You watch your own search behavior change. You ask Claude or ChatGPT a question and get a paragraph instead of ten blue links. Google’s AI Overviews now reach more than 2.5 billion monthly active users and, on the informational queries a prospect is likely to ask, push organic results below the fold. The website you were about to commission, or the one you already have, suddenly looks like an artifact from the previous era.
The question that follows is reasonable. If AI answers the question for them, why does the site need to be there at all?
The answer turns on where the AI got the answer it gave. Assistants don’t generate facts about your business out of thin air. They retrieve from indexed web content, synthesize a response, and cite the sources they pulled from. Your site is the source that gets cited, the authority that gets quoted, and the URL that gets linked when someone asks the assistant about your category. No site means no citation. No citation means no presence in the answer.
The site’s job has shifted. It used to be the place people landed after a search. It is now also the source the assistant pulls from before anyone lands anywhere. Both jobs run on the same site doing different work.
What AI assistants are actually doing when they answer
The mechanic is simpler than it looks. Assistants retrieve indexed content from across the open web, sometimes from a training corpus, increasingly from live retrieval at query time. They synthesize a response, then cite or link the sources that informed it. The sites that get cited are the sites with structured, parseable, authoritative content on the topic.
“Authoritative” means most of what search engines have rewarded for two decades: clear topical focus, internal linking, freshness, schema, and backlinks from credible sources. There is one new wrinkle: the content has to be readable to an agent, not just a human. A page that looks fine to a visitor but renders behind heavy JavaScript or hides its meaning in image-only content is invisible to the assistants citing the open web.
AI didn’t break the web. It promoted some sites and demoted others, and the rules for which side you end up on are more legible than people think.
The cost of not being there
A founder weighing whether to invest in a site is usually weighing one cost against zero. The actual zero isn’t zero.
You’re invisible in the AI answer. When someone asks Claude “who builds custom AI tools for small businesses,” the assistant cites whoever has indexed pages on that topic. If that’s not you, you weren’t in the consideration set. Your name didn’t appear, and the prospect moved on.
Your competitors define your category. Without your own pages on your own terms, the description of what you do gets written by other people, in their language, against their comparison criteria, and around their framing of which features matter.
You lose the soft trust signal. Even when an assistant surfaces your business by name, prospects still click through to verify. No site, or a thin site, fails that check. The AI did its job. The conversion died at the source.
You can’t update the record. Wikipedia, directories, third-party listings: none of those let you tell your story in your own voice. Your site is the only place where you control what gets cited.
The data on this gap is unambiguous. Pew Research found that only 8% of users click a result when an AI summary is present, compared to 15% without one, and 26% of AIO-present sessions end the search entirely. But Seer Interactive’s 2025 longitudinal study found that brands cited inside an AI Overview get 35% more organic clicks and 91% more paid clicks across the rest of that SERP than non-cited brands. The traffic that used to come from generic blue-link clicks is being redistributed toward cited brands. The penalty for not being cited is steeper than the penalty for ranking on page two used to be.
What changes about the brief
This changes what the site is structurally optimized for, which is a different job from redesigning the homepage.
Discoverability over decoration. Beautiful animations don’t help an assistant cite you. Clear topical pages, descriptive headings, and content that directly answers the question a prospect would actually ask do.
Structured data still earns its place. It earns it through Google’s rich results and by pinning down who your organization is. Organization, FAQPage, BreadcrumbList, Article: the basics. We ship these on every build, including this site, because the cost is low. Do not expect it to buy AI citations on its own: Google says no special schema is needed for its AI features, and Ahrefs’ 2026 test of 1,885 pages that added schema found no citation lift.
One clear answer per page. This is our editorial choice rather than an AI requirement: Google says its systems can pull the relevant piece out of a multi-topic page. But a page with one job is easier to write, easier to link to, easier for a human to skim, and leaves no ambiguity about which query it should own.
Freshness still matters. Published dates, updated dates, and recent posts signal that the site is maintained. Stale sites get demoted regardless of which engine is reading them.
Original perspective beats summarization. AI is excellent at compressing other people’s content. It is bad at inventing a take. Your point of view, your decisions, your project specifics: that is the part it can’t generate without you. That’s also the part it cites.
What stays the same
A lot of “AI search era” advice acts like the rules changed entirely. They mostly didn’t.
Site speed still matters for human conversion and as a ranking input the engines (AI ones included) use to decide what to surface. Accessibility still matters: agents read the same DOM screen readers do, and a site that fails WCAG often fails AI parsing for the same reasons. Overlay widgets that promised one-click compliance don’t fix the underlying parseability problem either way. Clear writing still matters, because the same structure that helps a skim reader helps an assistant extract the answer. Backlinks still matter, because citations from other authoritative sources are how engines decide who to trust on a topic.
Most of what was good practice for SEO and UX makes you more citeable. Most of what was bad practice for humans is also bad practice for agents.
The cost of being un-cited, in real numbers
The reallocation has been measured. Reuters Institute’s Journalism, Media, and Technology Trends 2026 report, drawing on Chartbeat data from more than 2,500 publisher sites, found Google search referrals to publishers fell 33% globally in the year to November 2025 and 38% in the U.S. Google Discover referrals fell 21% globally. Smaller publishers are hit harder than larger ones.
The pattern in the publisher data extends to small businesses too, just with less press coverage. Sites that were ranking on a thin layer of generic content are losing the click-throughs that used to drive their pipeline. Sites that show up as cited sources, or that have direct, branded, returning audiences, are absorbing the shift.
The architecture that wins this transition is the one designed to be both. A site that is the cited source on its category, and that converts the clicks that do arrive.
So, do you still need a website?
Yes. The questions worth asking have shifted, though:
- Is my site the source when AI answers questions about my category?
- If not, what would it take to be?
- What pages do I need that I don’t have?
- What signals (schema, freshness, depth, authority) am I missing?
A site that can’t answer those affirmatively is operating under the old rules. Getting to the new ones usually means reframing what the site is for, then making a handful of structural changes that compound, not commissioning a rebuild.
The implementation side (schema choices, llms.txt, semantic HTML, agent-readable structure) is its own post. Here it is. If you want the first question answered before reading it, our free AI citeability checker reports in a few seconds whether an assistant can read a given page at all, and an AI search visibility audit answers it for the whole site.
Your site is doing two jobs now
It used to convert traffic that arrived through a blue link. Now it also serves as the source the assistant pulls from before traffic arrives at all. Both jobs run on the same site doing different work. If the site is built to be cited, it converts when the visitor lands too.