A proposal crossed my desk this month: a little online shop, a modest budget, and an agency quoting the going monthly rate to run their Google Ads. Standard stuff. I’ve read a hundred of these. So I started typing the standard reply, and about two paragraphs in I stopped, because I realized I was about to charge somebody’s decision to a playbook that no longer exists.
Here’s what I actually wrote back, more or less: the fee you’re quoted is the kind of fee you pay when the playbook is mature and everyone knows what works. Right now nobody knows what works. So what you’re really buying is someone willing to adapt, not someone running the same strategy from five years ago.
That sentence kept bugging me after I sent it, because it wasn’t really advice about one budget meeting. The whole industry is standing where that little shop is standing, and I think it’s worth walking through why, with receipts… because the receipts are wilder than the vibes.
The machine underneath the machine
For twenty-five years, search marketing sat on one stable loop. A person types a query. A page of links comes back. The person clicks one, lands on a website, and maybe buys something. Every job title in the field is a bet on one step of that loop: SEO people fight for position in the links, paid-search people bid on the click, CRO people work the landing. Two industries, one loop, and everything in both of them is measured and billed in clicks.
If you want proof of how load-bearing the click is, look at Google’s own fine print. Their Ad Grants program gives qualifying nonprofits $10,000 a month in free search ads, which is genuinely great, and it comes with a rule: keep a 5% click-through rate every month, or miss it two months running and your account gets deactivated. The click isn’t just how the industry measures success. It’s how you keep the grant.
That 5% rule is about to matter more than Google intended.
The click is being rationed
In 2025, Pew Research watched 900 real people run 68,879 real Google searches. When an AI summary appeared at the top of the results, people clicked a traditional link 8% of the time. Without the summary, 15%. The AI answer cuts clicking roughly in half, and the links cited INSIDE the summary did even worse: clicked on 1% of visits. One.
And the summaries are spreading. Rand Fishkin’s SparkToro, working from Similarweb’s clickstream panel, found that 68% of US Google searches now end without a single click to the open web, up from about 60% in 2024. That’s the fastest jump in a decade, and he pins it squarely on AI Overviews. So Google’s nonprofit program suspends your free ads if your click-through rate drops below 5%… while Google’s own answer feature is the thing pushing click-through rates through the floor. I don’t think that’s malice. I think it’s one hand of a very large company not reading the other hand’s memos. But if you’re on the receiving end of that squeeze, the distinction doesn’t help you.
Meanwhile, the asking itself is moving. ChatGPT crossed 900 million weekly users in February. Semrush’s clickstream data shows referral traffic from ChatGPT out to the web grew 206% in a year. People are choosing answers over links, hundreds of millions of times a week, on purpose.
Both things are true, and that’s the weird part
Now, the honest counterweight, because this piece has opinions but it doesn’t get to have its own facts. Gartner predicted in 2024 that traditional search volume would fall 25% by 2026. It’s 2026, and that didn’t happen. Google says queries are at an all-time high, search ad revenue grew 19% last quarter, and their CEO will happily tell you AI Overviews make people search MORE. He’s not lying. The prediction was wrong on volume and right on behavior: people search plenty, they just increasingly don’t click.
So follow the money through that. Searches up. Clicks down. Revenue up 19%. The only way all three of those hold at once is that each surviving click costs more, and that’s exactly what the benchmarks show: the average Google Ads cost per click rose about 13% last year, in 87% of industries, the fifth straight year of increases. Google’s search share even slipped below 90% for the first time since 2015, and the auction still got more expensive. You’re bidding more for a shrinking supply of a thing people do less. If a client asked me to describe that market with a straight face, I’d draw them a picture instead.
That’s what the retainer conversation is actually about, underneath the line items. The old fee bought a mature playbook run by the book. The book is being un-written in real time.
The new discipline is being invented in public
Which brings us to the part I find genuinely exciting, because a blank playbook doesn’t stay blank.
There’s already a name fighting to stick: generative engine optimization. The academic paper that coined it tested what actually makes content show up in AI answers and found you can boost visibility by up to 40%. And look at WHAT worked: adding citations. Quoting credible sources. Including statistics. The machine’s favorite content, it turns out, is content that shows its work.
There’s money arriving too. Profound, a startup whose whole product is monitoring how AI engines talk about your brand, raised at a billion-dollar valuation in February, two years after the category didn’t exist. Semrush ran 100 million AI citations through a study to figure out which sources the engines actually lean on (Reddit and Wikipedia dominated for a while, then the mix lurched overnight when a model updated, which tells you exactly how settled this all is).
And a lot of it is still folklore. There’s a proposed standard called llms.txt, a plain text file that tells language models what your site is. I’ve written one for my own site. Google’s John Mueller compares it to the keywords meta tag, and server logs suggest most AI crawlers don’t even check for the file. That’s where this field is right now: the practitioners are shipping tactics faster than anyone can confirm the tactics do anything. Edition two is being drafted in public, in pencil.
This is a design problem in a marketing costume
Here’s my actual opinion, the one I sat down to write. Read that list of proven tactics again: cite your sources, quote credible people, publish real numbers, structure the page so a machine can lift the answer cleanly. There’s no bid strategy in any of that. It’s writing and information design for a new reader, one that reads the lines instead of between them. I’ve written about what that reader believed about me (it thought I sold lumber, it’s a whole thing), and about owning the layers of the web it reads. The short version: an answer engine doesn’t visit your brand, it retells it, and what survives the retelling is decided by how clearly you built the source material.
That’s a designer’s problem. It always has been. “How do we get represented accurately to an audience we don’t control” is the oldest brief in the field. The audience just stopped being entirely human. And the design work extends past the source material into the destination, because when someone DOES click through from an AI answer, they arrive mid-conversation. They’ve already read the summary, and now they’re checking whether you’re as good as the machine said. Early numbers back this up: ChatGPT referral traffic to ecommerce sites converts about 31% higher than non-branded search. Fewer people are coming through the door… but the ones who come were walked there.
Fair warning before anyone re-budgets around a trend piece: that traffic is still small, low single digits of most sites’ referrals. This is a shift in kind arriving ahead of a shift in scale. Which is precisely the window where learning it is cheap.
Who gets to write edition two
So, back to that little shop’s budget meeting, and to yours, whichever side of the table you’re on.
If you’re hiring for this: stop shopping for the person with the longest keyword résumé. Ask every candidate one question, “what’s your plan for the fact that people click less?”, and listen for whether the answer is a plan or a shrug. A confident shrug at a discount is still a shrug. And if the honest state of the field is “we’re all figuring it out,” then pay for figuring-it-out ability: curiosity, measurement, a willingness to say “that didn’t work” out loud.
If you’re a designer or a writer wondering whether this is your fight: it’s more yours than anyone’s. You already do the things the machines reward. You compress a company into a sentence, you keep the story consistent everywhere it appears, you know that the way a page is structured changes what it says. What’s left to learn is the reader: how these models retrieve, what they quote, what they mangle. If you use LLMs every day you already half-know it, the way you know a friend’s taste in restaurants. The career paid-search person has to unlearn an entire economy. You just have to learn one new audience… and the tools to reach it get cheaper every month.
Every mature playbook was once somebody’s nervous experiment. The ten-blue-links edition took years to harden into best practices and certifications and $5.26 average clicks. Right now the next edition is sitting there, mostly construction lines, and the people qualified to fill it in aren’t defined by their job title. They’re defined by whether they understand the new reader.
The playbook isn’t written yet. That’s not the problem. That’s the opening.