The direct answer
Choose review mode until a team can explain what makes a draft safe to schedule. Semi-automatic and automatic modes can reduce repetitive work, but only when they are bounded by clear content, timing, and volume rules.
Three modes, three responsibilities
| Mode | System responsibility | Human responsibility |
|---|---|---|
| Review | Prepare drafts and signals | Approve every post |
| Semi-automatic | Schedule drafts that pass configured gates | Set gates and handle exceptions |
| Automatic | Publish within configured limits | Own the policy and audit outcomes |
Signals are not proof
A brand-fit score can surface whether a draft resembles a reviewed voice profile. It cannot verify an external fact, predict how an audience will react, or replace a legal and product review. Treat it as a prioritization signal.
Operating safeguards worth documenting
- A daily maximum number of automatic posts.
- A minimum gap between posts.
- Quiet hours in the brand's time zone.
- A way to pause the queue and return to review mode.
- A clear list of claim types that always require human approval.
For the implementation surface, see the SmAgent approval workflow. For a transparent first-party example, read the SmAgent LinkedIn case study.
See the B2B SaaS workflow →