⏱️ The Essentials in 3 Minutes |
🧠 Understand the Pedagogical Value of Document Structure
Didask AI does not "read" a document the way a human does: it compares each question against your entire knowledge base and selects the passages closest to the learner's need. A poorly structured, too long, too vague, or redundant document will be hard to retrieve, even if it contains the right answer.
Ensuring your documents are well-structured allows the AI to pinpoint exactly what each collaborator needs at the right moment. It also guarantees that your business knowledge is transmitted faithfully, without noise or ambiguity.
✅ Evaluate a Document Before Adding It to the Base
Before any import, ask yourself these four questions:
Question | If the answer is no... |
Can I summarize this document in a short, precise title? | The document probably covers too many topics: split it. |
Is all the information still valid? | Clean up the document or do not include it. |
Are there other documents in the base that say the same thing? | Remove the overlapping parts or merge the documents. |
Is this document useful to all users who have access to this base? | Restrict access or separate into distinct bases. |
📌 Follow the Rule: One Document, One Topic
A document that covers multiple topics at once will be hard to associate with a specific question. The AI will struggle to identify it as the most relevant source, even if it contains the right information.
The tell-tale sign: you have trouble giving it a short, precise title.
Good example: "Client quote validation process" - explains only the validation steps, the stakeholders, and the timelines.
To avoid: "Sales team meeting - March 12" - covers quote validation, an HR update, Q3 strategy, and a bug post-mortem all at once.
What to do: split the document into as many documents as there are distinct topics. Each document should have a title that summarizes its content in one short sentence.
🧹 Remove Outdated Information
The AI cannot distinguish what is still valid from what is outdated. Old information can make the right information harder to find, or even cause incorrect responses.
Good example: an up-to-date expense reimbursement policy, containing only the rules currently in force.
To avoid: the same document with new rules added at the bottom, without removing the old ones. The AI cannot tell which ones are valid.
What to do: extract only the still-valid parts and create a clean new document. If a document is too cluttered to clean up easily, it is better not to include it. Fewer documents of higher quality always yields better performance.
🔁 Avoid Redundancies Between Documents
If multiple documents cover the same topic in different words, the AI will surface them simultaneously. These duplicates take up space that could be filled with complementary knowledge, and the overlapping parts mask the specific content of each document.
Good example: two documents on the same tool - one titled "Technical guide to tool X for developers", the other "Using tool X day-to-day for business teams". Each goes straight to what is specific to its audience, without a shared introduction.
To avoid: three versions of documentation on the same feature (marketing, support, developers) that all start with the same two paragraphs of general overview.
What to do: remove the overlapping parts, or isolate them in a dedicated document that you reference. Clearly distinguish documents in their title ("technical version" or "simplified version").
🔒 Adapt Access Rights to Users
Not all learners need all the knowledge. Highly technical documents intended for a small group can interfere with responses for other users whose questions use similar words.
Good example: technical documentation on the product architecture is accessible only to R&D teams. Sales teams have access to a separate base with product sheets and sales pitches.
To avoid: a base accessible to everyone that mixes client call transcripts, internal technical specifications, and onboarding guides.
What to do: define user groups and assign documents accordingly. Build the base progressively: add a document when a group expresses a real need, not in anticipation.
Keywords: document structure, knowledge base, Didask AI, document quality, performance, redundancies, access rights.