Answer Source Verification helps ensure that AI answers are grounded in your company’s content. When it’s enabled, the system checks answers against your documents and may adjust them before they’re finalized.
You can control this in General Chat → Advanced → Answer source verification.
What users see in the chat
When verification is enabled, answers appear in two stages:
Draft answer (grey text)
You’ll first see the AI’s initial answer streaming in grey.
This is the unverified draft.
Verified answer (black text)
You’ll see this message as the system analyzes the answer to assess if a redo is needed
After verification, the final answer appears in black.
If the system found issues, the black text may differ from the grey draft (sections may be tightened, clarified, or corrected).
The grey draft is only shown temporarily during streaming; the black answer is the one users should rely on.
You’ll see this message below a verified answer
Modes
Mode | What it means for users |
|---|---|
Off (default) | Only the standard black answer appears; no extra verification is performed. Best for brainstorming or creative work. |
Automatic | Draft streams in grey, then a verified black answer replaces it. Good default for most internal knowledge and support use. |
Always | Same visual behavior as Automatic, but with the strictest checking on every answer. Best for regulated, customer-facing, or high-trust scenarios. |
Stricter modes can slightly increase response time, especially for long or complex answers.
Admin recommendations
Internal knowledge & support: set default to Automatic.
Customer-facing or compliance-sensitive use: set default to Always.
Ideation and exploratory work: allow users to choose Off for faster, more flexible answers.
Configuration
Automatic mode sensitivity can be configured low, medium or high.