The machine, the rhythm,
the AI loop.
One search, start to finish
Ad Rank is recomputed at every search — there is no standing rank, only today's answer.
Specific = buyer; generic = browser
Keywords raise a hand; negatives veto first
Budget caps entry; verification gates volume
Ad Rank = bid × quality × assets, per search
RSA assembles from 15/4; brand pinned H1
CPC paid · GCLID attached — the thread
Form tags line + language; the reply clock starts
Bidding learns from
what you feed it.
Graded tours in, within 7 days · one primary conversion · secondaries observe
When you do the work
The discipline is as much about when NOT to touch the account.
Look, don't touch: pace, notices, reply speed. Same-day appeals. No edits.
The packet: terms → keep/block/promote · grade leads · upload conversions · post via Editor · settings check.
Cost per qualified tour + speed-to-lead. Budget vs calendar; ≤20%/week is routine.
Nov–Feb ramp +50–100% (manual) · quince calendar separate · Smart Bidding at ~30–50 conv/30d, not before.
Data-informed, human-approved
The watchdogs ride along: settings drift, the budget-limited regime flag, 'no new negatives in 30 days'.
Terms, leads, pace, policy, ledger, trends — every morning
Numbered items, each citing its rule + evidence
Approve · amend · hold · kill. Protected terms unproposable
Editor CSV only · the log line is the receipt
Verdicts tune the packet; tours feed bidding
Vendor figures retire as your data arrives
Drift, regime flag, staleness — flagged, never auto-fixed
Four numbers, one system
Everything else in the operation is derived from these.
Below it, manual. Per campaign, per 30 days.
Clicks / leads — our SLA is 7 days.
32% convert under a minute; 4% after a day.
Per week. Bigger gets a named decision.
The machine re-decides every search.The rhythm touches it once a week.The loop learns from your verdicts.
Companion visual: Google-Ads-Visual-Workflow-2026-10-16.html · crash course + sample tests in the same folder