How to Audit a Google Ads Account: Eight Checks, Tracking First

A sample account, and the finding that was not about waste

Harbour & Finch LLP is a Toronto employment and injury law firm that does not exist. Its Google Ads account is the sample behind this guide: three campaigns, CA$21,727 of spend in the 90 days from 5 August to 2 November 2026, 1,846 clicks, 94 conversions and 412 search terms. The audit of that account is published in full as a 12-page sample report, with its action plan and negatives file, so every figure below can be checked against the page it came from.

The audit found CA$5,571 spent on 38 searches that never booked a consultation. It is the obvious headline. The finding that changes more is on page 6 of the report: two conversion actions were counting the same phone calls, so the account may have been reporting up to 29 more conversions than it had, and both of its main campaigns were bidding toward that count. Each later check reads the Conversions column, which is why the column gets checked first.

The checks, in the order they depend on each other:

1. Conversions. List what counts, and find anything counted twice.
2. The spend. Split it into what converted, what never did, and what cannot be judged.
3. Search terms. Group the waste by what the searcher wanted.
4. Negatives. Write them with match types, and test each against the searches that convert.
5. Location. Check whether the ads reach people outside the area the business serves.
6. Match types and recommendations. Find where the waste got in.
7. Quality Score. Split it into its three parts and send each to whoever can fix it.
8. Paid vs organic. Find the searches the account pays for and already ranks for.

Then every fix becomes a task with an owner and a check, and the checks run again 30 days later. Besides the account itself, the audit needs three exports: the search terms report, the keyword report with its Quality Score columns, and a Search Console queries export for the same dates.

1. Start with what the account counts as a conversion

In Google Ads, the Summary page under Goals and then Conversions lists every conversion action. What matters for an audit is which ones are primary. Primary actions are reported in the Conversions column and used for bidding; secondary actions appear only in All conversions, according to Google Ads Help. For each primary action, ask three questions. Does it record a lead or a sale, rather than a page view? Does it count one conversion per click, which Google calls a good choice when what matters is whether a lead came in, not how many times? And does any other primary action record the same event?

Phone calls are the easiest place for the third question to fail. Google Ads has five sources for call conversions. Two of them sit on the website: "Calls to a phone number on your website", which swaps in a Google forwarding number and counts the call, and "Clicks on a number on your mobile website", which counts the tap. One person who taps the number on the site and then talks to the firm can be counted by both.

The way to see it is to segment the search terms report by conversions and pick Conversion action, one of the conversion segments Google lists. Each search's conversions are then split by the action that recorded them. Cost and clicks show as two dashes in the split rows, because the segment applies only to conversion columns. In the sample, "Calls from website" and "Phone number clicks" were both primary. On 33 of the 41 days with a call, the two moved together on the same search term. Up to 29 of the 94 conversions may be one call counted twice.

Both campaigns bid with Maximize conversions, which steers toward whatever the primary actions count, so the bidder was chasing the duplicates too. The fix is one setting: make "Phone number clicks" secondary, then check over the next 30 days that conversions roughly agree with the calls the front desk logged.

Three more things to settle before moving on. If the account spent money and recorded no conversions at all, stop there: tracking is broken, and blocking searches on that evidence switches off the ones that work, as our guide to finding wasted spend explains. Google counts conversions on the date of the click, not the date of the conversion, so the last week or two of any window are still filling in; leave a gap before you download, or treat late clicks with care. And a booked consultation is not a signed client. With enhanced conversions for leads, which Google now suggests over offline conversion import for new setups, an advertiser can send back what happened to a lead later, such as a signed contract, and Google matches it to the ad click. Without that, an audit judges searches by consultations booked, and the report should say so.

2. Split the spend three ways before calling any of it waste

Every dollar in the window did one of three things. It bought a conversion, it never did, or it cannot be judged from the data. Reported on its own, the middle figure looks larger and more certain than it is.

The third pile has two parts. CA$1,604, 7.4% of the total, is spend Google reports without the search behind it, because the export leaves out searches with too little activity, for privacy. Nothing can judge that money, and some of it may be wasted too, so any waste figure is a floor. The other CA$864 went on 313 searches with no conversion and fewer than eight clicks, each one a search a client could plausibly make. The report leaves them alone.

