Technical SEO checklist for Next.js sites in 2026
What changes when AI engines start citing pages, and what to lock down at the Next.js layer to make sure yours get cited.
Most of the technical SEO advice published before late 2025 was written for a world where the only customer was Google's ranking algorithm. That world is gone. Today's brief covers the second customer, generative engines like Perplexity, ChatGPT, Gemini, and Google's AI Overviews, and what that adds at the Next.js layer.
What's actually new in 2026
Generative engines extract passages, not pages. A 40–60 word block that cleanly answers a sub-query now matters more than the whole page's keyword density. Schema markup graduated from "nice to have" to "the structured representation of your brand that AI models can rely on." And technical Core Web Vitals are still a baseline, but they're a baseline you must clear before any of the new work pays off.
Server rendering is the floor
Anything important must be in the initial HTML. We default every Next.js project to React Server Components and treat "use client" as a budget. Every island must justify itself. Forms, theme toggles, and motion-heavy elements are fine; navigation, headlines, FAQs, stats, and testimonial bodies are not. If a query rendering tool sees a blank page where your H2 should be, your site won't be cited even when the user is asking the exact question you'd answer.
next/image is non-negotiable for LCP
The single highest-impact thing on most sites is image discipline. Hero images get priority. Every image gets sizes and a fixed aspect-ratio container. AVIF + WebP output is automatic. Done correctly, this routinely cuts LCP by 600–900ms on 4G mobile, which moves Lighthouse Performance from the high 70s to the high 90s. Same content, same design, just shipped properly.
Self-host fonts via next/font
Google Fonts blocks render. next/font self-hosts and inlines metric-matching fallback fonts so CLS stays near zero. We use Fraunces, Geist, and JetBrains Mono, chosen for character, not because they're the safe default. Variable fonts with font-display: swap give us per-axis control without paying multi-file weight.
Comprehensive structured data
Every page declares what it is. Site-wide Organization schema. ProfessionalService on each service page. Article on each journal post. FAQPage on every page with a FAQ. BreadcrumbList everywhere. The goal isn't to chase rich results (those are a side effect). It's to give AI models a structured handle on your brand they can use without hallucinating.
A passage discipline
Every H2 in this article opens with a 40–60 word paragraph that answers the H2's question directly. That's the unit AI engines extract. After the answer, write at length. That's what Google still rewards. The two audiences live in the same document; they just read it differently.
Sitemap, robots, and the rest
Dynamic sitemap built from the file system. robots.ts allowing everything important, disallowing /api, /thanks, /og. RSS for the journal. manifest.webmanifest. None of this is exotic in 2026, but it's still missing on most sites we audit.
Wrote this. Want to talk about your site?