10 min read
Connected Google Ads Audits: What an AI Agent Can Check
The reads a connected audit agent makes, the Google Ads resources behind each, the limits of the connection, and how findings turn into approved changes.
An AI agent that connects to Google Ads audits an account by reading it through the Google Ads API rather than from an export. Given a connected account, Parallel can read conversion action setup, campaign status and shared budgets, budget pacing, the search terms report, negative keyword conflicts, ad policy status and Ad Strength, Recommendations, Performance Max asset group performance and listing group filters, and change events for the last 30 days, and write what it found as a document that names the campaign, metric and period for each finding.
The connection has limits, and an honest audit states them. Change events through the API go back 30 days; the interface's Change history goes back two years. Product diagnostics read a bounded set of products from the account's shopping product data, not the Merchant Center console. The Performance Max channel comparison comes from an export of the channel performance report. Margin, the promotion calendar and lead quality reach the audit only as imported conversion values, offline conversions, or files and notes you supply. And nothing the audit finds changes the account: a proposed fix is drafted and waits for an authorized person to approve.
What a connected audit can read, check by check
Each row is a read Parallel can make through the API when asked to audit a connected account. The limits matter as much as the list of checks.
| Read | Google Ads resource it queries | Limit |
|---|---|---|
| Conversion tracking setup | Conversion actions: status, category, counting type, primary for goal, tag snippets | Setup only; it does not show when an offline import last delivered, or whether CRM or Data Manager payloads are healthy |
| Account health | Campaigns and shared budgets: status counts, budget utilization, warnings | Campaign status and budget structure, not a performance judgment |
| Budget pacing | Campaign spend against budget for the current period, with a projection | The projection is arithmetic on spend to date; demand-led pacing, whole-day schedules and campaign total budgets move spend within the period by design |
| Search terms | The search terms report for Search campaigns; search terms insight categories for Performance Max | Google omits some queries for privacy; for query-level Performance Max terms, use the Performance Max search terms report in the interface or its export |
| Negative keyword conflicts | Ad group and campaign criteria: negatives that may block positive keywords | Campaign and ad group negatives; account-level negative keywords are checked in the account |
| Policy and Ad Strength | Ad policy approval status; responsive search ad Ad Strength | Disapproved and limited ads; Ad Strength is Google's rating, not a performance measure |
| Recommendations | The Recommendations resource, by type and campaign | Returns Google's suggestions, up to 50 per read; applying one is a separate, approval-gated step |
| Performance Max | Asset group performance; listing group filters; campaign-level brand and URL exclusions; search terms insights | The implemented API channel query is not Google's complete channel report; export that report for the full comparison |
| Recent changes | change_event: campaign, ad group, ad, keyword, budget, asset and feed changes | 30-day maximum lookback through the API; conversion action changes are not in change events; the interface's Change history covers two years |
| Conversion lag | Days to conversion for the account | Used to mark which recent days are incomplete |
A sensible order, and why
Measurement before money. Reading the conversion actions first makes sense because every later finding is expressed in conversions. If the primary action counts every submission, or an offline import has stopped, the cost per conversion findings that follow are about the wrong number. The tracking read says what the account counts and how; whether the import is still delivering is confirmed in the conversion action's details in the interface or at the source, and whether the business agrees with what counts is the reader's call.
Changes before conclusions. Change events over the API's 30-day window come next. A bid target change, budget increase or status change gives you an explanation to investigate; its timing does not establish cause. The summary diagnostic also flags heavy change activity, more than 20 events in the lookback, as a prompt to read the history before the performance; that is a heuristic, not evidence of a problem. A new or changed conversion action does not appear in change events, so that check is made from the conversion actions read and, for the date, from the interface's Change history.
Then the reads that explain cost: search terms for Search and search terms insight categories for Performance Max, negative keyword conflicts, budget pacing against the period, policy status and Ad Strength, and Recommendations with the auto-apply state noted. Performance Max campaigns get asset group performance, listing group filters and the exclusions in place, with a note that the channel comparison needs the exported channel performance report.
Lag last. The Days to conversion view marks which recent days are incomplete, and every cost per conversion finding on a short window should carry that mark. This is one order that works; an audit asked a narrower question reads what that question needs.
What a finding looks like, and what it does not claim
Two hypothetical findings, in the shape a good request asks for: entity, metric, period, source, next step, and what is outside the read.
First: in the Brand Search campaign, the primary conversion action Lead form is enabled, category Submit lead form, counting type Every. A second action, Qualified lead import, is enabled and secondary. The campaign's selected standard goal includes Lead form, and no custom goal includes the qualified-lead action. In this example, the main conversion total therefore counts the form submission rather than the qualified lead. Proposed next step: confirm at the source, and in the action's details in the interface, that the import is still delivering, since the API read shows setup rather than delivery; then discuss with the owner whether the qualified action should be primary given the campaign's goals and the loss of comparability with earlier periods. No bid or budget change until then. What this finding needed from outside the read: the import's delivery status and the business's view of which action should count.
Second: in the Performance Max campaign, search terms insights show a new category of used and refurbished queries this month, the listing group includes the full catalog, and change events show the budget raised by 40 percent on the third. Proposed next step: before drafting any exclusion, confirm with the owner whether the shop sells used or refurbished items, and open the Performance Max search terms report for the actual terms, since the insight is a category. If the terms are outside what the shop sells, draft a campaign-level negative keyword list for those terms, with match types, for approval. What the finding does not say: that the budget increase caused the new queries, which channel took the extra budget, because that is in the channel performance report export, or whether the new queries produced worse orders. Category-level conversion totals do not answer that question; order values, margins, refunds or other business records may be needed.
That second half is the audit's honesty. Through this connection it can name the campaign, the query category, the date of a change event and the exact proposed exclusion. It can show the conversion totals Google reports for a search category, but those aggregates do not identify individual leads, orders or their quality. It also cannot read change events older than 30 days or establish margin that has not been supplied. Other sources can: the interface's Change history covers two years, conversion values and offline imports carry business outcomes into the account, and files you supply carry the rest. The finding should say which of these it used and which it still needs.
Questions and answers
What does a connected Google Ads audit read that an export does not?
The current conversion action setup, change events for the last 30 days, negative keyword conflicts across ad groups, ad policy status, Recommendations by type, and Performance Max asset groups and listing group filters, each with campaign and period attached, and it can be rerun the next day on the same account. An export is a snapshot of whatever was exported; some comparisons, such as Google's full Performance Max channel report, currently need the interface or an export for this workflow.
How far back can a connected audit read Change history?
Thirty days of change events through the Google Ads API. The interface's Change history covers the past two years, and the team's own records cover the rest; an audit that needs an older date says so and uses those.
Does the audit change my account?
No. Reading never changes the account. Any fix the audit proposes is drafted with the exact values and waits for an authorized person to approve, including applying a Google recommendation. Approval permits the attempt; Google's response and a fresh read confirm the result.