Two years ago, the three of us on my team handled three to four Google Ads accounts each. Today we handle six to seven. More revenue, more complexity, better results for our clients, and we still leave the office on time. Nobody new got hired. Well, nobody human.
I lead the paid search team at a German agency focused on explanation-heavy products: finance, insurance, and B2B lead generation. No e-commerce. Part of how we doubled the output is a new team member: it never sleeps, makes decisions without asking, and reads a client brief in eight seconds to write a better one than I would. Its name is Claude, the AI assistant I’ve spent the last year building into a second brain for the agency.
But this article isn’t about Claude.
It’s about my second non-human team member, the one I didn’t choose, don’t always like, and can’t fire. It changes the rules every quarter and tells me it’s for my own good.
Already know who I’m talking about? Yes – Google.
Google + Me isn’t a love story. It’s a roommate situation.
We didn’t pick each other, and we don’t always agree. But we share the rent, and I’ve stopped fighting it. My job now is to manage it.
Three shifts are happening at once, and together they explain why the old playbook stopped working.
How people search has changed: Yesterday someone typed “best ETF portfolio” into a search box. Today they speak a whole situation into their phone: “I’m 42, I have 50k in savings, I want to invest 1k a month at moderate risk and retire early. How should I allocate?” We’re not targeting keywords anymore. We’re targeting situations. And that context gets sent to Google, ChatGPT, and Perplexity all at once.
Where answers appear has changed: AI Overviews now show up on transactional queries, not just informational ones. “Best ETF for retirement.” “Compare business accounts.” The answer often arrives before the click can happen. And the click that does happen is worth more than ever, so your landing page and content have to carry far more weight. Paid and organic aren’t separate disciplines anymore. They’re the same conversation about whether you deserve the click.
Who decides has changed: Google now chooses which query matches, which ad shows, and which placement, bid, and asset combination wins. Performance Max, Broad Match, AI Max: three product names all pointing the same direction, which is to hand the wheel to the algorithm. Every quarter another lever disappears and another recommendation becomes the default.
So here we are: less control, more uncertainty, and the same ambitious targets as before.
Stop doing the work. Start managing the worker.
For years, we did the work ourselves. We moved the bids, wrote the ads, and picked the keywords. If it worked, we were the reason.
That job is shrinking. The new job is to manage the worker. Picture Google as your new direct report: fast, motivated, and processing more data in a second than you will in your career. Like any team member, it needs the same things from you: a clear briefing, real goals, regular check-ins, and someone who overrules it before it does something expensive.
If you’ve ever managed a person, you already know how to do this. Good managers do three things: they brief well, they verify the work, and they decide upfront what they’ll do if it succeeds and if it fails. The only difference now is that your direct report is an algorithm. Here’s how those habits translate into Google Ads.
Rule 1: Brief the machine well
A bad brief doesn’t produce a bad result. It produces a confidently bad result, with someone working very hard on entirely the wrong thing. The algorithm is exactly the same. It’s only as smart as the signals you feed it.
I audit an account or two every month, and most share the same broken foundation: conversions counting page visits, conversions double-counted, lead forms firing on every scroll, and soft and hard conversions mixed together with no values attached.
Three things Google actually needs from you:
- Conversion value. Most finance, insurance, and B2B advertisers don’t know the value at the moment of conversion. That’s not an excuse. It’s the problem to solve, with proxy values, lead scoring, and value rules.
- First-party data. Everything you know about customers that Google doesn’t: lifetime value, repeat behavior, and lead quality. Enhanced conversions and offline conversion imports are how you get it in.
- Offline outcomes. Whether a lead actually became a customer: closed, canceled, or returned.
One example from our world. We have an insurance client with 90 days from first click to closed deal. For years we optimized for lead form submissions, because form fills are easy to track. It worked in one sense: more leads, but not more closed deals. Google had no idea which leads were any good, so it kept chasing cheap form fills. Once we fed the closed-deal data back through offline conversions, with real values attached, Google finally understood what a good lead looks like and started finding more of them.
The lesson: if Google doesn’t know who your best customer is, it will happily optimize for your cheapest one. So before you touch a bid, fix the thing being bid on. Bid adjustments and budget shifts don’t matter if you’re measuring the wrong thing.
Rule 2: Always test, but test with structure
Performance Max is still largely a black box. The usual response is to make three changes at once and then have no idea which one worked. Every campaign you launch is a hypothesis, so treat it like one.
Before I launch any test, I answer four questions:
- What do I believe will happen?
- What result would prove me wrong?
- How long will I run it, and for how much?
- What changes on Monday?
Question four is the one everyone skips. If you can’t answer it, you don’t have a test. You have a wish. I failed at this for years. I once tested Broad Match on a top campaign, and six weeks in, CPA was up 30%. The honest call was to kill it. Instead my brain served up three excuses: a rate change hit demand, a competitor launched something, and the algorithm just needs more time. Pre-committing to the action fixed that, not because I got smarter, but because I took the decision away from future-me.
Then test the inputs, not the outputs. Match types, audience signals, value rules, feed slices, AI Max on or off: that’s where the algorithm listens. And treat AI Max as a reach feature, not an efficiency one. Expect the same CPA at higher volume and you’ll kill it in week two for the wrong reason. So: one test per campaign, one variable at a time. Change two things and you’ve learned nothing.
Rule 3: Decide with frameworks, not feelings.
This is the rule I’m still the worst at. Most weeks we react: the numbers come in, we look, we adjust. That works when targets are clear and the team is experienced, but it has a cost. We end up explaining the data instead of acting on it.
Google now makes a thousand small decisions a second so you don’t have to. The few big decisions left, which are scale, hold, and cut, matter more than ever. A simple frame for those:
- Scale when you’re hitting target and losing impression share to budget. Push budget and broaden signals.
- Hold when you’re hitting a target with no room to grow. Protect it and leave it alone. The most expensive habit in PPC is fiddling with a campaign that’s already working.
- Cut when you’re missing a target with no path back. Reallocate, without sentiment.
Most of us spend too much time tinkering with Hold campaigns and not enough time cutting the ones that don’t work. The fix is boring but effective: decide the rule before you look at the data.
On Monday morning, before you open any dashboard, decide what would make you scale, what would make you cut, and when you’ll do nothing on purpose. Then open the dashboard and follow your own rules. Put it in a workflow, so the decision exists before the data does. The point isn’t discipline. It’s taking the call away from a tired, pressured future-you.
You’re not the operator anymore. You’re the architect.
The operator job shrinks every quarter. The architect job grows. You now wear three hats. As signal architect, you decide what the machine learns from. As test designer, you ask the questions the dashboard can’t. As decision maker, you make the calls the algorithm isn’t allowed to.
Anything that can be automated in the next few years probably will be, so concentrate on the part of the job that can’t. The manual work Google automated, like keyword expansion, bid tweaks, first-draft copy, and weekly reports, is exactly what used to eat the day. What opens up is the work that moves the needle: understanding what makes a good lead versus a cheap one, auditing landing pages now that the click is worth more, and improving the offer and the message itself.
AI didn’t take our job. It took the parts that were never really ours to own, the ones we did because nobody had automated them yet, not because that’s where we added value.
So next Monday: fix one signal, ship one test with a kill date, and write one rule down. Your non-human team members need a good manager. That’s you.
