I set up my first Google Ads campaign a couple of weeks ago — a small, deliberately narrow test, a handful of specific search terms, ten dollars a day. I used AI heavily throughout it, the same way I use it for most of what I do running Keel on my own. It genuinely helped. It also, more than once, nearly talked me into something that wasn't in my interest — and both of those things being true at once is the actual point of this post.
Two different things wearing the same name
"AI" gets talked about as one thing, but for a small operation there are really two, and they behave differently. One is a general-purpose assistant you bring to a task — it has no stake in what you decide, its only job is to help you think clearly. The other is a feature built into whatever platform you're using, and that feature was built by a company whose revenue depends on your behaviour. Not maliciously. Just aligned with a different set of interests than yours, by design.
Setting up that campaign, I ran into the second kind constantly. Google's own "AI Max" feature offered to auto-generate my ad copy — and the draft it produced overstated Keel's scale in places, and leaned on "AI-assisted OKR writing" as a unique differentiator, which directly contradicts an argument I'd already published. A "New Customer Acquisition" bidding option offered to help me avoid wasting budget on repeat customers — a real problem for a retailer, not one I have. An audience-targeting suggestion would have narrowed an already-thin pool of searches even further, in the name of "more relevant" reach.
What actually caught it
Every one of those got caught before it went live, and the thing that caught it was the first kind of AI — an assistant with no stake in whether I spent more or less, just asked to check whether a suggestion was actually true and actually useful for what I was trying to do. Did the copy it drafted actually describe Keel accurately? Not in a couple of places. Does the New Customer Acquisition feature solve a problem I have? No — I have zero converted customers from paid search to acquire twice. Would narrower audience targeting help or hurt a campaign that's already short on volume? Hurt, clearly, once you look at the actual numbers instead of the recommendation's confident framing.
None of that required deep technical expertise. It required treating each suggestion as a claim to verify, not a decision to accept — the same discipline I'd apply to anything else someone tried to sell me.
The actual lesson, and it isn't "be careful with AI"
The lesson isn't to distrust AI, and it isn't really about Google Ads either — the same pattern shows up anywhere a platform has a financial interest in your decisions: a bidding tool nudging toward automated spend, an e-commerce platform's "smart" upsell suggestions, a social platform's algorithm rewarding whatever keeps people scrolling longest, not whatever's actually good for your business. The useful question isn't "is this AI," it's "who built this, and what does it want me to do." A general-purpose assistant with no product to sell you is a genuinely different thing from a feature embedded in the product itself.
For a small operation without a marketing team or an agency retainer, that distinction is worth real money. Used well, AI is a legitimate advantage — a second, tireless set of eyes checking whether a "recommended" setting actually serves you. Used uncritically, especially the version built into the very platform you're trying to navigate, it's optimizing for someone else's outcome and simply hoping yours goes along for the ride.