Intrigue DigitalGoogle Ads in the age of AI Overviews: what changed, and what didn't.
AI Overviews took the questions and left the purchases. An ecommerce operator running 90+ brands on which searches are still worth paying for, and how to write pages an AI will quote.
I run an ecommerce group of 90+ brands across more than 800 specialist stores, selling mostly to people who arrive from Google. I'm not an agency and I don't manage anyone else's budget. What follows is what the last couple of years did to ours.
The click you used to pay for is sometimes answered before it happens
The honest version of this change is narrower than the headlines suggest. Searches where somebody wanted a fact have largely stopped being clicks. If the question is how often you should replace a thing, or what size suits a particular job, the answer now sits at the top of the page and the visit never happens. We used to get some of those visits. We don't any more.
What survived is the search where somebody has already decided to buy and is choosing between options. An AI summary can tell you what a product does. It can't put it in a box and send it to you.
So the first thing that changed for us wasn't cost per click. It was which searches were worth buying at all.
AI Overviews took the questions and left the purchases. That's survivable.
The discipline that didn't change: set the ceiling before you bid
I use something I call the Golden PPC Formula, and it's outlasted every platform change I've seen.
Say a $300 product. You decide you'll spend ten per cent of its price to win the sale, so $30. If one in twenty-five people who land on that page buys, your ceiling on a click is a bit over a dollar.
That number is the decision. Everything after it is management. Auction prices move and new ad formats arrive every year, and none of it changes what a single order is worth to you.
Most advertisers I meet work the other way round. They look at what clicks cost, then work out what they can bear. That's how you end up explaining an overspend instead of setting a limit.
The ceiling also tells you when to stop. If a product can't be won inside its number, stop advertising it. We'd rather lose the line than fund it out of something else's margin.
Aim at the searches an AI answer doesn't finish
We moved budget toward queries that carry a decision inside them. Best-of searches. Product-level specifics. Fit and compatibility questions. Comparisons between two named things. Somebody typing "will X fit a Y" has a wallet out and a problem to solve, and no summary paragraph ends that job.
We moved away from the broad informational terms we used to buy. They stopped arriving, so there was nothing left to convert.
That shift sounds small, and it changed which half of the account got the money.
Write the page for the machine that will quote it
Here's the part most advertisers underrate. The assistant answering your customer's question is reading something. If it's reading a competitor's page because yours is all adjectives, you lost before the auction opened.
So our product pages carry what a machine can repeat without guessing: what the product is, what it fits, what it costs, what it weighs, what buyers said about it, and the questions people actually ask before buying. We write those answers plainly and we write a lot of them. It's unglamorous work that doesn't feel like advertising, and it's the same job it always was: being the thing that gets recommended at the moment somebody chooses.
You're probably under-counting what AI already sends you
One practical warning. Traffic arriving by way of an assistant often doesn't label itself cleanly, and standard analytics setups file a good share of it as direct. We only noticed when we measured the same period two ways across our properties and the totals disagreed.
The exact gap matters less than the habit. If you're deciding whether AI referral is worth anything by glancing at a default report, you're deciding on an undercount.
What I haven't done
I haven't run ads inside an AI tool. ChatGPT ads are new, I've no results to report, and I'd rather say so than speculate. When we do test it, the first thing we'll work out is the same thing as always: what one order is worth, and what that makes a click worth.
The short version
AI Overviews took the questions and left the purchases. That's survivable. For anyone selling a specific thing to a decided buyer it might even be an improvement, because what remains is the traffic worth paying for.
What it punishes is the advertiser who never knew what an order was worth to them. They're the ones watching costs climb with no number to measure them against.
Featured in this article?
Add this badge to your website's About or Press page. Copy the code below and paste it into an HTML or Custom Code block (WordPress, Wix, Squarespace, Webflow and Shopify all have one).
Want to be the brand AI mentions?
Intrigue Digital does SEO, AEO and GEO as one system, so your business shows up whether someone searches Google or asks ChatGPT. One business per industry, per postcode.
Book a free strategy call →Business Spotlight features are contributed by operators in our network and reflect their own experience and views. Intrigue Digital does not verify the internal metrics or figures reported and they should be read as the author's own account, not independently audited results. This article is not financial advice.