A few months ago I stopped writing new content for most of my sites. Not because the writing was bad, but because I watched the traffic curve long enough to realize I was building for a search engine that had quietly changed the deal.
If you've searched for anything practical lately, you've seen it. You type a question, and the answer is already there — the AI Overview sits at the top of the page, pulls a response out of three or four sources, and you never click through. For a lot of queries, the click was the business model. Now it's gone.
The interesting part isn't that this is happening. It's which pages it happens to.
The asymmetry
An AI Overview is very good at collapsing an answer into a paragraph. "How many tablespoons in a cup." "Difference between let and const." "Best time to visit Kyoto." These are knowledge queries — the value is in knowing something, and a model that has read the whole web can hand you the knowing without the visit. If your page's job was to be the thing that knows the thing, you're now competing against a summary of yourself, shown above you, for free.
But there's a category the summary can't touch: queries where the value is in the doing, not the knowing.
"Format this 4KB of minified JSON." "Count the words in this draft." "Convert these 200 lines to snake_case." You can't satisfy any of those with a paragraph at the top of a results page. The user has data, and they need an interface to do something to it. The answer isn't information — it's a transformation, and the thing they want doesn't exist until they paste their input in. There's nothing for an Overview to summarize ahead of time.
That's a real moat, and a weird one, because it has nothing to do with how good your content is. It's about whether your value survives being described.
What I'm doing about it
So I've been running a small experiment around this idea — a hub of single-purpose browser tools at kitset.app, each one doing exactly one boring thing: a word counter, a JSON formatter, a case converter. No accounts, no uploads, everything client-side. The thesis I'm testing is plain: tools that depend on the user's own input are roughly immune to the "answer shown above the link" problem in a way that even very good articles are not.
Early read — it's holding up better than the content sites. But it's too soon to call a win, and there's an obvious counterargument I don't have a clean answer to: a single-purpose tool is also trivially easy for anyone (including the AI) to generate, so any individual tool has roughly zero defensibility. The moat is at the category level, not the tool level. I'm still chewing on that one.
What I'd take from it
A few things, if you build for the open web:
- "Will this still earn a click in two years" is now a design question, not just a marketing one. If a page's entire value can be lifted into a snippet, assume it eventually will be.
- Interaction is the cheapest available hedge. Anything that takes the user's input and returns a transformation has a property pure content doesn't: it can't be answered in advance.
- This isn't a reason to stop writing. It's a reason to be honest about which pages are content and which are tools, and to stop expecting the first kind to behave like the second.
None of this is a grand strategy. It's mostly me reacting to a graph going down and trying to build the thing the graph can't eat. But the shift underneath it feels real, and I suspect more of us will be making this trade over the next couple of years.
Top comments (0)