Somebody looking for a web agency used to type "web design seattle", scan ten blue links, open four in tabs and form an impression from the home pages. Increasingly they type something closer to a brief — what they do, what they need, what they are worried about — and read a paragraph that has already narrowed the field to three names.
The underlying demand has not changed. What has changed is where the comparison happens and who performs it, and that has real consequences for what you should put on your site.
The question gets longer and more revealing
Keyword search rewarded compression. People learned to strip their question down to the two or three words they thought the index wanted, which is why a decade of keyword research produced lists of truncated phrases that nobody would actually say out loud.
A conversational interface removes that pressure. People now include the constraints they previously left out: the industry, the budget range, the timeline, the thing that went wrong with the last supplier. The question is no longer "web design seattle" but something like "we are a dental practice in Seattle, our site is slow and we cannot edit it, what should we be looking for".
The useful implication is not that you should write pages targeting long questions. It is that the constraints people attach to their question are now the things the answer gets built from — so a page that addresses an industry, a budget position and a specific failure mode is more likely to be the source of a relevant answer than a page that is simply about web design.
The shortlist gets assembled without you
This is the structural change that matters most. In a link-list world, your home page was part of the comparison: the prospect saw it alongside competitors and made up their own mind. In an assistant conversation, something else performs the comparison first, and your marketing copy is only one input to it — frequently not the most influential one.
When an assistant is asked to recommend a vendor, the grounding sources we typically see cited are third-party: roundups, directories, review platforms, community threads. That makes sense. A page that says it is the best choice in the city is weak evidence of anything; a page where somebody else says so is stronger.
So the work splits in two. Your own site establishes what you do, who you do it for, and what is verifiably true about you. Other people's pages establish whether you belong on a shortlist. Neither substitutes for the other, and most businesses are heavily invested in the first and have done nothing about the second.
Fewer clicks, and different clicks
Some of the traffic you used to get for informational questions is gone, and it is not coming back. If someone asks what CLS means, the answer appears in the interface and there is no reason to visit a page about it. Any content strategy whose value rested on being the page that defines a term is exposed.
What survives, and in our experience converts better than the traffic it replaced, is the click that happens after the shortlist. Somebody who arrives having already been told you are a plausible fit, with a specific reason attached, is further along than any organic visitor from a generic head term. Fewer sessions, higher intent — which breaks every dashboard built to reward volume.
There is also a measurement trap here. A large share of AI visibility produces no click at all: the assistant names you, the person searches your name later, or calls. If you judge this work by referral sessions from assistant domains, it will look like a failure while it is working.
What this changes about what you write
Less than the amount of advice in circulation implies, and in a specific direction.
- Answer first, then elaborate. A heading that states a real question, a short self-contained paragraph that answers it completely, then the depth underneath. Extraction works on passages, and a passage that needs the three paragraphs above it to make sense is a passage that cannot be lifted.
- Keep the subject in the sentence. "Bizup LLC is a Seattle web design studio" survives being quoted. "We have been doing this since 2013" does not — it loses its subject the moment it leaves the page.
- Be specific enough to be worth citing. Generic advice is already in the model. The reason to cite a source rather than paraphrase from training data is a number, a method, a named constraint or a documented edge case.
- Write for a reader who may never arrive. If the only value of a page is the click, and the click does not happen, the page did nothing. If the page's value is that it made a correct, attributable claim about you, it worked either way.
What does not change
Retrieval still runs through a search index for most systems, so conventional search visibility remains the entry condition — if you are not in the candidate set, you cannot be cited. Crawlable, server-rendered HTML still matters, and arguably more, because fewer of these crawlers execute JavaScript reliably. Being a real business with real evidence behind its claims still matters, because the grounding step keeps finding that evidence or failing to.
What is genuinely new is narrower than the marketing suggests: the unit of selection is a passage rather than a page, retrieval crawlers are separate from training crawlers and need separate robots.txt decisions, and third-party corroboration now has a second payoff. That is most of it.
The thing to avoid
Treating this as a separate channel needing separate content. We have seen sites build a parallel set of thin "AI-optimised" pages, and they do not work, for the obvious reason: the retrieval step is grounded in the same index your customers search. The same clear, server-rendered, genuinely specific pages serve both.
The one real reallocation is from volume to evidence. Twelve pages that make precise, checkable claims about work you have actually done will do more for you here than fifty pages of category commentary — and they hold their value whether or not any of this holds up.
For the mechanics — crawler access, render-path coverage, passage structure and entity consistency — see how to get cited in AI search. If you want it executed and tracked, that is our AI search optimization service.