Updated 2026-07-13 · 14 min read
Vibe-coded sites rank under the same rules as hand-coded sites. The difference is failure probability: an AI builder can make a page look complete while shipping an empty HTML shell, repeated metadata, inaccessible links and no internal architecture. SEO must be part of the founding specification, not a final “optimize this” prompt.
Open View Source or fetch the URL with JavaScript disabled. The title, primary heading, body copy and important links should exist in the response. Google can render JavaScript, but rendering costs time and introduces failure modes; many AI and specialist crawlers do not execute a complete browser session. Prefer static generation or server rendering for content intended to rank.
Each route needs a concise title, useful meta description, canonical URL and one descriptive H1. Dynamic templates must derive these from real page data rather than repeat a generic brand string. Do not index filter states, empty searches or thin parameter combinations.
Group content into hubs and clusters. A guide hub links to tool comparisons, tutorials and technical SEO; each article links back to the hub and laterally to the next useful answer. Use real <a href> links with descriptive text. Buttons with JavaScript navigation can be invisible or semantically meaningless to crawlers.
The sitemap should include canonical, indexable URLs only and update when content changes. Submit it to Search Console. Robots rules should block private/admin/API surfaces, not CSS or JavaScript required for rendering. A sitemap helps discovery; it does not make weak pages index-worthy.
Use Organization and WebSite consistently, Article for editorial guides, BreadcrumbList for visible paths, and CollectionPage/ItemList for real directories. Structured data is not a place to add invisible claims. Stable @id values connect entities across pages.
AI tools add animation and client libraries enthusiastically. Set explicit budgets: properly sized images, dimensions to prevent layout shift, limited fonts, delayed non-critical scripts, no autoplay media above the fold unless it carries real value. Measure LCP, CLS and INP on deployed pages, not only the local preview.
Search engines and answer engines need information worth citing: original measurements, reproducible methods, direct answers, concrete examples and sources. Rephrasing the ten current top results produces no reason to rank. Norrly’s strongest SEO asset is not the phrase “vibe coding”; it is the measured dataset behind hundreds of real sites.
Before launch, crawl the site locally and fail the build on missing titles, duplicate canonicals, broken internal links or invalid JSON-LD. On day one, submit the sitemap and inspect the homepage, one hub, one article and one dynamic page. During week one, watch server logs and Search Console for crawl errors. During weeks two to four, compare submitted versus indexed URLs and improve pages Google crawled but declined to index rather than publishing more of the same.
Measure search performance by cluster, not individual keyword rank: impressions to the guide hub, clicks to comparison/tutorial pages and whether visitors continue into the product. Rankings without useful next actions are vanity traffic.
All public content routes must render meaningful HTML on the server or at build time. Give each route unique title, description, canonical, H1 and Open Graph tags. Use semantic landmarks and crawlable href links. Generate sitemap.xml from canonical routes and block private/API routes in robots.txt. Emit valid page-specific JSON-LD that matches visible content. Optimize images and fonts for Core Web Vitals. Add automated checks for duplicate titles, missing canonicals, broken internal links and invalid schema.
An llms.txt file can provide a useful map for AI tools that choose to read it, but it is not a ranking switch and does not replace crawlable pages, internal links or structured data. Treat it as documentation layered on top of a sound website.
Not inherently. They become weak when they ship client-only content, repeated metadata or poor architecture. Rendering and implementation choices determine SEO quality.
Google can render JavaScript, but server-rendered or static HTML is more reliable and accessible to a wider set of crawlers.
There is no evidence that it directly improves Google rankings. Use it as an AI-readable directory, not as a replacement for established SEO foundations.