How to build a website that ranks with Claude Code

Building a website has never been faster. You can describe a business to Claude Code and watch a polished, working site appear in minutes. But speed hides a trap: a site can look completely finished and still be invisible to search engines and AI answer engines. It ranks for nothing, gets cited by no one, and the traffic never comes.

The fix isn’t to slow down — it’s to give the agent the right methodology. Claude Code is excellent at the implementation of SEO and AEO; what it needs is a playbook for what to do and the data to prioritise it. This guide walks through both: the seven build phases of the build-site methodology, and where traffix.dev plugs in the measurement. We use Claude Code as the example throughout, but the same phases work in any AI coding agent — Cursor, Copilot, Windsurf, Gemini, and more — over the traffix MCP.

Why AI-built sites go invisible

It’s rarely the design or the code quality. AI-built sites underperform for mundane reasons: no keyword intent behind the pages, content that only renders client-side, missing structured data, heavy unoptimised images, no internal linking, and no analytics to learn from. None of these are hard to fix — they’re just easy to skip when the site already looks done. The methodology below makes them non-optional.

The seven build phases

Each phase maps to a reference in the build-site methodology. Run them in order — every one assumes the previous is in place.

1. 01

   Brief before you build

   Start from a one-page SEO brief, not a blank canvas: the primary keyword and its intent, the audience, the pages you need, and the entities you represent. This is what stops an AI build from producing a beautiful site about nothing in particular.

   build-site ref · seo-page-brief

2. 02

   Build on solid, crawlable HTML

   Server-render the content that matters. Semantic headings, real links, and text in the initial HTML — not content that only appears after JavaScript runs. Engines and agents both reward pages they can read without executing your app.

   build-site ref · agent-readiness

3. 03

   Mark up your entities

   Add Organization, WebSite, and page-type JSON-LD (Article, Product, FAQ, Breadcrumb). Structured data is how you tell engines who you are and what each page means — the difference between being indexed and being understood.

   build-site ref · structured-data

4. 04

   Pass a performance gate

   Treat Core Web Vitals as a build gate, not an afterthought. Ship images in modern formats at the right size, avoid layout shift, and keep the main thread free. Slow pages get demoted and rarely get cited.

   build-site ref · lighthouse-perf-gate

5. 05

   Make it answer-engine ready

   Publish an llms.txt, keep AI crawlers unblocked, and write answer-shaped content — a clear question as a heading, answered in the first sentence. This is what gets you quoted in ChatGPT, Perplexity, and AI Overviews.

   build-site ref · geo-answer-engines

6. 06

   Interlink with intent

   Connect pages with descriptive internal links so authority flows to the pages you most want to rank, and so engines understand your topic structure. A flat pile of pages ranks worse than the same pages well-linked.

   build-site ref · interlinking-pass

7. 07

   Instrument and measure

   Wire up Search Console and analytics from day one. You can't improve what you can't see — and the measurement loop is what turns a one-time launch into compounding rankings.

   build-site ref · instrumentation

Where traffix.dev fits

The methodology tells Claude Code how to build well. traffix.dev tells it what to work on next. It connects to your Search Console, measures every page’s performance, index status, and Core Web Vitals, runs keyword research, and produces one prioritised backlog — striking-distance keywords, content gaps scored by difficulty, low-CTR pages, and AEO/GEO gaps. Each item is exposed to your Claude Code over MCP with the matching build-site reference attached, so the agent fixes the highest-impact thing first, with evidence, on your own tokens.

Get started

Two commands connect your Claude Code to the methodology and the data. Install the plugin, mint a key, and ask the agent to pull your top opportunity and execute it.

Frequently asked

Can Claude Code do SEO?

Claude Code can implement SEO — titles, metadata, structured data, performance fixes, internal links, content — extremely well, because those are concrete coding tasks. What it lacks on its own is the data and judgement about what to do: which keywords are winnable, which pages underperform, what the current Core Web Vitals are. That's the half traffix.dev supplies over MCP, so the agent works from evidence instead of guessing.

What is the build-site methodology?

It's the methodology behind hundreds of ranked sites — a set of references covering SEO page briefs, structured data, answer-engine (AEO/GEO) readiness, Core Web Vitals performance gates, interlinking, conversion, instrumentation, and agent-readiness. Install the traffix plugin and your AI coding agent can pull it over MCP with get_methodology, so the agent builds to a consistent, proven standard.

Do I need to know SEO to use this?

No. The point of pairing the build-site methodology with traffix.dev is that the methodology and the prioritisation are encoded for you. You describe the business; the agent applies the playbook and measures the result. You stay in control and approve the work, but you don't need to be an SEO specialist.

Will an AI-built site actually rank?

Yes, if it's built right. Search and answer engines don't care that a site was built with an agent — they care that it's fast, crawlable, well-structured, genuinely useful, and authoritative. Those are exactly the things the build-site methodology enforces. The risk with AI-built sites isn't the AI; it's skipping the fundamentals because the site looked done.