Eight clicks is a reason to look, not proof. The sample converts 5.1% of its clicks. A search exactly that good records no conversion on eight clicks about two times in three (66%), and on 17 clicks still about two times in five (41%). So the click count is the weaker test, and the sample barely leans on it: only 11 of the 38 wasted searches had eight clicks or more. The other 27 are on the list for what they say, such as a job, a form or another province. The plausible ones with enough clicks to notice, like "severance calculator ontario" at 17 clicks and CA$402, carry a Review mark rather than a plain recommendation.

3. Group the waste by what the searcher wanted

A list of 38 searches is hard to act on. Five groups, each with its own reason, can be decided one at a time. Read every search and ask what the person who typed it wanted:

Each group points to a different fix, which is the point of grouping. Free help and jobs need negatives and nothing else. Other cities need negatives and a setting, which is check 5. Calculators and templates need a decision from the owner before anyone touches the account. Someone pricing their severance may be weeks from needing a lawyer, so the firm can block those searches or build a calculator page that ends with an offer of a free first call. The report sets that out as a choice instead of making it.

4. Write the negatives, then test each one against what works

Negative keywords match literally. Google Ads Help says they "won't match to close variants or other expansions", so an exact negative blocks one search, and a phrase negative blocks every search that contains its words in that order. Our guide to wasted spend and negative keywords covers the match types and where each negative should live. For an audit, one rule matters most: before a negative goes in, run it against every search that converted in the window.

The tempting shortcut in the sample is a phrase negative on "free", since 6 of the wasted searches came from people looking for free help. It would also have blocked "employment lawyer free consultation" and "free consultation wrongful dismissal", which booked 8 consultations between them. The firm offers a free first consultation, so those people are its buyers. The report adds the free-help searches one by one instead.

Two more were narrowed for the same reason. "ontario labour board phone number" goes in as an exact negative, because a phrase negative on "phone number" would block people looking for the firm's own number. "workers compensation claim form" is exact as well, because a workers' compensation claim gone wrong is a case the firm takes.

The ten searches that converted most booked 62 of the 94 conversions, and none of the 38 negatives touches them. That sentence belongs in every audit, with the check behind it, because a negative that blocks a working search costs more than the waste it removes. The sample's negatives come as a CSV, sorted by spend, with the campaign and match type of each.

5. Check who the ads are allowed to reach

Location targeting has two options, and the default is the wider one. Google's help page calls "Presence or Interest" the default and recommended option, and describes it as reaching people in, regularly in, or who've shown interest in the locations you target. "Presence" reaches people in, or regularly in, those locations.

Both non-brand campaigns in the sample used the default. Nine searches for lawyers in other cities cost CA$1,388 in 90 days, and three of the cities, Vancouver, Calgary and Montreal, are outside the province the firm works in.

The setting and the negatives do different jobs, so the audit recommends both. Presence stops people elsewhere who have only shown an interest in Toronto. It does nothing about someone in Toronto who searches for a lawyer in Hamilton, because that person is in the target area; only a negative stops that click. The city negatives block the nine cities already seen, and the setting stops many of the next ones before they cost anything.

6. Find where the waste got in

The search terms report also shows the match type of the keyword that triggered each search. In the sample, 37 of the 38 wasted searches came through broad match keywords, and one, "legal aid", through phrase match. Broad match is built to reach searches related to a keyword as well as those containing it, which is how a firm bidding on employment lawyers ends up paying for "legal assistant jobs". Google's help page on match types says "it's critical to use Smart Bidding with broad match", and these campaigns did. But Smart Bidding learns from conversions, and this account's were partly duplicated.

Before narrowing match types, check which converting searches came in through broad keywords, because switching those keywords to phrase removes the good searches along with the bad. The safer order is the one this guide follows: fix the conversion count, add the negatives, and look at match types again once the search terms report has a month of clean data.

While you are in the account, look at auto-applied recommendations. Advertisers can let Google apply some types of recommendation automatically, among them "Add keywords" and "Add broad match keywords", so an account can end up with keywords nobody chose. Check which types are switched on, and whether anyone meant to switch them on.

