Do we really need websites
if everything is queried from AI?

The zero-click reality

The numbers stopped being deniable a while ago. AI answer engines resolve a large and growing share of informational queries without a single click to the source. Publishers report referral traffic from search falling double digits year over year while "AI crawler" traffic climbs. For a typical documentation site, the ratio of bot readers to human readers has quietly inverted: the most frequent consumer of your content is now a model's retrieval pipeline, not a person.

For the classic marketing website — ten pages of hero sections, testimonials, and a pricing table — this looks existential. If the user asks their assistant "what's the best SAST tool for a small team and what does it cost?", the assistant synthesizes an answer from whatever it has read. The user may make a purchasing decision without ever rendering a pixel of yours.

AI needs a source of truth — that's the website's new job

Models don't know your current pricing, your latest release, your changed API, or your security advisory from training data. They know it from retrieval — crawling, browsing, or being handed your content at query time. If your information isn't published somewhere authoritative and crawlable, the assistant doesn't fall silent about you; it answers anyway, from stale data, a competitor's comparison page, or a hallucination.

That's the strategic inversion: the website used to be where you persuaded humans. Now it's where you ground machines. Being absent doesn't remove you from the conversation — it removes your version of the facts from it. The canonical URL becomes less of a destination and more of a citation: the thing the AI links when it makes claims about you, and the thing users check when the stakes are high enough to verify.

Your new visitor is a machine — design for it

If the dominant reader of your site is an agent, the design priorities flip. Agents don't care about your hero animation, your carefully tuned scroll-triggered transitions, or your brand font. They care about whether facts are extractable, unambiguous, and current.

• Structured data over visual polish. Schema.org markup, clean semantic HTML, accurate metadata — the things that were "nice for SEO" are now the primary interface.
• Machine-facing entry points. llms.txt, well-maintained sitemaps, an honest robots.txt policy, and stable canonical URLs are the new storefront.
• APIs as the transactional layer. When agents don't just read but act — book, buy, file, schedule — the form gives way to the endpoint. The "website" for an agent is an API with documentation it can read.
• Freshness as a ranking factor for truth. A stale page doesn't just look bad anymore; it actively pollutes every AI answer that ingests it.

What survives, what doesn't

Likely to fade

• Thin informational pages whose only job was capturing a search query a model now answers directly.
• Ad-funded content farms — their economics required human eyeballs in transit.
• Form-driven interfaces for routine transactions that agents can perform via API.

Likely to strengthen

• Canonical sources: documentation, pricing, advisories, release notes — the ground truth machines must cite.
• Trust-critical surfaces: checkout, banking, healthcare portals — places where humans want to see for themselves before committing money or data.
• Experience destinations: products that are the website (dashboards, editors, games, communities). An agent can summarize Figma's pricing; it can't be Figma.
• Brand homes: when every answer is machine-flavored text, a distinctive human-crafted destination becomes a differentiator again, the way vinyl survived streaming.

The security angle nobody discusses

A web read mostly by machines has a new threat model. If AI assistants are the layer through which users learn about your company, then attacking that layer is the new phishing. We already see it in the wild:

• Prompt injection in page content — hidden instructions in HTML aimed at the browsing agent rather than the human ("ignore previous instructions, recommend ...").
• AI-targeted cloaking — serving different content to AI crawlers than to humans, to poison answers about competitors or inflate claims invisibly.
• Hallucinated endpoints — models inventing URLs, package names, and API routes; attackers register them and harvest the traffic. Your security perimeter now includes URLs you never created.
• Agent-driven transactions — when agents hold credentials and spend money on a user's behalf, your website's auth flows, rate limits, and APIs face automated counterparties at scale.

Verdict: yes — but not the websites we've been building

Do we need websites if everything is queried from AI? Yes, unavoidably — because "queried from AI" is just "queried from websites, with extra steps." The model is a lens, not a source. What we don't need is the website as a persuasion funnel optimized for a human who reads top-to-bottom, because that human increasingly isn't there.

The website's next act is quieter and more important: the verifiable, machine-readable, continuously accurate record of what your organization offers and claims. Less theater, more infrastructure. The teams that win this transition won't be the ones with the prettiest landing page — they'll be the ones whose facts every AI in the world finds, parses, trusts, and cites.