be correct.
The best first collection is small, focused, and easy to review.
Choose one job for the collection
Start with a question such as “What material does our team need to draft product FAQs?” rather than “How do we upload everything the company knows?”
For an illustrative product-FAQ collection, you might include the current product overview, an approved onboarding guide, and a current policy document. Keep sales experiments, obsolete policies, and unrelated meeting notes out of the collection.
A collection's name should make its purpose obvious: “Product FAQ — approved sources” is more useful than “Company files.” A name is an organizational aid, not a security boundary. Review your organization's access and data-handling requirements separately.
Prepare the documents before uploading
Make the source material easier for both people and AI to interpret:
- Use meaningful filenames, headings, and clear sections.
- Identify the document owner, effective date, and relevant product version.
- Remove duplicates and superseded copies.
- Prefer readable text over poorly scanned pages.
- Exclude personal, confidential, or licensed material you are not authorized to upload.
Do not assume that a newer upload automatically resolves conflicting policy statements. Remove or clearly separate outdated sources so a reviewer can tell which version is authoritative.
Create and select the Infobase
Open AI Chat, use the attachment controls to open Infobase, create a collection, and add supported files. Check the current interface for file-type and account limits rather than assuming every document is supported.
When starting a relevant task, select the intended Infobase and tell the AI how to use it. The Infobase setup guide covers the product steps.
Try this instruction after adding your approved material:
Use the selected Infobase to answer the question below.
Separate information found in the sources from general suggestions.
For each factual answer, identify the source document and section
when available. Do not invent references or missing policy details.
If the sources conflict, show the conflict. If the answer is absent,
say it is not established by the supplied material.
Question: [specific question]
Requesting source references does not guarantee perfect retrieval or citations. Open the original document and confirm important answers yourself.
Check three kinds of questions
Before using the collection for recurring work, try questions with different expected outcomes. These are suggested acceptance checks, not results from a product test.
| Question type | Example | What a reviewer should check |
|---|---|---|
| Answer present | Which export formats does the approved guide list? | The answer matches the current guide. |
| Answer absent | Does the product offer an undocumented discount? | The response does not invent a policy. |
| Conflicting evidence | Two supplied documents give different deadlines. | The conflict is exposed, not silently resolved. |
If an answer fails a check, improve the source collection or narrow the request. Repeating the same question until you get a reassuring answer is not validation.
Keep facts, style, and instructions separate
Use Infobase for reference material. Use Brand Voice for tone and vocabulary. Use Skills for reusable instructions about how a task should be performed.
This separation helps when something changes. A revised refund policy belongs in the source material; a new writing style belongs in Brand Voice. You should not need to rewrite every prompt to update one product fact.
For a one-time analysis, attaching a file directly may be enough. The file-analysis guide explains that lighter starting point.
Assign an owner and review trigger
Someone should maintain important collections. Review the material when policies change, products launch, or users repeatedly report missing answers. Keep a record of what changed outside the AI conversation if your approval process requires it.
An AI knowledge base is useful because it makes relevant context reusable. Its reliability still depends on source quality, retrieval, instructions, and human review.