TheMattBerman Google Ads Draft Summary
Why this skill exists
After an audit, search-term review, or tracking diagnosis, the copilot produces individual draft files. But individual files don't answer the operator's real question:
"I have 20 minutes. What should I review and apply first?"
This skill answers that.
Read first:
drafts/DRAFTS.md— draft system documentation
Read workspace:
workspace/ads/drafts/_index.md— current queue- All
workspace/ads/drafts/*.md— individual draft files workspace/ads/account.md— account contextworkspace/ads/findings.md— for tracking confidence context
What This Produces
A Draft Summary — a single document that:
- Lists every pending draft (status:
proposedorapproved) - Classifies each by priority, impact, risk, and reversibility
- Shows dependency chains (what must be applied before what)
- Produces a recommended apply order (the sequence an operator should follow)
- Flags blocked actions (e.g., budget scaling blocked by tracking problems)
- Estimates total impact of applying all pending changes
This is the current backlog snapshot, not the audit-run packet. _summary.md should always reflect the latest pending queue. Durable audit packets belong in _batch-*.md.
Priority Classification
Each draft gets a priority level based on these criteria:
P0 — Fix First (Blocking)
- Tracking problems that gate all other decisions
- Actively harmful configurations (e.g., paying for store visits as conversions)
- Budget bleeding on clearly wrong traffic
- Apply these before anything else
P1 — Quick Wins (High Confidence, Easy, High Impact)
- Clear negative keyword additions with significant waste (>$100/month)
- Keyword pauses with obvious waste
- Simple tracking cleanup (remove micro-conversion from primary)
- Apply these in the same session as P0
P2 — Strategic Moves (Requires Thought)
- Structure changes (campaign splits, ad group reorganization)
- Budget reallocation between campaigns
- Landing page changes
- Schedule a focused review session for these
P3 — Refinement (Low Risk, Incremental)
- RSA headline/description refreshes
- Minor negative additions (<$50/month waste)
- Value setting updates
- Apply when convenient, batch with other changes
Impact Assessment
Each draft gets an impact estimate:
| Impact Level | Description | Signal |
|---|---|---|
| Critical | Stopping active harm or fixing broken measurement | Tracking broken, massive waste |
| High | Significant waste reduction or conversion improvement | >$200/month waste, or >20% of spend affected |
| Medium | Meaningful improvement in efficiency or clarity | $50-200/month impact, structure improvement |
| Low | Minor optimization, incremental improvement | <$50/month, cosmetic structure |
Risk Assessment
| Risk Level | Description | Reversibility |
|---|---|---|
| High | Could disrupt Smart Bidding learning, reduce impression share, or block good traffic | Moderate — takes 2-3 weeks to recover |
| Medium | Temporary performance dip possible, but recoverable | Easy — undo within a day |
| Low | No meaningful risk of regression | Easy — instant undo |
Dependency Chain Logic
Drafts can depend on each other. Common dependency patterns:
Tracking Fix (P0)
├── blocks → Budget Scaling (P2)
├── blocks → Bid Strategy Changes (P2)
└── informs → Structure Changes (P2)
Negative Keywords (P1)
└── before → Structure Changes (P2)
└── before → RSA Refresh (P3)
└── before → Budget Scaling (P2)
Rules:
- Tracking fixes always come first (they change the data everything else is based on)
- Negatives before structure (clean the traffic before reorganizing it)
- Structure before RSAs (know the buckets before writing the copy)
- Everything before budget scaling (scale only what's working)
Draft Summary Document Format
# Draft Summary — [Date]
Account: [Name] (CID: [ID])
Tracking confidence: [HIGH / MEDIUM / LOW / BROKEN]
Pending drafts: [N]
Total estimated waste addressable: $[amount]/month
## Recommended Apply Order
### Step 1: [Draft name] — P0
- **File:** `workspace/ads/drafts/[filename]`
- **Type:** [tracking-fix / negatives / structure / budget / rsa / landing-review]
- **Summary:** [one sentence]
- **Impact:** [Critical / High / Medium / Low]
- **Risk:** [High / Medium / Low]
- **Reversibility:** [Easy / Moderate / Hard]
- **Estimated waste stopped:** $[amount]/month
- **Blocked by:** [nothing / other draft]
- **Blocks:** [list of drafts that depend on this one]
### Step 2: [Draft name] — P1
[same format]
### Step 3: [Draft name] — P2
[same format]
[... repeat for all pending drafts]
## Dependency Map
```text
[visual or text representation of dependencies]
Blocked Actions
- [action] — blocked by [reason]. Unblocks after [draft] is applied.
Quick-Apply Candidates
Drafts that are low-risk, high-confidence, and have no dependencies:
- [draft name] — [one-line summary]
Deferred / Needs More Data
Drafts that shouldn't be applied yet:
- [draft name] — [reason: needs more data / confidence too low / depends on client input]
Total Impact Estimate
If all pending drafts are applied in the recommended order:
- Waste reduction: ~$[amount]/month
- Tracking improvement: [description]
- Structure improvement: [description]
- Time to full effect: [estimate]
---
## How to Generate the Summary
### Step 1: Read the Index
Load `workspace/ads/drafts/_index.md` to get the list of all drafts and their current status.
### Step 2: Read Each Pending Draft
For each draft with status `proposed` or `approved`, read the full file to extract:
- Action type(s)
- Confidence level
- Risk and reversibility
- Dependencies
- Estimated impact (waste $ if quantified)
### Step 3: Check Tracking Confidence
Load the most recent tracking diagnosis from `workspace/ads/findings.md`.
Tracking confidence determines which drafts are blocked.
### Step 4: Build the Dependency Chain
Map out which drafts depend on which. Apply the standard ordering:
1. Tracking fixes
2. Keyword pauses (stop active bleeding)
3. Negative keyword additions
4. Structure changes
5. RSA refreshes
6. Landing page improvements
7. Budget reallocation
### Step 5: Assign Priority and Impact
Using the P0-P3 framework and impact/risk tables above.
### Step 6: Write the Summary
Output the Draft Summary document to `workspace/ads/drafts/_summary.md`
---
## Output Locations
- **Summary document:** `workspace/ads/drafts/_summary.md` (overwritten each time)
- **Audit packets:** `workspace/ads/drafts/_batch-YYYY-MM-DD-[account-slug].md` (created by `/google-ads audit` when one run emits 2+ drafts)
- **Index update:** Update `workspace/ads/drafts/_index.md` if any status changes are warranted
- **Console output:** Also display the summary inline for immediate review
---
## Multi-Account Support
When `workspace/ads/drafts/_index.md` contains drafts for multiple accounts (identified by account slug in the filename):
- Group the summary by account
- Show per-account tracking confidence
- Show per-account total impact
- Cross-account dependencies are rare but possible (shared negative lists, MCC-level changes)
---
## Rules
- **Read every pending draft.** Don't summarize from the index alone — the index is a title list, not a summary.
- **Always check tracking confidence.** It gates everything.
- **Be honest about confidence.** If a draft says "Medium confidence," don't promote it to P1.
- **Surface the 20-minute review.** The operator should be able to read the summary, pick the top 2-3 actions, and apply them in one session.
- **Don't create new drafts.** This skill summarizes existing drafts — it doesn't generate new recommendations.
- **Flag stale drafts.** If a draft is >30 days old, note it. The data it was based on may no longer be current.
- **Flag conflicting drafts.** If two drafts recommend contradictory actions, call it out explicitly.