Google Ads Apply

Commands

Apply a Draft

/google-ads apply [draft-file]

Flow:

1. Read the draft file
2. Parse all proposed actions (manifest-first for budget drafts; legacy parsing for v1 negatives/pauses)
3. Validate each action (correct account, valid targets, within live apply scope)
4. Resolve human-readable names to Google Ads resource IDs via GAQL
5. Display dry run with every mutation listed, action type summary, and risk assessment
6. Wait for operator confirmation ("confirm" or "cancel")
7. Execute each action via Google Ads REST API v20
8. Verify changes via GAQL re-query
9. Write audit trail (session log, master log, reversal registry)
10. Update draft status to "applied"
11. Show post-apply summary with undo instructions

Review a Draft (No Apply)

/google-ads apply review [draft-file]
/google-ads apply review --all

Flow:

1. Parse the draft into structured actions
2. Show action breakdown with counts by type and risk levels
3. Show confidence and dependencies
4. Provide a proceed/defer recommendation
5. Show next-step commands (dry-run or full apply)

No network calls, no API access needed. Pure local analysis.

Check Operator Status

/google-ads apply status

Shows at a glance:

• Connection state (account, API version, credentials)
• Pending drafts with status
• Active reversals grouped by draft
• Recent apply sessions
• Suggested next step

Undo a Single Action

/google-ads undo [reversal-id]

Flow:

1. Look up reversal record in workspace/ads/audit-trail/reversal-registry.json
2. Display what will be reversed
3. Warn if change has been live >7 days
4. Wait for confirmation
5. Execute reversal via Google Ads API
6. Verify via GAQL
7. Update reversal registry (status: "undone")

Undo an Entire Draft

/google-ads undo-draft [draft-file]

Reverses ALL active actions from a specific draft. Same confirmation flow.

View Apply Log

/google-ads apply log

Shows recent apply sessions from workspace/ads/audit-trail/_log.md.

Scope Guardrails (v1 + v2 budgets)

Allowed Actions

Why These Are Safe

• Negative keywords can only REDUCE traffic. They cannot increase spend, break ads, or corrupt tracking.
• Pausing keywords stops traffic to one keyword. All history, quality scores, and configuration remain intact.
• Pausing ad groups stops traffic to a set of keywords. Same preservation guarantees. Broader blast radius than keyword pause, hence Medium risk.
• All are instantly reversible: remove the negative, or set status back to ENABLED.

Blocked Actions (Hard Reject)

If a draft contains actions outside the supported scope, the apply command will:

1. List the out-of-scope actions
2. Explain which are out of scope and why
3. Offer to apply ONLY the in-scope actions
4. Never execute out-of-scope actions regardless of confirmation

Draft Parsing

The apply layer parses these sections from draft markdown files:

Both negative-draft.md and pause-draft.md templates are supported. Mixed-type drafts (negatives + pauses in one file) are fully supported — the parser extracts from all applicable sections and deduplicates.

Budget drafts are different:

• They MUST include ## Apply Manifest
• The manifest is authoritative when present
• Draft-level meta.budget_policy controls whether any net increase is allowed

Draft Templates

• drafts/templates/negative-draft.md — negative keyword additions + removals + narrows
• drafts/templates/pause-draft.md — keyword pauses + ad group pauses + impact summary

Validation Checks (Before Dry Run)

Before showing the dry run, validate:

1. Account match: Draft account ID matches connected account
2. Campaign exists: Campaign referenced in draft is still present (GAQL lookup)
3. Ad group exists: For ad-group-scoped actions (GAQL lookup)
4. No duplicates: Negative keyword doesn't already exist at the same scope
5. Match type valid: PHRASE, EXACT, or BROAD only
6. Target exists: Keyword being paused still exists and is currently ENABLED
7. Ad group status: For ad group pause, target is currently ENABLED
8. Within scope: Supported actions are add negative, pause keyword, pause ad group, and manifest-backed campaign daily budget changes

If validation fails, report which actions failed and why. Don't block the entire apply — allow valid actions to proceed.

Dry Run Display

═══════════════════════════════════════════════════════
 DRY RUN: [draft-file]
═══════════════════════════════════════════════════════

Account: [Name] (CID: [ID])
Actions: [N] valid / [M] total

#    Action           Target                       Detail                      Risk
---  ---------------  ---------------------------  --------------------------  ------
1    ADD NEG          Campaign: [name]             "near me" [PHRASE]          Low
2    ADD NEG          Campaign: [name]             "debris chute" [PHRASE]     Low
...
13   PAUSE KW         AG: [name]                   "waste mgmt" [EXACT]        Low
14   PAUSE AG         Campaign: [name]             [ad group name]             Medium

Summary:
  • 12 negative keyword addition(s)
  • 1 keyword pause(s)
  • 1 ad group pause(s)
  • Reversibility: All actions reversible

⚠️  This will make REAL changes to account [CID].
Type 'confirm' to proceed, or 'cancel' to abort:

Post-Apply Verification

After execution, run verification queries:

Audit Trail

After every apply session, these files are updated:

1. Master log: workspace/ads/audit-trail/_log.md — append-only session summary
2. Session detail: workspace/ads/audit-trail/YYYY-MM-DD-apply-session.md — full action-by-action log
3. Reversal registry: workspace/ads/audit-trail/reversal-registry.json — machine-readable undo records
4. Draft file: Updated with applied date and reversal IDs
5. Draft index: workspace/ads/drafts/_index.md — moved to Applied section
6. Change log: workspace/ads/change-log.md — record of what changed

CLI Scripts

Safety Rules

1. Never apply without dry run. The operator must see exactly what will change.
2. Never auto-approve. The operator must explicitly type "confirm."
3. Never apply out-of-scope actions. Hard reject, clear explanation.
4. Never delete anything. Pause is the maximum severity in v1.
5. Log everything. Every action, every error, every reversal.
6. Fail gracefully. If one action fails, continue with the rest. Report the failure.
7. Verify after apply. Re-query to confirm changes took effect.
8. Make undo easy. Every action has a stored reversal instruction.
9. Rate limit. Max 50 mutations per apply session. 500ms delay between mutations.
10. No batch apply across accounts. One account per apply session.