AI Search
What AI search changed about the way I write
For most of the time I have been doing this work, writing for search meant one thing: earning a place in a list of ten blue links, and trusting that someone would click through to find the actual answer. That assumption is no longer safe to make. People increasingly ask ChatGPT, Perplexity, or Google’s own AI Overview a question and receive a paragraph in return, sourced sometimes from somewhere on the web, but read in full inside the answer itself, with no obligation on the reader to go any further.
I noticed it first almost as an aside, while writing a report for a client. Kahana Baler, a boutique bed and breakfast in Baler that I built the website for, has been picking up visits from Bing for a little while now, and lately, in numbers too small yet to build a strategy on, from AI assistants as well. I want to be honest about the scale of it: it is not yet a meaningful share of the site’s traffic. But the direction is the one thing I am fairly certain of. A few visits today from assistants that did not exist eighteen months ago is worth paying attention to, the way an early impression on a brand-new keyword is worth paying attention to, long before it becomes a real number.
The articles I write now open with the direct answer, stated plainly, in the first two or three sentences, because that is the part most likely to be lifted whole.
What that has changed, practically, is how I write. I used to build an article the way most of the industry still does: an introduction that eases in, a body that develops an argument, a conclusion that ties it up. That structure serves a human reader who arrived with time to spend. It does very little for a system trying to extract one clean, quotable answer from a page and decide whether to cite it. So the articles I write now open with the direct answer, stated plainly, in the first two or three sentences, not because it reads more elegantly but because that is the part most likely to be lifted whole. Everything after it exists to support that opening, not to delay it.
The rest of the shift is less about prose and more about scaffolding. Comparison tables where a buyer would genuinely want to compare, rather than as decoration. FAQ sections marked up in schema, so the question and answer pairs are legible as data, not just as text on a page. Product and article pages carrying structured data that states, in a format a machine can parse without guessing, what a thing is, who made it, and what it costs. None of this is exotic. Most of it is available to any WordPress or Shopify site with the right plugin and a bit of care. What is not automatic is the discipline of doing it on every page, every time, rather than as an afterthought on the pages that already rank.
I have built this into two content systems for client work now: one for a European home and kitchen retailer restructuring its entire product catalogue, and one for an industrial distributor whose readers are tradespeople comparing thermostats and actuators, about as far from search-engine glamour as a subject can get. Neither client asked me, in as many words, to optimise for AI search. They asked to be found. It happens that structuring content so a machine can read it cleanly is, at the moment, indistinguishable from writing it well for a person in a hurry. I don’t know how long that alignment will hold. For now, it is the whole of the job.