Assistant Output Verification Before Client Delivery (QA)

Assistant Output Verification Before Client Delivery (QA)

1. 01

   Pull the assistant's draft against the original client brief line by line

   Open the source request (email thread, Slack message, or ticket) beside the generated output and confirm every asked question has a corresponding answer. Flag any item the assistant silently skipped.

2. 02

   Verify all names, figures, and dates against the CRM record

   Cross-check contact names, deal values, and scheduled dates against the client's CRM entry. Assistants that pull from stale context will confidently repeat outdated numbers.

3. 03

   Confirm no cross-client data leaked into the draft

   Search the output for any client name, project codename, or figure that belongs to a different account. This is the highest-severity failure mode when assistants share a memory store across accounts.

4. 04

   Check tone against the client's communication history

   Compare the draft's register to the last three approved messages sent to that client. A formally correct reply can still read wrong if the relationship runs casual.

5. 05

   Run the claim audit on any factual assertion

   For every statistic, capability claim, or competitive reference in the draft, require a source or remove it. Vendor announcements are not evidence of tested performance.

6. 06

   Log the correction in the assistant's context notes

   Record what the assistant got wrong and why, so the next generation cycle inherits the fix rather than repeating the error.

7. 07

   Route the verified draft to a named human approver before send

   The approver is accountable for the output, not the assistant. No client-facing message leaves without a person's name attached to the approval.