Sift Docs

Learned Patterns is Sift studying your own team's behavior. It watches how agents edit AI-drafted replies before sending, and proposes new voice rules based on what it sees - nothing applies automatically until someone accepts it.

At a glance


Runs on demand ("Run learning now")	✅
Proposes voice rules from real agent edits, not guesses	✅
Nothing applies until a suggestion is accepted	✅
Accept a suggestion, or write your own rule directly	✅
Filter active rules by channel	✅
Managers and above can edit	✅

What you're looking at

Found under Settings > AI > Learned Patterns. The page has two parts: Learned from your team, where Sift surfaces proposed rules after a learning run, and Active rules, the list of rules currently shaping every AI-drafted reply.

The Learned Patterns page in AI Studio, showing the Learned from your team and Active rules sections. — Learned Patterns: proposed rules from real agent edits, nothing applies until accepted.

How it works

Click Run learning now to have Sift study recent agent edits to AI drafts and propose new voice rules. If there's nothing new to learn from, the page says so plainly rather than inventing a suggestion.
Each proposed rule can be accepted (moving it into Active rules) or dismissed.
Active rules are rules you've either accepted from a learning run or written yourself. Each one shapes every AI-drafted reply until it's removed. Rules can be scoped to a specific channel or left applying to any channel.

Tips

Run learning periodically, especially after onboarding a new agent whose edits might reveal voice patterns Sift hasn't picked up on yet.
Because nothing here applies without explicit acceptance, it's safe to run learning often - review what it surfaces on your own schedule.

Voice & Tone

The voice profile every rule here ultimately refines.

Brand Profile

The business context every AI stage grounds on.

Learned Patterns