Pull up a Google search on your phone and count what sits above the first organic result. Up first, a block of ads.
Then the AI Overview, tall enough to fill most of the screen and built to answer the question before you scroll.
That stack is most of the story behind what drove up paid search search costs this year.
Advertisers who used to compete for the top of a page now compete for a narrower strip of it, and the auction has done what auctions do when inventory tightens: skyrocket.
Across the paid media accounts we manage in software, financial services and destination marketing, most campaigns are up 5 to 10% over the last seven months. The climb is gradual enough to miss on a monthly report and steep enough to reshape what a lead costs.
One account makes it plain. Same campaign, same keyword list, same landing pages, running for more than a year. Average cost per click used to sit between nine and ten dollars.
It’s fifteen now.
Fifty percent more for the same position, the same audience and the same work, on a page that gives that position less room than it had a year ago. That part is straightforward. The harder question is what you are buying when they win that click.
Less Room at the Top
That shrinking strip at the top of the page is the engine behind all of it, and the costs are only the first place it shows up. The number itself isn’t what’s new. Cost per click has drifted up for years, and anyone who runs search has a story about a rough quarter. What’s new is how the number behaves.
Cost per lead used to swing on us. It might double one month and drop back the next, with nothing we changed to explain either direction. That was the ordinary volatility of the channel. Over the past year, it stopped swinging. Now it climbs, and it stays.
That’s not the same problem. A number that keeps climbing is different than one that swings and settles. It’s the clearest sign we have that the shift at the top of the page is structural rather than seasonal, and it’s why “costs always go up” as a reason doesn’t cut it this time.
You can always buy your way back to the top. Google will happily take an absolute-top-of-page bid from anyone who wants one. For the accounts we’ve run the math on, that’s about fifty dollars a click.
“More views, fewer clicks, a higher price on every one that lands.”
Nobody Can Tell You What You’re Buying
Impressions went up over the same stretch. Ads served more often, not less, and impression growth outran spend growth. More people saw the ad and the cost to show it didn’t climb as fast, which sounds like the one piece of good news in the whole story.
It isn’t, because you don’t pay for impressions, you pay for clicks. Here’s the chain that turns rising impressions into a rising bill:
- The AI answer sits at the top, so fewer people scroll down and click through.
- Google serves the ad more often to make up for the lost clicks.
- The clicks that do come through cost more.
More views, fewer clicks, a higher price on every one that lands. That’s the no-click internet they’re inventing, and advertisers are paying to be part of it.
What even counts as an impression now
When a report says the ad was seen, what was actually served?
- A sponsored slot sitting above the AI Overview.
- A placement woven into the Overview itself.
- A standard result three screens down that nobody laid eyes on.
The reporting doesn’t separate them. We honestly can’t tell you which one you paid for, and that’s the problem. The number you’d use to diagnose all of this has quietly gone soft.
Part of the reason is Google still working it out in public. They’ve sketched how ads show up alongside AI results, but the operational detail an account manager needs hasn’t arrived. Reps have hinted, off the record, that you may need AI features switched on to appear in those results at all. Take it as a hint, because a hint is all it is. Either way, advertisers are making budget calls on incomplete information, and that gap is the real friction, more than any single cost line.
None of which means every lost click hurts. Plenty of those searches were never going to convert anyway, like the ‘what is an ad ops platform’ queries from people who were never going to buy one. The AI Overview is welcome to them. The real problem is that the same shift clearing out the junk is repricing the good clicks right along with it.
Why We Haven’t Flipped On AI Max
Google’s answer to a lot of this is AI Max, the feature suite that lets the system generate keywords, write copy and pick landing pages on its own. It works, and for a high-volume consumer account it works well. We wrote a full breakdown of how it works and where it fits, so we won’t rehash the mechanics here.
What it does is loosen the reins on which searches trigger your ads.
For a business selling something specific to a narrow buyer, that’s the one control you can’t afford to give up. Open the gates and the system chases every search that looks even loosely related, and you pay for a swath of an audience you already know is wrong.
Take a financial planning software company. It’s not for you or me, it’s for your financial advisor to use. Turn the keyword generation loose, and it starts bidding on every search that looks financial, and most of those are consumers who will never buy enterprise software.
You pay for those clicks from a budget you fought to protect, while the system works out they were the wrong ones. That’s the part that gives us pause: spending a client’s money to teach Google who their customer isn’t.
The tuition isn’t cheap, either. AI Max needs a learning period, somewhere between two weeks and a month, before it settles into anything useful. On a six-thousand-dollar monthly budget, that’s a big slice of the year handed over to the algorithm’s education, with no guarantee of results on the other side. Accounts with room to eat a rough month are where we test it first.
“That’s the part that gives us pause: spending a client’s money to teach Google who their customer isn’t.”
For all that, the honest read is that it hasn’t been as detrimental as we expected.
When we’ve opened it up on the right account, it came in softer than the worst case, and the results were often fine. AI Max isn’t dangerous. It’s a poor fit for a lean account that lives or dies on precision, and those are the ones we guard hardest.
Control the Inputs You Still Own
The industry is moving toward automation, and there’s a decent chance opting out of AI features won’t be a choice for much longer. So the goal isn’t to hold automation off forever. It’s to get your account clean enough that when you do hand the system the wheel, it’s working within boundaries you set on purpose.
- Build the negative keyword base first. Every irrelevant search you exclude now is one the automation can’t chase later on your dime. A tight negative list is the guardrail that makes turning on AI features survivable instead of reckless.
- Question the campaign, not just the keywords. Sometimes the fix isn’t a better bid, it’s admitting you’re advertising the wrong thing. We’ve pulled a client back from pushing hard on one part of the business and leaned into another that was actually converting, and that one call did more than a month of keyword tuning.
- Bid toward the people who buy, and away from the ones who don’t. We’ll push bids up around 25% on searchers who match the real buyer and drag them down on the traffic that shows up, pokes around and leaves. There’s always a pile of that. No sense paying full freight for it.
None of this stops costs from going up. They’re going up, period, and we tell clients as much.
What it changes is the quality of what you get back. On the accounts where we’ve kept up this work, the leads come through noticeably better even as the cost per click keeps climbing. Better leads, off a page that’s fighting you the whole way.
What to Do Monday
Pull cost per click and cost per lead for the last seven months and set them next to the same stretch last year. Then look at your negative keyword list and ask how much of it you built before AI Overviews took over the top of the page. If the honest answer is “most of it,” you’re running a defense you designed for a search page that no longer exists.
Most people reading this will land in a meeting explaining a cost increase they didn’t cause. That conversation goes better when you separate the two things happening: the auction got more expensive because the page changed, and your account did whatever it did inside that. Run them together and good work takes the blame for the chart.
We keep saying we’re trying to combat this, but honestly, combat isn’t the word. You don’t beat this. The page isn’t changing back, and Google is still working out how AI search pays for itself, so the rules are going to keep moving while they do. What you can do is stay close to it, keep your inputs clean and put the money on the people who actually convert.
If you want a team that does this every day, let’s talk.