7. Split Quality Score into its three parts and send each to whoever can fix it

Google gives each keyword a Quality Score and calls it a diagnostic tool: "Quality Score is not an input in the ad auction". Its three parts are expected clickthrough rate, ad relevance and landing page experience, each rated above average, average or below average. In an audit, read the parts, not the score, because they have different owners. Whoever writes the ads can change expected clickthrough rate and ad relevance. Only whoever runs the website can change landing page experience. Sort by spend rather than by score: a 3 on a keyword that spent CA$40 matters less than a 5 on one that spent CA$900.

In the sample, four keywords with scores of 3 to 5 spent CA$3,140, and all four had a landing page experience rated below average. Three of them send people to /wrongful-dismissal, where on a phone the first screen is a photo and a heading, and the phone number sits below it. That is CA$2,228 of spend behind one fix for a web developer: put the number, as a tap-to-call link, above the photo. The fourth, "workplace harassment lawyer", is also rated below average on ad relevance, because it sits in an ad group whose ads never mention harassment. That fix belongs to whoever writes the ads: an ad group of its own, with ads that use the word.

8. Find the searches you pay for and already rank for

Join the search terms report to a Search Console queries export for the same dates, and look for searches where the site already ranks in Google's top three. The sample had six. Four were the firm's own name: CA$1,411 in 90 days through a brand campaign, where the site ranks first. Two were questions the firm's guides answer, "severance pay ontario" at an average position of 2.4 and "notice period ontario employment" at 2.9, which cost CA$486 and CA$319 and booked one conversion each.

None of that is a saving yet. When Google's researchers analysed 390 experiments in which advertisers paused their search ads, half of the ad clicks were incremental, on average, when the advertiser already held the top organic position: they disappeared instead of moving to the free listing. That is an average across advertisers, and yours can sit on either side of it. So the audit writes these as tests. Pause the brand campaign for four weeks and add paid and organic brand clicks together. If the total holds, the ads were buying clicks the site would have had anyway; if a rival starts bidding on the name, switch them back on. Our guide to bidding on keywords you already rank for shows how to design a pause that a small account can read.

The sample report puts the verdict on page 2 and the twelve tasks on page 9.

Write every fix as a task with an owner and a check

The sample's twelve fixes are written as tasks, each with the change as an instruction, the evidence with its figure, where it goes, who does it, and how to tell it worked. They run in the order to do them: account changes that take minutes, then the website, then the one test. Here is task 8, the conversion fix from check 1, as it appears in the action plan file:

## 8. Make "Phone number clicks" a secondary conversion action
Why: "Calls from website" and "Phone number clicks" both count as conversions and record the same calls: on 33 of the 41 days with a call, they moved together on the same search term. Up to 29 of the 94 conversions may be one call counted twice, and both campaigns bid to that number.
Where: Goals, Conversions, Summary. Open "Phone number clicks" and set it to Secondary.
Who: Whoever manages the ads
Check: For the next 30 days, conversions and calls answered at the front desk should roughly agree.

Each task can be pasted into Claude on its own, handed to an agency, or checked by the owner without reading the rest of the report. If a task has no evidence in the report, it does not belong in the plan.

Run the checks again in 30 days

The sample ends with four checks for the month after the fixes go in:

• The 38 searches are gone. None should appear with spend in the next search terms report.
• What works is untouched. The ten best searches should keep their clicks and conversions. If one drops, a negative is catching it.
• The calls add up. With one call action left as primary, conversions should come close to the calls the front desk logged.
• Google rates the pages better. Four weeks after the phone number moves up, the landing page column for the four keywords should improve.

Judge cost per conversion the same way: the weeks after the changes against the same weeks before, and against a campaign you did not touch, so a seasonal swing is not credited to the audit. Our guide to proving an SEO change worked sets out the control group method for organic search, and the logic carries over. Two limits belong in the report too: the CA$1,604 Google shows without a search term cannot be judged by anyone, and the conversions are consultations booked, not clients signed.

Everything above can be done with the account and the three exports, and the free Ads leak check does the search terms part in your browser. If you would rather hand it over, our Google Ads audit runs these checks on the same exports for $79 and sends the report and action plan within 8 hours.

Questions