Safety & guardrails

Automation is only safe if it's cautious by default. These guardrails are always working in the background to make sure it never does damage you can't undo.

What's protected

• Terms making sales: a converting term is only ever blocked if it's performing far worse than target (> 2.0x CPA, or < 0.50x ROAS). Healthy converters are never auto-blocked.

• Your brand: with brand handling on Protect, brand searches are never auto-blocked and are kept out of the learning model.

• Local searches: a search naming your service area is never auto-blocked, even on a quiet week.

• Terms the learning model rates well: protected from the "nobody clicks" rule, so a useful term isn't dropped just for low clicks.

Two account-level safety nets

• Exact-by-default negatives: auto-applied negatives use exact match, so a single negative can never over-block a whole theme of traffic.

• Flag-rate guard: if your always-block list ever matches more than about 10% of a run, those terms are held for review instead of auto-applied. So a mis-tuned block word can't flood an account with negatives - you'll see them all, but nothing applies during a suspected flood.

Example. Someone adds "service" to the flag list on a business where half the searches contain "service". Instead of blocking hundreds of terms, the flag-rate guard holds them for review and you catch the mistake in seconds.

Good to know. Because every applied change is logged with an Undo, even something that slips through is quick to reverse from Recent Changes.

RELATED ARTICLES