I write for a living, which sounds more impressive typed out than it feels most days. A lot of what I actually do is search things. For the PhD, for pitches, for the ten small logistical questions that come with running a life across two countries. Somewhere in the last year or so, without ever deciding to, I noticed I’d stopped clicking through on most of it. Google just answers now, in a gray box before the actual results, and I read the paragraph and move on with my day. I hadn’t thought once about who used to get that click until I read about a woman whose entire business was built on being the person who answered exactly those questions, and who watched the number that funded her business fall by eighty percent in about two years.
Her name is Carrie Forrest. She runs a site called Clean Eating Kitchen, and until recently, it was a genuine success story of the kind the early food-blogging internet promised and rarely delivered: a single person, a decade and a half of unglamorous consistency, and a business built entirely on people typing questions into Google.
The blog that grew for fifteen years on the strength of very small questions
Forrest started blogging in 2009, documenting the recipes and habits behind her own health, and kept doing it long enough that documenting her health journey quietly turned into a career. By 2022, Google’s own case study on her site was using her as an example of what a small, independent operation could earn through advertising alone, and by her account in her Food Blogger Pro interview, the site “went from, I don’t know, a few hundred thousand page views and up to about a million page views a month.” A million monthly pageviews is not viral-hit territory. It’s the much less glamorous outcome of writing thousands of specific, useful posts and having Google send people to them, reliably, for years.
That reliability is the part that’s gone. Forrest now puts her traffic at somewhere between 100,000 and 200,000 monthly pageviews, an 80 percent drop she traces almost entirely to one thing.
How to cut a banana was never supposed to be load-bearing
The posts that used to bring in the steadiest traffic on Clean Eating Kitchen were not the ambitious ones. They were the small, almost embarrassingly simple ones: how to freeze sweet potatoes, how to boil an egg, how to cut a banana. Content like that ranks fast, gets searched constantly, and used to be one of the easiest ways for a new or growing blog to build an audience, because the competition for a keyword that small was thin and the intent was crystal clear. Forrest built a meaningful share of her early growth on exactly that logic. In her own words, “I was so keyword focused, they were very kind of simple terms like that.”
Those are also, it turns out, the first questions an AI system can answer with total confidence and zero risk. Nobody needs a nuanced, sourced, multi-paragraph take on banana-cutting technique. They need three steps and a photo, which is precisely the shape of thing a language model can generate from a search index without sending anyone anywhere. “Every single one of those now is answered by an AI overview,” Forrest says of her simple how-to content. The traffic didn’t get redistributed to a competitor. It just stopped needing a destination at all.
She’s the specific case, not the exception
It would be a cleaner story if Forrest had simply built her business on the wrong kind of content. In practice, she’d built it on exactly what the entire industry was told, correctly, to build for a decade. Other food writers are describing the same year in different numbers. Eb Gargano, who runs the UK site Easy Peasy Foodie, told Fortune that her turkey-recipe traffic was down 40 percent year over year. The same AI Overviews, she said, can also be dangerously wrong.
Describing an AI-generated version of her own Christmas cake, she didn’t mince words: “You’d end up with charcoal!” Adam Gallagher of Inspired Taste told Fortune that click-through rates on his cocktail queries specifically dropped 30 percent once AI Overviews started serving up what he calls “Frankenstein AI recipes” – ingredients and steps stitched together from several sites at once and presented as one.
It worked for a while, and now it’s not.
That’s Forrest’s own summary of the last two years, and it’s a strange kind of sentence to have to say about a job you built from nothing. Not bitter, exactly. Just accurate, in the way people get accurate once they’ve stopped arguing with a fact and started living inside it.
What she’s doing with the years she has left of this
The part of Forrest’s interview I keep coming back to isn’t the traffic number. It’s what she’s decided not to do about it. She isn’t suing anyone, and she isn’t writing more content faster in the hope of outrunning the algorithm. She’s leaning on the email list she started building back in 2014, long before anyone treated an inbox as a hedge against a search engine, and she’s putting real effort into video and YouTube, formats an AI overview can summarize but can’t actually replace. She’s also, by her own account, tried to sell her audience workshops, ebooks, and a Substack, and mostly been told no. People came to her for a recipe, not a relationship with a paid product, and that distinction turned out to matter more than any content calendar.
What she’s landed on instead is less measurable and, I think, more honest: content built around transformation and human connection. A search engine has no interest in summarizing a relationship. There’s no clean paragraph to lift out of trust built over a decade, no bullet point that captures why someone keeps coming back to the same person’s recipes. She sounds, in the interview, less like someone who solved a problem and more like someone who’s stopped waiting for the old numbers to come back. “I’m feeling more creative than I was,” she says. “I’m back in touch.”
I think about my own habits again here, the questions I no longer bother clicking through on, and I notice the shape of what got lost isn’t really information. The banana still gets cut correctly. What disappeared is the fifteen years of a specific person’s attention that used to sit underneath the answer, the trial and error, the version that didn’t work, the small human fact that someone tested this enough times to be sure.
An AI overview can tell you what to do. It has never once told you who figured it out, or what it cost them to keep figuring things out long after the internet stopped sending anyone to say thank you.
