How to Run a Google Ads Audit in 2026: The 10-Point Checklist We Use

Most Google Ads accounts leak money quietly. Not through one dramatic mistake, but through a dozen small ones: a broad match keyword nobody reviewed, a conversion action that double-counts, a campaign whose budget was set in January and never revisited.

An audit is how you find the leaks. This is the 10-point checklist we built into Karloe's ad-audit skill — usable by hand, in order, with what "good" looks like at each step. Nothing here requires an agency.

Before you start: pull the right window

Audit at least 30 days of data (90 if your spend is modest), segmented by campaign. Small accounts generate thin data; short windows make noise look like signal. Have the account's goal written down — literal target CPA or ROAS — because "underperforming" is meaningless without it.

The 10-point checklist

1. Conversion tracking (do this first)

Everything else in the account is decided by this data, so verify it before judging anything downstream.

• Every conversion action: is it firing? Is it firing once per conversion?
• Are primary vs. secondary conversion actions set deliberately? (Smart Bidding optimizes to primary.)
• Does the conversion count in Google Ads roughly reconcile with your source of truth — Stripe, your CRM, GA4?
• Any imported goals double-counting alongside native tags?

Good looks like: one clean primary conversion per real business outcome, reconciling within sane tolerance of your backend.

2. Account structure

• Do campaigns map to something you'd actually budget differently (product lines, funnel stages, geographies)?
• Are there overlapping campaigns competing for the same queries?
• Single-keyword ad groups everywhere, or thousand-keyword dumps? Both extremes age poorly.

Good looks like: you can explain why each campaign exists in one sentence.

3. Search terms report

The highest-yield fifteen minutes in the audit.

• Sort by cost, descending. Read every term above your average CPA's worth of spend.
• Flag terms that are irrelevant, competitor-navigational queries you don't want, or informational when you're bidding transactional.
• Check how much spend flows through terms Google hides as "other" — high share means less control.

Good looks like: the top 20 cost-bearing search terms all deserve the click.

4. Keywords and match types

• What share of spend runs on broad match, and is every broad keyword paired with Smart Bidding and real negatives?
• Duplicate keywords across campaigns fighting each other?
• Negative lists: do they exist, are they applied, when were they last touched?

Good looks like: broad match is a deliberate choice with guardrails, not a default.

5. Ads and assets

• Does every ad group have at least one strong responsive search ad with distinct headlines (not twelve rephrasings of one sentence)?
• Ad strength is a rough proxy — weak everywhere is still a finding.
• Are sitelinks, callouts, and structured snippets present and current? Stale promos in assets are a classic quiet leak.

Good looks like: ad copy that mirrors the search terms actually converting in step 3.

6. Bidding strategy

• Does the strategy match data volume? Target CPA/ROAS on a campaign with a handful of conversions a month starves the algorithm.
• Are targets realistic against trailing performance, or aspirational numbers that throttle delivery?
• Recent strategy switches without a learning-period annotation?

Good looks like: bidding automation fed by the clean conversions from step 1, with targets set from history.

7. Budget allocation

• Rank campaigns by ROAS or CPA against their share of budget. Is money where performance is?
• Campaigns capped by budget while outperforming? Campaigns limping along that would fund a winner?
• "Limited by budget" flags on your best performers are free money on the table.

Good looks like: budget follows demonstrated performance, reviewed monthly.

8. Audiences and geography

• Are location settings "presence" (people in your target) rather than "interest" where that matters?
• Any geo, device, or schedule segments burning spend with no conversions across the window?
• Remarketing lists attached — at least as observation?

Good looks like: no segment quietly consuming double-digit spend share with zero return.

9. Landing pages

• Does every high-spend ad group land on a page that matches its promise — message, offer, and next step?
• Mobile experience and load time on the top three pages by spend.
• Do forms and checkout actually work? (Test them. Post-migration breakage is more common than anyone admits.)

Good looks like: the searcher's query, the ad, and the page tell one continuous story.

10. Change history and hygiene

• Who changed what in the last 90 days? Unexplained performance shifts usually have an entry here.
• Auto-applied recommendations: on or off? If on, review what Google has been changing on your behalf — then decide deliberately.
• Disapproved ads, limited keywords, or policy warnings pending?

Good looks like: every material change traceable to a person and a reason.

Turning findings into a fix list

Don't fix as you go — collect findings, then order them by expected impact:

1. Tracking defects — fix immediately; everything else waits on clean data.
2. Spend on irrelevant terms — negatives and match-type fixes; instant savings.
3. Budget misallocation — shift toward winners; fast, low-risk gains.
4. Creative and landing-page gaps — slower to fix, compounding payoff.

Re-run the checklist after 30 days to confirm the fixes took. That before/after is also how you evaluate anyone — human or AI — who manages ads for you.

The honest alternative: don't do it by hand

This checklist takes hours by hand, which is exactly why most accounts never get audited. It's also perfectly mechanical until the judgment calls at the end — which makes it ideal work to delegate to an AI marketing agent. Karloe runs this audit with read-only access to your account and returns the findings document in Slack; the fix list stays under your approval. The checklist above is the same one it uses — so you can verify its work, which is how it should be.