What Google Ads automation looks like when it runs a real account

Search for Google Ads automation and you mostly find two kinds of pages. Tool roundups written by people who have clearly never run an account, and agency pitches where the automation turns out to be a monthly meeting with slides. This post is neither. It is a walk through what our system actually did inside real accounts, taken from its own logs, with the numbers it wrote down at the time it acted.

Quick context so the numbers make sense. Attribution is software that operates Google Ads accounts on its own. It reads the account, decides, acts, and records why. The owner sleeps through most of it. That last detail matters more than it sounds, and you will see why.

The nightly reading nobody wants to do

Every account produces a stream of search terms, the literal phrases people typed before clicking your ad. Almost nobody reads them. There are too many, and most are noise. So the waste hides in plain sight, a few cents at a time, for years.

Our system has judged 112,516 search terms one by one so far. Not with a blocklist of dirty words, each term judged against what the account actually sells. Out of those, 9,204 were blocked for spending money without ever selling anything. People call this automated negative keywords. Internally we just call it reading.

One entry deserves its own paragraph, from an account that sells car tires. The search was "pneu dianteiro 100 80 r17", a front tire in the 100/80 R17 measure. To a blocklist, to a rules engine, honestly to most people skimming a report, that term looks like a customer. The system blocked it and wrote the reason next to the action: 100/80 is a width that only exists on motorcycle tires, and this store sells tires for cars, pickups and SUVs. In the same account it left "pneu 225 60 r18" alone, an SUV measure, numbers that look identical to anyone who does not know tires. The obvious junk, "free", "used", the wrong cities, any careful advertiser negates on day one. The money leaks through the terms that look right.

Adding keywords is easy. Refusing them is the job

Google will happily suggest thousands of new keywords, and accepting them is one click. That is how accounts drown. Our system considered 279,534 keyword ideas across accounts and refused 96.6% of them. The refusals never show up in a report, and honestly, they are the product. Discipline is invisible until you lose it.

The ones that survive earn their place. In one account, 714 keywords added by the system, with no human asking for any of them, now drive 64% of all sales, at a cost per sale 18% lower than the account average. Those two numbers together are the whole argument: it is not about adding volume, it is about adding buyers.

Pausing is easy. Unpausing is the honest part

Any script can pause an underperforming ad. Ours paused 2,340 of them. The number we are prouder of is different: 875 came back. Not because the pauses were mistakes. An auction never stands still: costs drift, active ads saturate, and an ad paused months ago can quietly become the best cost per sale in its group again. The system keeps reading the paused ads against the live ones, and when the comparison flips, it acts on that too, in its own log.

My favorite entry in that log is a single ad that returned to the account with no human involved, because its replacement was converting 10% worse. The system had made a call, watched the consequence, and admitted the original was better. Most human account managers never re-audit their own pauses. There is no glory in it. Software does not need glory.

By the way, none of what you have read so far involved an approval email. This is what separates real Google Ads automation from tools that generate suggestions for you to review: here, reviewing is not your job. The system executes directly, and the log plus the reversibility are the contract that makes that acceptable.

Is Google Ads automation safe for a small account?

Fair question, and the answer depends entirely on what the automation does when data is thin. Thin data plus aggressive rules is how small accounts get wrecked. The approach that works is the opposite: the smaller the account, the more patience per decision.

A small example. One account had a quiet little theme running: 23 clicks that turned into 11 sales. Tiny numbers. But that ratio is loud, so the theme was promoted into its own ad group with its own budget, where it could grow without competing for attention. Nobody had noticed it. At that scale, nobody ever does.

Small accounts do not need less judgment. They need more of it, applied to smaller signals.

What it will not do

It will not rescue an offer nobody wants. It will not fix a checkout that fails on mobile. And it will not produce a miracle in week one, because the first days are mostly reading, and reading takes the time it takes.

What it will do is show up every night, read everything, refuse almost everything, act on the rest, and write down the reason and its confidence for every single move. Every action reversible, every action explained. That is what Google Ads automation looks like when the word means something. It can start reading your account tonight.