- Twelve checks, 60–90 days of data, read-only access. No changes needed to run any of them.
- The four that most often move real money: conversion actions set up wrong, budget-capped winners, search-term waste, and brand searches leaking into non-brand campaigns.
- Rank findings by dollars per month, not by how interesting they are, and state the sample size behind each one.
- In one real account the checks found $1,022 a month in one campaign alone, a 34–49% impression share lost to budget on every campaign, and 129 of 749 search terms running in several campaigns at once.
- Change one thing at a time and set checkpoints (week 1, week 3, day 90) before you change it.
This is the checklist we run on every local service account, in the order we run it. Each check names what to look at in Google Ads, what a red flag looks like, and what we found in a real account (an auto film shop in Minneapolis, names changed, numbers unchanged). It's long on purpose; skip to the plan if you only want the method.
Before you start
- Data window: 60–90 days, ending at least two weeks ago so late-closing jobs have landed.
- Access: read-only is enough. If someone else runs the account, ask them to add you (or us) as a read-only user.
- What you're measuring against: paid jobs if your CRM or shop software reports them back to Google Ads; leads otherwise. Decide this first, because every check below is ranked by it.
The 12 checks
1. Conversion actions: what is Google actually counting?
Look at: Goals → Conversions → Summary. Every action, its source, whether it's primary or secondary, and how many it recorded in the window.
Red flag: page views, “clicked phone number”, chat opens or directions counted as primary conversions; booked or paid jobs set as secondary; an action with hundreds of conversions that nobody can explain.
Real account: a pop-up form action had 255 primary conversions and wasn't in the owner's list of lead sources; completed jobs were partly secondary. See primary vs. secondary.
2. Cost per lead vs. cost per job, per campaign
Look at: Campaigns view with columns for cost, conversions, and (if you have it) the job conversion action. Divide.
Red flag: a spread of 5× or more between campaigns, or the biggest budget on the most expensive jobs.
Real account: $4–$26 per lead became $20–$483 per job. The largest campaign (33% of spend) produced 5% of jobs. Full walkthrough: cost per lead vs. cost per job.
3. Budget-capped winners
Look at: Campaigns → Columns → Competitive metrics → Search lost IS (budget).
Red flag: a campaign with a low cost per job losing 20% or more of impressions to budget while a high-cost campaign spends freely.
Real account: every campaign lost 34–49% to budget, including brand searches at $20 per job. Details: budget-capped winners.
4. Search terms that will never buy
Look at: Keywords → Search terms, sorted by cost, 90 days.
Red flag: job seekers (“technician salary”), DIY and research (“how to”, “vs”, “DIY”), competitors' brand names you can't win, and wrong service entirely.
What to do: add negatives at the campaign or account level. Expect 5–15% of search spend to be recoverable in an account that hasn't done this in a year.
5. Brand searches leaking into non-brand campaigns
Look at: Search terms containing your business name, filtered to campaigns that aren't your brand campaign.
Red flag: any meaningful spend. Those searches would convert anyway, and they make the non-brand campaign look cheaper than it is.
Real account: $209 and 19.9 conversions of brand searches inside non-brand campaigns over 90 days.
6. Campaign overlap
Look at: the same search term appearing under two or more campaigns.
Red flag: a large share of spend on terms that several campaigns compete for. Google shows at most one of your ads per search, so you aren't bidding against yourself, but the data gets split and moving budget between those campaigns stops meaning anything.
Real account: 129 of 749 search terms ran in more than one campaign, 53% of search-term spend.
7. Duplicate keywords
Look at: the same keyword and match type in more than one ad group or campaign.
Red flag: duplicates with different bids or landing pages. Pick one home for each keyword.
8. Geography: where the money goes vs. where the jobs come from
Look at: Locations report (user location, not just targeted location) and, if available, distance from your location.
Red flag: a location campaign spending heavily in another location's area; spend on towns that produce clicks but no leads; a cliff in lead rate beyond a certain distance.
Real account: the suburban campaigns spent about 22–23% of their budget in the city.
9. Hour of day and day of week
Look at: Campaigns → Segment → Time → Hour of day / Day of week, with cost and conversions.
Red flag: paid clicks when nobody answers the phone; a block of hours with spend and no leads. Don't just turn ads off at night: check whether night clicks produce leads first. This view only shows primary conversions, so it can't show jobs.
10. Devices
Look at: Segment → Device.
Red flag: one device with twice the cost per lead of the others and no bid adjustment; a landing page that only works well on desktop while 80% of clicks are mobile.
11. Change history
Look at: Change history for the last 30 days (Google only keeps details that long in the API; the interface shows more).
Red flag: budget or status changes right before performance shifted; changes nobody remembers approving.
Real account: two budgets were raised from $20 and $14.50 to $30 a day on the same date; one of them was the account's most expensive campaign per job.
12. Attribution lag and reporting window
Look at: the date range on every report you've been judging the account by.
Red flag: decisions made on the last 7 or 14 days. Jobs are credited to the click date and close weeks later. The most recent two weeks of any cost-per-job report will understate jobs; a report that doesn't say so is misleading you.
Turning findings into a plan
- Price every finding in dollars per month. “Campaign X costs 6× the average per job” becomes “$1,022 a month that could buy jobs elsewhere.” Rank by that.
- State the sample behind each one. 761 clicks and 5 jobs is solid. 40 clicks and 1 job is a hint. Say which.
- Decide where the money goes before you cut. Usually: the campaign with the lowest cost per job and the highest lost impression share to budget.
- Estimate pessimistically. We assume moved dollars buy jobs at twice the receiving campaign's current cost per job. If the plan still looks good under that assumption, it's a good plan.
- Record the baseline, then set checkpoints. Week 1: did spend move and did tracking break? Week 3: cost per lead in the receiving campaign. Day 90: cost per job, with time for jobs to close.
How often to audit
A full twelve-check audit once, then a short monthly review: the receipt (cost per job per campaign against the baseline), lost impression share, new search-term waste, and what changed in the account. That's what our monthly report is. Re-run the full checklist after any big change: a new agency, a new website, a new location, or a budget change of 50% or more.
Questions people ask
How long does a Google Ads audit take?
With read-only access and the official API, pulling 90 days of data takes minutes. Reading it properly takes a few hours for a typical local account with 3–10 campaigns. The free SameSpend audit is delivered in 2–3 business days; most of that is writing it so a non-specialist can act on it.
Can I run this checklist myself?
Yes. Every check names the Google Ads screen or column. The parts that are hard to do by hand are matching spend to paid jobs across campaigns, measuring overlap across hundreds of search terms, and keeping a baseline so you can prove whether a change worked. Those are what our software does.
Do I need to pause anything while auditing?
No. An audit is read-only. Nothing should change until you've seen the evidence and chosen what to act on.
What's the minimum spend for an audit to be useful?
Roughly $1,500–$2,000 a month over 90 days gives enough clicks for the findings to be more than hints. Below that the checklist still works, but expect more 'early signal' and fewer conclusions.