Message source shows you why the AI replied the way it did. For any AI-generated message, you can see the customer intent the AI detected, a plain-language explanation of its reasoning, and the exact products and content sources it retrieved and used. This helps you spot wrong or incomplete answers quickly and fix the data behind them, so future replies get better.
Message source is available in the Testing Playground (while you test your AI agents) and in Helpdesk (on live bot conversations).
Why use Message source
Understand why the AI gave a particular reply.
See the customer intent the AI detected.
Check which product and content sources were retrieved, and which were actually used.
Troubleshoot unexpected or wrong replies, and fix the data behind them.
Where to find Message source
Where | What you see under each AI message | Best used for |
Testing Playground (Journeys > Testing Playground) | Message source | Correct response | Checking and fixing replies before your AI agent goes live |
Helpdesk (Helpdesk > All bot tickets) | Message source | Response cannot be edited, plus an AI icon next to the message | Understanding replies the AI already sent to real customers |
The selected message is highlighted with a green border in the Testing Playground, so you always know which reply the panel is explaining.
How to view the Message source
In the Testing Playground
Go to Journeys in your BIK dashboard and open the journey you want to test.
Open the Testing Playground.
Click the settings (gear) icon and choose the Channel (for example, WhatsApp). Enter a Name and Email for the test customer. Phone is optional.
Type a customer query in the message box,
Wait for the AI agent to reply, then click Message source below the reply.
The Message source panel opens on the right with the customer intent, AI Reasoning and View sources tabs.
Click Message source under any other reply to see its details. Use the reset icon to start a new test conversation.
In Helpdesk
Go to Helpdesk and open All bot tickets.
Select a conversation handled by the AI.
Click Message source below an AI reply.
The Message source panel opens on the right. Click the collapse icon (») to close it.
Support teams can check any AI reply this way without leaving the conversation.
Understanding the Message source panel
The panel has three parts: Customer intent, AI Reasoning and View sources.
Customer intent
At the top of the panel, Customer intent shows what the AI understood the customer wanted, for example Store information, Order tracking or Product search. If the intent looks wrong, the reply that follows will usually be off too.
AI Reasoning
The AI Reasoning tab answers the question "Why did the AI answer this way?" in plain language.
While the explanation is being prepared, you'll see Thinking... or Working on your answer....
The explanation tells you which source the AI relied on, what it retrieved but did not use, and what to update if the answer is wrong. For example: the AI said the store is online-only because it used the FAQ "Where is your shop located?".
Use the thumbs up and thumbs down icons to rate the explanation.
Under Follow ups, click a suggested question (such as "How do I update an FAQ?") to get the next step.
Type your own question in the Type your question box to dig deeper into the same reply.
AI can make mistakes. Please double-check the explanation before you change your data.
View sources
The View sources tab lists exactly what the AI looked at before replying.
View trace: click it, or the external link icon next to it, to open the full trace of how the reply was generated in a new tab. Use it when you need to dig deeper.
Product sources referred: for product queries, shows how many products were retrieved and used, for example 12 retrieved · 1 used. Expand it to see the products used, the products retrieved but not used, and the Out of stock products behaviour selected for your store, such as "Recommend products but inform buyers that the product is out of stock".
Content sources referred: the knowledge sources the AI searched, such as FAQs, uploaded files and documents, manual links and other configured sources. The summary shows the count, for example 10 retrieved · 1 used. Expand it to see:
Used in response: the sources the reply was actually based on.
Retrieved but not used: sources the AI found but did not use.
External link icon: click it next to a source or summary to open All Data Sources, where you can edit the FAQ, file or link.
Reading the counts
Count | What it means |
Retrieved | Total sources the AI fetched for the question |
Used | Sources the reply was actually based on |
Not used | Sources fetched but left out of the final reply |
For example, 10 retrieved · 1 used · 9 not used means only one of the ten sources shaped the reply. If the source you expected is under Not used, or missing, check and update it.
If a reply was sent by an agent or a saved template (for example, an order selection message), you'll see No sources for this message. This is expected and doesn't mean something is broken.
How to fix a wrong reply
Once Message source shows you why a reply went wrong, you can fix it in one of these ways:
Correct response (Testing Playground): click Correct response under the reply, edit the answer and save it. The corrected answer is saved as an FAQ so similar questions get the improved reply.
Fix response: in the Message source panel, click Fix response below the question box to correct the reply from the panel itself.
Update the source: open the source from View sources using the external link icon. Edit the FAQ, file or link in All Data Sources, then test the same question again in the Testing Playground.
Replies in Helpdesk show Response cannot be edited, because they were already sent to the customer. Fix the source instead, and the next reply will use the updated data.
FAQs
1) What is Message source?
It shows why the AI gave a reply: the customer intent it detected, an explanation of its reasoning, and the products and content sources it retrieved and used.
2) Where can I use Message source?
In the Testing Playground while testing your AI agents, and in Helpdesk under AI replies in bot tickets.
3) What does "10 retrieved · 1 used" mean?
The AI found 10 sources related to the question but based its reply on only 1. Expand the summary to see which source was used and which were retrieved but not used.
4) Why do I see "No sources for this message"?
The message was sent by an agent or a saved template, not generated by the AI, so there are no sources to show.
5) Can I ask follow-up questions about a reply?
Yes. Click a suggested question under Follow ups, or type your own in the Type your question box in the AI Reasoning tab.
6) Can I edit a reply in Helpdesk?
No. Helpdesk replies were already sent, so they show Response cannot be edited. Update the source from View sources, or use Correct response in the Testing Playground.
7) Does fixing a source change future replies?
Yes. Once you update the FAQ, file or link, or save a corrected response, the AI uses the updated data for similar questions.
8) What is the difference between AI Reasoning and View sources?
AI Reasoning explains why the AI replied the way it did. View sources lists the product and content sources it retrieved and used to write the reply.
9) Can I see which products the AI used in a reply?
Yes. Expand Product sources referred in View sources to see the products retrieved, the ones used, and the ones retrieved but not used.
For further assistance or to raise feature requests related to Message source in BIK, please contact the BIK Support Team.