Earlier this year, two of our clients told us they were leaving for an "AI agency."
Not a competitor. Not a cheaper freelancer. An agency whose whole pitch was that AI did the work, so it cost less. I didn't fight either one. I said what I always say: I get it, go see for yourself, the door's open.
One of them is already back. The other I'm still giving room — no hard feelings, these things take the time they take.
I'm not telling you this to gloat. I'm telling you because the most common thing I hear right now is some version of "can't AI just do this?" And the honest answer is: we use AI more than most of the agencies bragging about it. We just don't confuse using it with handing it the keys.
What we actually use it for
I'll be straight about this without giving away the parts that took us years to build.
AI is all over our day. It drafts the first version of a weekly client update so the strategist can spend their time on the thinking instead of the typing. It chews through a 40-minute call recording and hands back the five things that actually mattered. It spots patterns in a messy account export faster than I can scroll. It runs first-pass QA against a checklist so nothing dumb slips through. It turns a wall of search terms into something a human can actually make decisions from.
Notice what's not on that list: deciding what to do.
Every one of those is grunt work — the stuff that used to eat the hours our experts should be spending on judgment. AI gave us those hours back. That's the real win, and it's a big one. But "write the draft" and "decide the strategy" are two completely different jobs, and the second one is the one you're actually paying for.
The day AI built a Google Ads strategy
Here's the example that made this click for me.
My husband runs a business. He's brilliant at what he does — sharp guy, builds things, figures stuff out. Paid search just isn't his world (it's mine). One night he got curious and asked AI to build him a full Google Ads strategy for his business. Him prompting it. Not me. I just watched.
What came back looked incredible. Pages of it. Campaign structure, audiences, a keyword plan, ad copy, the works. If you didn't know what you were looking at, you'd think you just saved yourself a few thousand dollars.
I knew what I was looking at. So here's what was actually in there:
- Max CPCs set higher than the daily budget. Read that again. The bids were set so high that a single click could blow the whole day's budget — which means the campaigns would barely serve, if at all. Dead on arrival.
- A structure built for a Fortune 500 budget. Eight campaigns, dozens of tightly themed ad groups, for a budget that should've been one or two campaigns kept simple. It would've spread the spend so thin nothing ever got enough data to learn.
- Negative keywords that blocked his actual buyers. Some of the "negatives" it added were terms his real customers search. It was politely filtering out the people most likely to convert.
- Conversion tracking it just assumed existed. The whole plan leaned on signals that weren't set up and it never flagged that.
Any one of those kills the account. It had all four. And it delivered every word of it with total confidence — no hedging, no "you may want to check this with someone." That's the part that should scare you, honestly. It's not that it was wrong. It's that it was wrong and sure.
A non-expert reads that plan and sees a thorough, professional document. An expert reads it and sees a budget that would've lit money on fire for a month before anyone noticed.
AI is only as good as its prompt — and the person reading it back
This is the whole thing, so I'll say it plainly.
The value was never the typing. It's knowing what to ask, what to constrain, and what "good" even looks like when it comes back. I'd prompt that same tool completely differently — I'd hand it the real budget, the real margins, the platforms to avoid, the conversion setup we actually have, the stuff that's burned us before. And then I'd still red-pen every line, because it will still be confidently wrong about something. It always is.
That's not a knock on AI. I love the tools. They make my team faster and our clients better served. But "fast" and "right" aren't the same word, and the gap between them is exactly the job.
My husband's a smart guy and the AI still handed him a plan that would've never served a single ad. That's not an AI problem. That's a "this needs an expert holding the prompt and the red pen" problem.
The client who came back didn't need a long conversation about it. They'd seen the difference themselves. That was enough.
We'll keep using AI everywhere it earns its place. We just won't ever email you its first draft and call it strategy.