Most agencies and automated audit software will run a 30-90 day lookback on an Ads account, run a checklist, and then produce a deck full of information that doesn't mean much. We don't work that way and haven't in some time. We have built a full AI-driven toolbox to help us better evaluate an account and look back at 6-12 months of search terms, Quality Scores, account change logs, conversion data, ad copy iterations, bid strategy shifts, and about 85 other points of relevant data.
The human element of how we do audits hasn’t changed. Our team still does a full manual review of accounts and does an actual analysis and feedback of what we, and our AI, actually saw before presenting it as fact. AI just allows us to go so much deeper and look at a long history of complex data to better understand the account.
Here are a couple of examples of what our team and AI tools have caught.
Geographic Waste, Calculated Properly
Take geographic waste. Most agencies handle it by eyeballing the search term report and adding cities to a negative keyword list. Someone searched for your service in a city outside your area, clicked your ad, and didn’t convert, so you block the city and move on. That’s the standard audit move.
We don't trust those reads at their face value. We calculate.
For every search term in a report that contains a place reference, our system identifies which campaign the term came from, pulls that campaign’s actual targeting radius and the office’s coordinates, geocodes the city the searcher mentioned, and calculates the real distance using the haversine formula. The haversine formula accounts for the curvature of the Earth when measuring distance between two points on a map. It sounds academic until you realize it’s the difference between “this city is 8 miles away from the office and probably fine” and “this city is 45 miles away and definitely not your customer.”
On a recent multi-location healthcare audit, the search term report had roughly $10,000 in annual spend sitting in entries that contained city names with no conversions attached. The eyeball method would have called all of it waste and moved on.
When we ran the haversine analysis, that $10,000 broke apart in three directions:
| Category | Annual Spend |
|---|---|
| Real geographic waste (cities outside the targeting radius) | $4,800 |
| Patient traffic the eyeball method wrongly flagged | $1,600 |
| Competitor names and insurance website lookups | $3,600 |
| Total flagged in surface analysis | $10,000 |
The most striking false positive was a small town that turned out to be the actual city where one of the offices sits. An agency working off the surface-level numbers would have negated real patient traffic alongside real waste, which is the worst kind of mistake an account can absorb.
The Patterns That Only Show Up at Scale
Different account, different vertical, same shape of problem. A national ecommerce retailer we audit had a search term report with hundreds of unique entries. Reading any individual term, nothing in particular jumped out. Most spent under $50 and plenty had no conversions, which is normal in ecommerce because you expect 90% or more of clicks not to convert.
Run those same terms through an N-gram analysis and the picture changes.
N-gram analysis breaks every search term into one-word, two-word, and three-word fragments and aggregates spend, clicks, and conversions for each fragment across the entire report. Brand names and product categories that are scattered across hundreds of individual search terms surface as concentrated patterns at the level they actually exist in the data, not the level any single search term shows them at.
The analysis surfaced roughly $14,800 a year in misallocated spend across three patterns no individual search term report made obvious. A specific brand name the retailer doesn’t usually carry showed up across 12 different terms with no conversions. A product category the retailer definitely doesn’t sell appeared as the primary phrase across 13 different search terms with no conversions either. And the costliest one was searches for clearance pricing on a category the retailer sells but doesn’t discount.
That clearance pattern was more than 2,300 clicks a year landing at a site that doesn’t run clearance sales. None of them converted. A casual reading of the search term report sees those clicks as 2,300 unconnected events. Our AI analysis surfaces them as one pattern with one fix.
Putting a Dollar Number on Quality Score
Back to the healthcare account.
Quality Score is Google’s per-keyword rating from 1 to 10, and it controls both how often your ad shows up in the auction and how much you pay per click when it does. Most accounts sit around QS 5, which is average. That’s the level where you pay what Google considers a fair CPC for your slot in the auction. Below 5 you pay a penalty on every click. Above 5 you get a discount.
The penalty math gets rough fast. A keyword sitting at QS 1 costs you roughly five times what it would at QS 5. QS 2 costs about two and a half times. QS 3 costs about one and two-thirds. The scaling is multiplicative, which means it doesn’t add a small surcharge to your click cost, it multiplies the entire amount.
Most CMOs or business owners have never seen what this looks like in dollars on their own account because nobody calculates it for them.
Our analysis pulls the Quality Score on every active keyword and runs the actual spend against the multiplier table. For each keyword, the calculation answers a simple question: what would that same spend have been if the keyword were at QS 5 instead of where it actually is? The difference between the two numbers is the Quality Score tax the account is paying.
For this healthcare account, the answer came out to over $22,000 a year in CPC overpay attributable to Quality Score alone. The bids weren’t wrong. The targeting wasn’t wrong. The offer wasn’t wrong. The keywords were just priced higher in the auction than they would be if their Quality Scores were sitting at average.
The fix isn’t bidding less on those keywords. The fix is moving the keywords up the QS ladder, which means real work on ad copy, landing page experience, and ad group structure. None of that work happens until somebody has done the math that justifies it in dollars.
What This Adds Up To
Three findings from recent audits, each one invisible to the standard 30 to 90-day checklist. The first kept a client from negating their own customer base across multiple offices. The second surfaced waste hiding inside hundreds of low-spend search terms that wouldn’t individually trigger any auditor’s threshold. The third gave a CMO a five-figure annual overpay number on a line item that nobody on the account team had ever bothered to calculate.
That’s what we mean when we say an AI-based PPC audit. It isn’t a chatbot reading your account and giving you generic suggestions. It’s our methodology, run across a full year of your account’s history at the row level instead of the summary level, with our team verifying every finding before any of it gets presented as fact. AI is what lets us actually read every row of every report instead of skimming the top of each one.
If your agency has never run any of these analyses on your own account, that’s a conversation worth having. We do these complete audits for our clients regularly and offer a free, no-pitch version of this Google Ads audit for those that need it. Then you can decide what to do with that information.
