FAQ
FAQ is where short, hand-written answers live: one question a customer asks, and the answer you want the assistant to give. It is the third place the assistant gets its material from, and the three do different jobs.
| Store | What belongs in it | How it reaches the assistant |
|---|---|---|
| Knowledge Base | Long documents, and everything synced from your catalog. | Found by search, when the customer's question matches. |
| FAQ | A short fact, written by a person. | Found by search, when the customer's question matches. |
| Guidance | Procedure - what the assistant should do about an intent. | Each rule's name and trigger are in every conversation; its instructions are read when the trigger matches. |
Two things follow from that table, and they are the reason FAQ exists.
Nothing you write here is ever overwritten by a sync. A knowledge base document belongs to the daily Magento sync: it is rewritten whole when its identifier syncs again, and deleting its category deletes it. A FAQ entry belongs to you. No sync run creates, changes or removes one.
An entry costs nothing until it is needed. A FAQ entry reaches the assistant only when a search for the customer's question returns it; in a conversation about something else it adds nothing. A guidance rule is lighter than it used to be - only its name and trigger sit in every conversation, and its instructions are read when the trigger matches - but that list is still there in every conversation, so a fact is cheaper as a FAQ entry than as a rule.
A FAQ entry is not a canned reply, though. Every customer message - including one sent by clicking a suggested question - is answered by the AI model, which searches the knowledge base and FAQ and writes the reply, and that conversation uses tokens as usual. A matching entry makes the answer accurate and the search short; it does not make the message free.
Creating an entry
| Field | Description |
|---|---|
| Question | What a customer asks, in a customer's words. This is what the assistant matches against, so phrase it the way someone would really type it. |
| Answer | What the assistant should tell the customer. |
| Status | Enabled or Disabled. |
The list shows the Question, its Status, its Source, and when it was last Updated. Source reads chat for an entry written from a conversation, and links back to it; an entry you typed here yourself shows nothing. The search box above the list matches both the question and the answer.
Example
Question
Can I collect my order from your warehouse?
Answer
Yes. Warehouse collection is available Monday to Friday, 9am to 4pm. Bring your
order number and photo ID. Your order is ready for collection two hours after you
receive the confirmation email.
Status: only Enabled reaches the assistant
Only entries with the status Enabled are searched. A Disabled entry stays in the list and stays editable - you can work on it, and it reaches no customer until you enable it. Disabling is therefore the safe way to retire an answer that is no longer true, and it is reversible in a way that deleting is not.
How an entry is matched
The assistant searches FAQ the same way it searches the knowledge base, and your entries appear as their own FAQ group in the results, alongside Products, Categories and the rest. Up to three of the best matching entries are offered per search.
Both the question and the answer are matched, and the question counts for more. So an entry whose question is phrased the way customers actually ask is found more reliably than one that hides the same words in a long answer.
An entry can only be found for a question the assistant is willing to engage with. If you ask it something it considers outside the store's business, it declines before it searches anything, and a FAQ entry on that subject is never reached. Entries about your products, orders, delivery, returns and policies are the ones that fire.
Correcting an answer from a conversation
Most FAQ entries are easiest to write at the moment you find the mistake - reading a past conversation in Conversations. Every AI message there has a Correct action that writes an entry without leaving the page, with the customer's own question already filled in. See Correcting an AI answer.
Deleting an entry
Deleting asks for confirmation, and it cannot be undone. The entry is gone, the assistant stops answering from it, and if it was written from a conversation the marker on that message goes with it. When the answer might be right again later - a seasonal policy, a product coming back - Disable it instead.