Two days ago, Google flipped the switch. The search bar that's sat unchanged at the top of the page for twenty years is now an AI. Type a question, get a synthesized answer, with small grey citations beneath it. Three years ago, a search for "best dentist in Miami" sent traffic to ten websites. Today, it sends traffic to one paragraph of AI. If your website was built before 2025, it wasn't designed for the page it's now competing on. The old version of SEO is becoming secondary.
What actually changed
AI engines don't rank pages the way Google's blue links did. They cite statements.
ChatGPT reads your homepage and quotes a sentence. Perplexity pulls a paragraph from your blog. Gemini compresses your service description into a single line at the top of an AI Overview. The blue links still exist below, but for many queries the AI answer above already gave the user what they came for.
The job shifts. It's less about being on page one of search results, and more about being citable in the answer that now sits above page one.
Old SEO optimized pages to rank. AI SEO optimizes statements to be cited. Those are not the same job.
The page that gets cited is the page engineered for it
An AI engine deciding whether to cite a sentence from your website is running a short checklist behind the scenes. The exact weights are proprietary, but the signals are well documented:
- Page speed. Slow pages tend to be deprioritized as source candidates. Core Web Vitals + sub-second LCP show up repeatedly in published Google guidance.
- Semantic HTML. AI crawlers parse heading hierarchy, lists, articles, and section tags to understand which content is important.
- Schema markup. LocalBusiness, FAQPage, Service, Article, and Organization schemas tell the engine exactly what each block means.
- Crawler access. llms.txt at the root, robots.txt rules for GPTBot, ClaudeBot, PerplexityBot. Block them and citation is unlikely.
- Authority signals. Real author bylines, real addresses, real credentials. E-E-A-T signals (Experience, Expertise, Authoritativeness, Trustworthiness) consistently surface in both Google and AI-engine quality docs.
If your website doesn't pass most of this checklist, citation becomes meaningfully less likely. And without citation, AI-driven search traffic mostly bypasses you.
Why most pre-2025 websites struggle to keep up
Most business websites currently online were built between 2015 and 2024. They were designed for a Google that no longer exists, the one that ranked ten blue links instead of synthesizing an AI answer above them. They often look fine and might still rank reasonably well on traditional results pages. But the page they're competing on changed, and the foundation underneath them wasn't built for the new conditions.
What's typically missing:
- Slow mobile load times — frequently four to eight seconds. Performance signals heavily into how AI engines prioritize sources.
- Heavy JavaScript-rendered content. Many AI crawlers don't fully render JS, so the actual prose is invisible to them.
- No structured data. Without LocalBusiness, FAQ, Service, or Article schema, the engine can't confidently identify what the page is or who runs it.
- No llms.txt at the root, no explicit allowance for GPTBot / PerplexityBot / ClaudeBot. AI crawlers find nothing guiding them.
- Marketing-fluff copy rather than citable statements. "We provide world-class service" isn't a sentence any AI is going to quote in an Overview.
- Credentials buried in About pages instead of in machine-readable markup where E-E-A-T signals are read.
None of these were problems in 2019. They're real headwinds in 2026, and they compound.
What an AI-era website actually looks like
On the surface it looks polished. The aesthetic is editorial, the typography is intentional, the imagery is real. But the engineering underneath is doing different work than a 2019 site was built to do.
Sub-1-second global load
Cloudflare CDN, optimized assets, no framework bloat, performance budget enforced at build time.
Schema on every content type
Organization, LocalBusiness, FAQPage, Service, Article, BreadcrumbList. Each one helps an AI engine cite confidently.
llms.txt at the root
Formal guide for AI crawlers explaining what the site is, who runs it, and what content matters.
Statement-structured copy
Real claims with verifiable evidence. "We see 200 patients a week" beats "we provide trusted care."
Semantic HTML5 throughout
Proper h1 to h2 to h3 hierarchy. Real article, section, and nav tags. No decorative div soup.
Real credentials in markup
Author byline, address, phone, license number, founding year. AI engines weight E-E-A-T signals heavily.
Mobile-first by default
Most AI search queries originate on mobile. The site is engineered for the phone first, the desktop second.
ADA-aware accessibility
Proper alt text, semantic landmarks, keyboard navigation. AI engines use accessibility as a quality signal.
The cost of doing nothing
Every quarter, more search traffic gets absorbed by AI overlays and never reaches your website's analytics. The customer often finds their answer in the AI summary and moves on. They don't click. They don't see your booking form, your phone number, your hero pitch.
Businesses without AI-era websites will likely look fine for the first few months. Their analytics will show a slow decline in organic traffic. The pattern tends to compound: by the second half of the year, the decline becomes harder to ignore, and the gap to competitors who did adapt widens.
The window to migrate is open now. The cost of waiting is mostly invisible until it's not.
This is why we rebuild from the foundation, not just make the site prettier
Every website we build at World Tech Miami is engineered for the AI-search era from the foundation up. Not retrofitted. Rebuilt.
Sub-1-second global load on Cloudflare. Schema markup baked into every page type. Semantic HTML5 throughout. llms.txt at the root, properly formatted. Content structured for citation, because we work on the copy with you, not just the visual design. Performance budget enforced at build time. Mobile-first responsive across six breakpoints. Real author bylines, address schema, and credential markup in the HTML where AI engines can read it.
You can repaint a house with a cracked foundation, and it'll look better for a year. The foundation still cracks. Most "website refreshes" being sold in 2026 are repaints. We don't do repaints. The honest answer is almost always the same answer: a new build, engineered for the page your business is now competing on.