There’s an old line about fish not being able to describe water. They’re in it, so they can’t see it.

I’ve watched two decades of internet marketing prove that about the default settings.

Every platform ships with a set of recommendations. They’re right there in the account, with a little score telling you how compliant you’re being. Apply them all and the score goes green. Nobody has ever been fired for a green score. And so an entire profession grew up around fluency in someone else’s recommendations, and started calling it strategy.

I’m not being clever about other people here. It’s the water I swam in too.

What actually changed

For most of my career, agreeing with the platform still left you holding decisions.

Which keywords. Which match types. Which ad against which query, which landing page against which intent, which searches you’d pay for and which ones you’d let go by. You could take every one of Google’s suggestions and still be doing the work, because the suggestions came at you one at a time and each one had a door you could close.

The newer campaign types close the doors for you. Hand over a budget, some assets, and a conversion goal, and the system decides the audience, the query, the creative, the placement, and the moment. That’s the pitch, and I want to be fair to it: for a lot of advertisers, that pitch is honest and the results are real.

But notice what moved. Agreeing with the platform used to be a posture. Now it’s the architecture. There’s no longer a door to close, because there’s no longer a door.

Nets and lines

A trawler drags a net through the water and keeps what comes up. It’s the most efficient way to catch fish ever invented, as long as you’ll eat whatever you catch.

That’s what a fully automated campaign is, and it’s genuinely good at it. If you sell something people buy quickly, at a price they don’t agonize over, in volume, then the net is exactly right. Pull it in, sort it, sell it.

Now picture the manufacturer I usually work with. Long sales cycle. Technical buyer. A purchase somebody has to justify to a committee. The order that matters this quarter comes from one plant engineer in one industry with one unusual tolerance requirement, and there might be forty of him in the country.

You don’t catch that fish with a net. You catch him with a line, in a spot you know, with bait you chose on purpose.

Drop a net for him and here’s what you get: bycatch. A boat full of form fills from hobbyists, students, tire-kickers, and competitors, plus one guy who wanted the desktop version at a tenth of the price. Your cost-per-lead chart will look wonderful. Your sales team will quietly stop returning the lead list.

I’ve watched that exact thing happen more than once, and the report always looks like a win right up until somebody asks what closed.

Why the average isn’t a bug

Here’s the part I think gets misread as cynicism. It isn’t. It’s just how the machinery works.

These systems optimize toward where the data is thickest — the broad middle of the distribution, the patterns that repeat, the behavior that looks like most other behavior. That’s not a flaw in the model. That’s what a model is. It finds the center and it swims there.

So the platform isn’t failing you when it averages you. Averaging is the product. And for most of the market that’s a genuinely good deal, which is exactly why it’s sold so hard.

The question is whether you’re most of the market.

Because if your business is ordinary — and plenty of profitable businesses are, with no shame in it — being efficiently ordinary is a fine outcome. But if you’ve spent thirty years becoming the shop that handles the alloy nobody else will touch, then the strange, narrow, specific thing about you is the margin. It’s the whole reason anyone pays you more than the cheapest quote.

An averaging system will file that down. Not maliciously. It’ll just keep finding you more of the common customer, because the common customer is what it can see, and the specific one is a rounding error in its training. Feed it long enough and your marketing describes a company slightly more generic than the one you actually run.

That’s the race to average. It’s a fine race to enter, if average is a place you can afford to finish.

The part that gets lost first

There’s a specific casualty here and I want to name it, because it’s the one my clients feel and can’t always articulate.

Some things have to be taught before they can be sold.

A buyer who doesn’t yet know that his real problem is fixturing, not the machine, cannot search for the answer — he doesn’t have the words yet. Somebody has to teach him. That’s a piece of content, a conversation, a specific person patiently explaining a specific thing, and it may be three touches before he’s even ready to have a price conversation.

Automated optimization is blind to that work. It optimizes toward the click that converts today, which means it will systematically defund the teaching that creates buyers eighteen months from now. You’ll never see it in the report. You just slowly stop being the company people learn from — and being the company people learn from was your moat.

Give a man a fish and you’ll close him this quarter. Teach him to fish and he’ll buy his equipment from you for twenty years. Only one of those shows up in a conversion column.

I’m not against any of this

I should be plain, since this reads like a man yelling at a river.

I use AI constantly. It has changed my work more in the last year than anything in the previous ten, and I’ve written elsewhere about how much I’ve been able to build with it. I’m not nostalgic for manual bid adjustments and I don’t want to go back.

The argument isn’t about the technology. It’s about which way you’re pointing it.

Point it at your own capacity and it multiplies what you know — you keep the judgment, the machine carries the load. Point it at your decisions and it takes the judgment and hands you back a very efficient average.

Same tool. Opposite direction. The whole disagreement fits in that sentence.

What I’d ask, if it were my money

Not exciting questions. That’s usually the tell.

1. If we turned the automated campaign off tomorrow, would we still know who our best customer is — or does only the platform know?
2. Which of last quarter’s leads actually became quotes? Not leads. Quotes.
3. What are we teaching this year that nobody can search for yet?
4. Can we say, in a sentence, what we’re deliberately choosing not to advertise for?
5. Who on our side is still making decisions, and could they explain the last one to a skeptical person?

If the honest answer to any of those is “the system handles that,” you’ve found the water.

Don’t let the river steer

A current is a wonderful thing when it’s going where you’re going.

Everything about these tools is a current — the defaults, the recommendations, the green score, the whole industry agreeing at once. Stop paddling and you’ll still move. You’ll move at exactly the speed and in exactly the direction as everybody else who stopped paddling.

For a commodity seller, that’s the trip. Go with it, save the effort.

For everybody else: the reason a customer chooses you is the thing the current is slowly rubbing off. Keep a hand on the tiller. Keep somebody on your team who can still read the water.

And before you let a platform make one more decision for you, ask what it can’t see — and then go look at that yourself.

Related: The Dark Side of AI and Automation in B2B Manufacturing — the same argument from the operations side, where the efficiency itself becomes the liability.