Auditing 8 Google Ads Accounts: 1,369 Dead Ads and Three Accounts Tracking Nothing

People come to us with the same sentence, phrased slightly differently every time: "the advertising runs, the money goes out, and nobody can tell me what happens in between". The owner sees clicks and impressions in a report, sometimes conversions, but cannot explain why last month produced twice the enquiries for the same spend.

Advice from a distance is worthless in that situation. What is needed is access to the account and a few hours spent looking at how it is genuinely put together. We did exactly that for eight accounts at once and combined the results into a single picture.

What the accounts were

The sample was deliberately mixed, so we could see which problems repeat regardless of industry:

• E-commerce - three accounts running shopping campaigns and product feeds.
• Healthcare services - two accounts with local targeting, where phone calls are the main enquiry channel.
• Industrial products - two accounts with long sales cycles.
• B2B - one account with a narrow audience and high cost per click.

Across twelve months the totals came to 63.5 million impressions, 1.11 million clicks and 23,434 recorded conversions. The accounts varied in age from newly launched to more than four years old.

Why reading the interface does not work

The trouble with most self-checks is that people look wherever something glows red. Google Ads does not work that way: the most expensive problems are never highlighted at all. A campaign optimising towards the wrong goal looks perfectly healthy. An ad group holding a hundred and fifty keywords carries no warning. A conversion action that records nothing simply shows zero, which is easily mistaken for an absence of sales.

What we set out to establish

Before starting, we wrote down five questions the audit had to answer concretely:

1. Can the system see what actually counts as a result for this business?
2. Does the account structure match the way people phrase their searches?
3. How much residue is left from decisions taken years earlier?
4. Is the budget protected from queries that never convert?
5. Which of these should be fixed first?

The last question matters most. A list of thirty observations without priorities is useless: no business has the capacity to fix everything at once, and the impact of different corrections varies by an order of magnitude.

We run audits through the API rather than the interface. The reason is practical: the interface shows what Google considers important, while a data export shows everything, including the parts nobody thought to ask about. Here is how it went, step by step, and what turned up.

Step 1. Export rather than browse

For each account we pulled five slices of data covering twelve months:

• every enabled conversion action, with its type and primary status;
• every enabled campaign, with channel type, bidding strategy and network settings;
• negative keywords at campaign level;
• every enabled ad, with its format;
• ad groups and keywords with their match types.

That set gives a complete structural picture in ten minutes per account instead of two hours of clicking. The actual work begins afterwards, in the interpretation.

Step 2. What the conversions revealed

This turned out to be the most damaged area in the sample. Across eight accounts there were 73 enabled conversion actions, and 51 of them were marked as primary.

The primary flag determines where automated bidding steers a campaign. When ten goals carry that flag, the algorithm picks whichever occurs most often and costs least. In practice that is never the sale.

The starkest case: one account had clicks for driving directions and taps on the phone number in the business listing sitting among its primary goals. Both occur many times more often than a genuine enquiry. The algorithm dutifully steered the campaign towards them.

Step 3. Archaeology in the ads

Expanded text ads stopped serving in 2022, and plain text ads earlier still. In five of the eight accounts they remain marked as enabled.

Together that is 1,369 ads in formats that physically cannot appear, against 181 responsive search ads. Almost eight dead ads for every working one.

Step 4. Ad group structure and match types

The spread between accounts came to a factor of 88: from 1.7 keywords per ad group to 149.5. The median sits at 20.4, which is already the upper edge of sensible.

The logic is direct. Ad text should echo the query. With five related keywords in a group, the headline can be written to match them literally, quality signals improve and cost per click falls. With a hundred and forty-nine, the ad becomes generic by necessity.

We assessed match types together with negative keywords, because neither is meaningful on its own:

• The highest share of broad match in the sample was 86.7% (8,008 of 9,235 keywords) against 2,183 negatives.
• The lowest was 0.1%, where the question of keyword control does not arise at all.
• Negative keyword density varies just as sharply: 184 per thousand clicks in one account against 0.3 in another.

Step 5. Prioritisation

An audit without an order of work is just a list of complaints. We score every finding on two axes: how much money it costs right now, and how many hours the correction takes.

That order is deliberate. Correcting conversion goals produces an effect within days, because automated bidding immediately starts seeing the right target. Restructuring ad groups delivers more over the long run but takes an order of magnitude more time, so it follows the quick wins.

What the client receives

• A written report of findings, each pointing to a specific location in the account.
• A priority table scoring impact against effort.
• A split between what can be fixed internally and what needs a specialist.
• Baseline metrics recorded before any changes, so there is something to compare against in a month.

What this audit deliberately excludes

We did not evaluate ad copy, landing pages or creative. Those matter, but they are subjective and resist being reduced to figures. We did not calculate money lost to irrelevant queries either, because that requires manually reading the search terms report for every account rather than exporting through the API.

Eight accounts is not a statistical sample, and we do not present it as market research. But when the same problem appears in five accounts out of eight, it stops being coincidence and becomes a pattern worth checking for in your own account.