The 4 AI prompts I run before every weekly client call

A 35-minute pre-flight that replaces 2-3 hours of report assembly. Includes the Google Ads Scripts setup, the four skills, and the download links.

Every weekly client call has the same risk: the client comes in with a question I haven't anticipated, and the next 45 minutes turn into defense instead of direction.

Eighteen months ago I started running a fixed pre-flight before every weekly call. Four prompts. Same order. Same outputs. About 35 minutes of structured pre-work. Today the call starts with the agenda already proven by data, the three ranked actions in front of the client, and the dead-end conversations cut off before they begin.

If you haven't read the first post — Why I built actcenter — start there. This post assumes you've felt the day-to-day cost it describes.

The pre-flight in 30 seconds

I run them in order. Output of one informs the next. Total: ~35 minutes for a single account.

Setup: the data layer (Google Ads Scripts → Drive)

Before the prompts work, the data has to be where they expect it. The cleanest way I've found is Google Ads Scripts that run daily and write snapshots into a shared Drive folder. No connector, no API, no engineering ticket.

The pipeline: paste the script into Google Ads → script writes CSVs to Drive nightly → Claude reads via the Drive connector when you invoke a skill → output lands in the same client folder.

A simplified excerpt:

// snapshot.gs · runs daily 04:00 in Google Ads UI
function main() {
  const folder = DriveApp.getFoldersByName("snapshots").next();
  const accountId = AdsApp.currentAccount().getCustomerId();
  exportReport("CAMPAIGN_PERFORMANCE_REPORT", folder, accountId, "LAST_7_DAYS");
  exportReport("SEARCH_QUERY_PERFORMANCE_REPORT", folder, accountId, "LAST_14_DAYS");
  exportReport("KEYWORDS_PERFORMANCE_REPORT", folder, accountId, "LAST_30_DAYS");
}

~20 minutes per account the first time, zero after that. If you're inside an MCC you can authorize at the manager level and target a list of customer IDs.

Prompt 01 — Weekly Health Check (8 min)

Skill: actcenter_weekly-health-check · download below

The first read. Macro-level state of the account.

The framework is the tCPA Health Check. Every account, every week, lands in one of five states:

• HOLD — CPA within ±10% of target, volume stable. Don't touch.
• RAISE — CPA at or below target, impression share lost to rank ≥ 30%. Lift tCPA 10-15%.
• LOWER — CPA ≥ 20% above target for 3+ weeks. Cut tCPA 10-15%.
• DIAGNOSE — CPA volatile, or conversion volume dropped > 30%. Investigate before adjusting.
• RISK — Tracking broken, pacing < 50% or > 200%. Stop and fix.

The skill outputs top 3 actions for the week, ranked, with one-line rationale. Anything not in the top 3 lives under "what I'm explicitly not changing this week" — equally important for keeping the call honest.

Replaces: the hour cross-referencing four tabs + the hour writing it up in Google Docs.

Prompt 02 — Search-Terms Sweep (10 min)

Skill: actcenter_search-terms-analyst · download below

Where is money being wasted on queries we shouldn't serve, and which queries deserve their own ad group?

Runs a Query Classification Grid: every query lands in a cell of spend × intent × conversion. Scans for the ten standard waste clusters: price-sensitive ("free", "cheap"), DIY, job seekers, wrong service, wrong geography, competitors, educational, reviews, adult/off-topic, brand drift.

Outputs two files: negatives CSV ready for Google Ads Editor + summary markdown with the top three wasted clusters, winners worth promoting, and structural flags.

Replaces: 90 minutes of squinting at a 4,000-row spreadsheet.

Prompt 03 — RSA Review & Refresh (9 min)

Skill: actcenter_rsa-review · download below

Are the ads doing their job, or has Ad Strength slid?

Applies a 7-axis review: character compliance, variety (15 headlines × 5+ angles), pinning discipline, Title Case, punctuation policy, message-match to landing page, Ad Strength signal.

If a refresh is needed: 15 new headlines mapped to 8+ angles + 4 descriptions, pin guidance, 14-day test plan. All in a CSV ready for Google Ads Editor.

Replaces: 45 minutes of staring at the current ad, hand-counting characters, failing the policy check on the third upload.

Prompt 04 — GBP Audit (8 min)

Skill: actcenter_gbp-audit · download below

For local-services accounts especially: the Google Business Profile is doing real conversion work. If the GBP is weaker than the top three competitors, paid is fighting uphill.

Audits four areas: completeness, reviews health, local SEO signals, competitive position (local-pack rank vs top 3 competitors).

Outputs operator-facing markdown + optional client-facing PPTX deck.

Replaces: the bi-weekly "I'll check the Maps listing and circle back" that never actually got done.

What the call actually looks like now

• First 5 min: the state, the three actions, the one risk. (Prompt 01)
• Next 8 min: wasted spend removed + winning queries promoted. (Prompt 02)
• Next 8 min: ad refresh status, test plan. (Prompt 03)
• Next 5 min: GBP / off-paid context. (Prompt 04)
• Last 4 min: open questions, next week's commitments.

What used to be a defensive call is now a directive one. The client sees the same shape every week. Predictable cadence builds trust faster than any single brilliant insight.

Caveats

The skills are opinionated. They assume Smart Bidding, lead-gen verticals, accounts above ~$500/month.

The skills don't act on the account. Every change is staged for the operator to approve. Algorithm-native means amplifying judgment, not replacing it.

The Google Ads Script is the linchpin. Monitor the success log every Monday.

Adapt the prompts to your house style. The downloadable versions are clean and operator-facing — add your compliance pack on top if you're in a regulated vertical.

Download the skills

Each skill is a folder with SKILL.md + README.md that drops into any Claude client that supports skills. Free to use, modify, and rename.

The pre-flight isn't magic. It's just the smallest set of frames I could find that turn 25 noisy accounts into a calm Monday morning.