Every few months someone publishes a piece asking whether websites are dying. AI search. Conversational interfaces. Zero click results. The argument runs like this: if buyers ask ChatGPT instead of Google, and Perplexity answers before anyone visits your site, what is the website even for?
It is the wrong question. The website is not dying. It is splitting in two, and most companies are still treating it as one thing.
Two shifts, at the same time
On the build side, making an interface is collapsing toward zero cost. v0 by Vercel has six million developers generating production-ready interfaces from plain English. Framer raised $100 million betting that AI-generated sites become the default. Builder.io now calls itself an AI frontend engineer rather than a CMS. How fast a page can exist has stopped being interesting. What the page is for is the whole question.
On the buying side, 73% of B2B buyers now use ChatGPT or Perplexity somewhere in their research. Perplexity grew 800% year over year. Organic click-through drops around 35% when a Google AI Overview sits above the results.
Buyers have not left the web. They arrive differently, and many of them were qualified or disqualified before they ever saw your homepage.
Put those two together and your site now has two jobs that have almost nothing to do with each other.
Job one: be legible to machines
Agents, procurement tools and LLM search read your site before any human does. They summarize it, compare it against alternatives, and cut the vendor list before anyone types your URL.
Gartner expects 40% of enterprise applications to have task-specific AI agents by the end of 2026. Those agents are doing vendor research right now. If your content is not structured for retrieval, you are not on the shortlist. You have not been rejected either. You are simply absent, which is worse, because nobody tells you.
The numbers are stark. Perplexity citations convert at eleven times the rate of traditional organic search. And only 11% of domains are cited across both ChatGPT and Perplexity. Most sites are invisible to at least one of them.
This is not quite an SEO problem. It is a content architecture problem. Bullet points written for skimming are not the same as claims written to be quoted. Schema matters again, differently. The question has moved from “does Google index this” to “can a model reproduce an accurate answer about us from this page.” Those two require different writing.
Job two: work for the human who already decided to look
By the time a person lands on your site, they usually know who you are. They are evaluating, not discovering. The site should behave accordingly, and mostly it does not.
Nearly 70% of B2B buyers say they struggle to find relevant products once they are already on a site. Navigation built for a 2015 buyer journey does not serve someone who arrived in 2026 knowing exactly what they need to confirm.
This is where conversational interfaces actually belong, and it is not as a chatbot bolted onto a static homepage. It is a retrieval layer that helps people find what they came for. Research presented at ACM SIGIR in 2025 found that sessions mixing conversational and traditional interfaces ran longer and more thoroughly than either one alone. People use chat to synthesize and navigation to verify. They want both.
The companies doing this well are not replacing their navigation with a chat window. They are adding a layer beside it. Ruggable ran landing pages that changed with the ad source and saw click-through rise sevenfold. Personio personalized by company characteristics and lifted conversions 45%. That is personalization as navigation help, not personalization as a party trick.
Where this goes
Over the next year, AI-generated interfaces become normal for the fast-moving pages: campaigns, landing pages, tests. Not everything, but the disposable end of the site. Conversational search keeps eroding traffic to informational pages. Companies that have not structured content for retrieval feel it first in attribution, where the pipeline stops making sense, and then in deal velocity.
Three years out, I think the monolithic website gives way to something with two surfaces. Humans navigate one. Agents read and act on a structured layer underneath. Content platforms without a retrieval story start to look like the systems that could not handle responsive design in 2012.
The strategic weight moves upstream: to positioning, to content architecture, to being quotable. The cost of building a page approaches zero. The cost of being absent from AI search approaches everything.
What this changes about how I build
I build for industrial companies. Most of them have sites resting on 2018 assumptions: a buyer who searches, clicks, scrolls and fills in a form.
That buyer still exists. They are now one of two. The other one sent an agent ahead to filter the list, and that agent needed clean, factual, well-structured content in order to recommend you accurately.
So the shift is not “add a chatbot.” It is deciding what your site is actually for now. If the answer is still “explain what we do and capture leads,” that is a perfectly good answer. It just requires two separate conversations. One about being found by machines before buyers arrive. One about what to show the humans when they do.
Most companies are having neither conversation clearly. That is the opening, and it will not stay open for long.