A Google Ads account-audit Claude Skill, as a SKILL.md you can edit - The Vault

A prompt gets you one pass at an audit. A Skill gets you the same methodology every time, edited once when your standards change rather than re-typed into a chat box. This is a complete Claude Agent Skill for a Google Ads account audit: an editable audit brain, not a black box, that names the exact GAQL resource behind every finding so the agent's reasoning is grounded in real data rather than a guess at what the account probably looks like.

The skill

Save this as SKILL.md in your Skills directory. It executes no code and mutates nothing: it reads Google Ads data, for example through the Google Ads MCP server pick on this shelf, and reasons over what comes back.

MARKDOWN

---
name: google-ads-account-audit
description: Use when auditing a Google Ads account for wasted spend, tracking gaps, structural problems or budget pacing issues. Triggers on requests like "audit this Google Ads account", "review our PPC setup" or "find where we're wasting spend".
---

# Google Ads account audit

You are conducting a structured audit of a Google Ads account. Read data only,
via GAQL against the Google Ads API (a connected MCP server, or a pasted
export). Never propose or execute a mutation: no bid, budget, status or
targeting change. Your output is a findings table, not an action.

Work through each section below in order. For each section, pull the named
GAQL resource, look for the listed symptoms, and add a row to the findings
table for anything worth flagging, whether it is a problem or a genuine
strength worth confirming.

## 1. Conversion tracking health
Resource: `conversion_action`, `campaign`
Pull every conversion action's status, category and whether it counts
primary conversions. Flag: no primary conversion action defined, a
conversion action left in "needs attention" status, or a campaign with an
optimisation goal that has no matching enabled conversion action.

## 2. Wasted spend
Resources: `search_term_view`, `geographic_view`, `campaign` (device segment)
Pull search terms by cost with zero or near-zero conversions over a
meaningful lookback window, geographic performance by country and region,
and cost by device. Flag: high-spend, zero-conversion search terms with no
matching negative, spend concentrated in a location outside the campaign's
intended targeting, and a device split badly out of line with the account's
device bid adjustments.

## 3. Account structure and naming
Resource: `campaign`, `ad_group`
Pull the full campaign and ad group list. Flag: inconsistent or
undecipherable naming, campaigns that duplicate the same keyword theme
against each other, and ad groups with a single, overly broad keyword theme
covering unrelated intents.

## 4. Bidding strategy fit
Resource: `campaign` (bidding_strategy_type, campaign_budget)
Pull each campaign's bidding strategy and recent conversion volume. Flag:
a Target CPA or Target ROAS strategy running on a campaign with too little
conversion volume to have exited the learning phase, and a Maximize Clicks
strategy still active on a campaign with conversion tracking in place.

## 5. Budget pacing
Resource: `campaign_budget`, `campaign`
Pull each budget's amount and the campaign's month-to-date spend. Flag: a
budget that is fully limited (capped) on a campaign that is otherwise
performing well, and a budget so large relative to spend that it is not a
real constraint on delivery at all.

## 6. Performance Max and Demand Gen coverage
Resources: `asset_group_asset`, `asset_group`, `campaign`
Pull asset group asset counts and ratings where available. Flag: an asset
group below the recommended minimum image, headline or video counts, a
missing listing group or audience signal on a PMax campaign, and overlap
between a PMax campaign and a Search campaign targeting the same terms
with no exclusion in place.

## 7. RSA and asset strength
Resource: `ad_group_ad_asset_view`, `ad_group_ad`
Pull Responsive Search Ad asset performance labels and ad strength ratings.
Flag: an ad group with only one active RSA, an RSA rated "Poor" ad
strength, and pinned assets that collapse the ad down to only one or two
real headline combinations.

## 8. Account history
Resource: `change_event`
Pull recent account changes. Use this to cross-reference any anomaly found
above against a recent edit, a bid strategy switch, a paused campaign, a
budget change, rather than reporting it as unexplained.

## Output format

Present findings as a single table:

| Area | Finding | Severity | Recommended action |
|---|---|---|---|

Severity is High, Medium or Low. Order the table High to Low. Close with a
one-paragraph summary naming the two or three findings that would move the
account most if fixed first. Do not suggest or take any mutating action;
recommendations are for a human to action.

Inputs

The Skill expects a connected Google Ads data source, an MCP server such as the official Google Ads MCP server pick on this shelf, a GAQL runner, or a manual export pasted into the conversation, and the account or customer ID to audit. It works best when Claude can run its own GAQL queries against the named resources rather than being handed a single flat export, since several sections cross-reference two resources (for example bidding strategy against conversion volume).

Failure modes

Without a live data connection, the agent will reason from whatever is pasted into the chat, which is only as complete as the export you provide, an audit run on a search-terms export alone will silently skip sections 1, 4, 5, 6 and 7 for lack of data, not because those areas are clean. On a very large account, pulling every resource in one pass can be slow or hit API row limits, scope the audit to one campaign type or date range first if that happens. The Skill has no access to Google Ads Editor, Google Analytics or Search Console data, so any recommendation is limited to what GAQL itself can see.