Context is everything
Outcomes
Three sources, one answer
An AI Employee is only as good as the context it is given, and that context is assembled from the same three places every time. Know them and you can diagnose almost any answer an AI Employee gives. Flip each one:
Instructions set behavior, knowledge supplies facts, and the conversation supplies the moment.
A library, not a script
The Knowledge Base is the reference library your AI Employees search when answering. Four kinds of sources live in it:
- The business profile: hours, location, services, and attributes
- Website content: the pages that already answer customer questions
- Uploaded documents: price lists, FAQs, policies, procedures
- Written entries: knowledge you type in directly, for exactly the questions customers ask
The word "search" is doing precise work here. The AI does not read the entire library for every message. When a question calls for business facts, it searches by meaning rather than by keyword, retrieves the most relevant pieces, and composes the answer from them. A question about "parking" can surface the entry about the lot behind the building even if the word "parking" never appears in it. General conversation may not touch the library at all.
Instructions lead, knowledge informs
The two configured sources have a clear order of authority: instructions lead, knowledge informs. When the two say different things, the instructions win. That gives you a clean placement rule:
Rules and behavior go in instructions. Facts go in knowledge.
"Quote only published prices" is a rule: it belongs in instructions. The price list itself is facts: it belongs in knowledge. Keep each in its place and the two reinforce each other; the employee behaves consistently and quotes accurately. The same logic tells you what stays out of the library entirely: anything the business would not want repeated in a reply. The library is material for answers, so stock it accordingly.
Fresh beats abundant, too. A lean library of current documents outperforms a large one where old and new price lists compete to be retrieved. When information changes, replace it rather than pile on top of it.
See exactly what an answer used
Every AI Employee answer is auditable. Open any response in Conversations and select Explanation to see which knowledge the AI retrieved and referenced for that specific reply. It is the fastest way to tune a library: when an answer misses, the Explanation shows you whether the right knowledge was missing, outdated, or crowded out.
Open one client's website and write two short lists: three questions their customers ask that the site answers, and three it does not. The second list is the start of that client's written knowledge entries, and it took you two minutes.
The data is the moat
Generic AI knows the world; an AI Employee knows one business. The model is the same technology everyone has. What makes an AI Employee valuable is the knowledge behind it: the services, the prices, the policies, the voice. Gathering, cleaning, and maintaining that knowledge is real work, and it is the work that makes the difference.
Knowledge Check
Three quick questions on the three sources, the placement rule, and how retrieval works.