Crop Pages

I accidentally optimized our website to the point that we were capturing about 2% of global Google search traffic for the word "beet." We were popping in and out of the first-page SERP. Not bad for a tiny CSA farm. None of it was the plan. The plan was getting our members to eat their kohlrabi.

Blue Glass Farm is my farm, and about half of our sales come through a program called a CSA, where customers pay in advance of the season for a weekly share of whatever we're harvesting. The reality is that a lot of those purchases are aspirational. Some customers come in knowing exactly what they're getting, seasoned CSA members who didn't like the program they were using, or who moved to the area and needed a new source of vegetables. But for a big swath of our customers, buying a CSA share is like buying a $300 pair of running shoes for your first 5K because you just picked up a new hobby. They arrive with high ambitions and don't realize how hard it is to eat fresh, home-cooked, seasonal meals, especially if they aren't already cooking.

I know, because that was me when I signed up for my first CSA. I had been cooking for a decade, but not with fresh ingredients, and when I decided I wanted the best ingredients, a little Googling told me the way to get them was a CSA, a hodgepodge of whatever was in season every week. I'm embarrassed to say I composted half of that food. Members get handed ingredients they don't recognize or feel intimidated by, fresh food means real prep, and most of us were taught to cook from complicated recipes that send you to the store for a dozen more ingredients.

I wanted to solve that with information. So we added a service layer on top of the CSA, online education for every crop we grow: what it is, what to do with it, how to store it, what goes with it, and which recipes are worth your time. Members said they genuinely appreciated them, so we kept building them out over the year.

The problem I quickly ran into was that this kind of content takes a ton of time to write by hand, and I wasn't always sure what to write about. We grow 50 to 90 crops in a season, across some 150 plantings of about 270 varieties, and I run the farm full time. So I tackled two problems in order. First, figure out what to write about. Second, figure out how to write it fast enough.

For the first, I signed up for Ahrefs, and quickly realized the thing I should have done years earlier was set up a Google Analytics profile so I could see what was actually happening on our site. Without that history I ran keyword analysis instead, looking at what people searched around particular terms, and it wasn't as useful as I'd hoped. Most of the keywords surfacing had little to do with what we did.

That pushed me toward a different idea, borrowed from a podcast where someone described finding business ideas by feeding terms into Google's autocomplete API to see which verticals to target. I could do the same with a vegetable. Type how, what, why, or where in front of "beet" and Google will tell you, verbatim, what the world wants to know about beets. So I took a pass at it and built a bunch of FAQs around the real questions people had.

The second problem was speed. We're talking tens of thousands of words of content, and my first attempts to have AI write it ran into the fact that AI doesn't know as much about food, and especially fresh vegetables, as you'd think. It hallucinates, particularly around storage, and it repeats the old wives' tales that circulate around the internet. I knew I'd need a heavy editorial lens. So I built a pipeline where the AI goes out and researches, gathers examples of the tone and language I want, and makes a first pass, and then I come back and run it through a manual editorial process.

Before generating a page I wrote a spec for what a crop page is: storage advice tuned to farm-fresh shelf life instead of grocery-store shelf life, flavor described through comparisons a home cook already knows, and guardrails, no food-blog fluff, no health claims, and if you're not sure a fact is true, leave it out. The recipes needed the tightest leash, because a language model will happily invent a plausible URL on a plausible food blog with total confidence, so a Playwright script opens every cited link in a real browser and reports back the page title before anything ships. My thesis on AI content is that it's a megaphone. It takes your voice, your knowledge, and your taste and produces at a scale one person never could, but it doesn't supply the voice.

The pipeline got me to good results much faster. Then the busy farm season of 2026 started while I was still working through the crop pages, and I realized I simply didn't have enough time in front of a computer. That's where the real unlock happened. I piped Claude Code into a Discord call with text to speech, so a full Claude Code instance could read me a piece of text, take my edits out loud, and make the corrections while I went about my daily tasks on the farm. It worked so well it became its own project, Claude Voice, and it has its own page on this site.

The results were better than I could have imagined. I had put this content out hoping it would help my CSA customers, and then I started watching the numbers come in on Ahrefs and was flabbergasted. I had made the lowest-lift content I could. Just text, no images, no video, decent on-page SEO but a weak internal linking strategy and almost no time spent on backlinks. And there was a period where our little farm sat right alongside blogs like Serious Eats in the results for the word beet, talking about a badly misunderstood root vegetable.

I realize now that what I'd done was more useful than it looked. As agentic AI took off and search engines started tuning their algorithms toward quick answers, our huge corpus of FAQs, geared toward exactly the questions people type into Google, started showing up in search results and in AI searches. There's a scramble right now to brand this as AEO, or GEO, or whatever acronym wins. We had optimized for it before anyone clued in that it was a useful thing to do. The takeaway is simple. If you want AI to surface your information, answer the questions it would put into a search.

What I liked best about this project was watching small efforts compound in ways I never expected. To notice that, though, we had to stop, measure, observe, and then give the work time to perform. Time to cook. I've taken a much more intentional approach to the website since, and we're seeing slow, steady growth in other categories. We'll see next year whether we can break 10,000 organic views a month.