Scale SEO content updates with Claude Code

SEO teams often invest heavily in new pages while existing URLs quietly lose rankings, clicks, and revenue. A structured content update process addresses exactly that: it diagnoses content decay, protects built-up SEO equity, and makes results measurable. This guide outlines a 14-step workflow and how to scale it with Claude Code—from the Search Console baseline through CMS approval.

Why existing pages deserve priority

New pages start at zero and often feel more exciting editorially. A page that already ranks and drives revenue is a live asset: internal links point to it, schema is in place, and a historical baseline can be destroyed by careless full rewrites. Content decay rarely appears as a penalty or algorithm update; it shows up as slow drift—positions slip, impressions hold, CTR falls, and AI Overviews take visibility, often before analytics dashboards raise an alarm.

On an Antalya airport transfer page, the 56-day diagnostic before the update showed about 148,500 impressions, an average position of 14.89, a 1.49% CTR, and only 2,215 clicks. The page was not invisible; it was stale. A full rewrite would have been risky; a targeted delta was required that preserves equity and intervenes only where data proves gaps or staleness.

The manual 14-step process

Diagnosis and section tagging

Every update starts with a 56-day window in Google Search Console—wide enough for reliable data, narrow enough for one season. That baseline later becomes the measurement window and matches the longest test period of common SEO testing tools. What matters most are top queries that must be preserved, striking-distance queries in positions 5–20 with weak CTR, and zero-click queries with high impressions but almost no clicks—a signal of intent the page is not yet answering.

Then every section gets a label:

• Keep: still ranks and is accurate—do not touch
• Fix: right idea, stale execution—rewrite in place
• Remove: wrong, redundant, or harmful
• Add: the data shows something is missing

Competitors, keywords, and personas

Query data reveals concrete gaps: striking-distance keywords, private-hire intents, and comparison queries without matching content. Personas come from two sources—a sitewide GSC taxonomy over many months and synthetic query fan-outs for LLM-like questions. For Antalya, that produced four personas: shuttle shoppers, private-hire buyers, group travelers, and day-trippers. Without that differentiation, the page stays a brochure instead of an argument.

Local knowledge is refreshed: terminals, taxi ranks, scam patterns, tipping norms, peak congestion, and review themes. From that comes a clear angle: different travelers need different transfers—the page should surface the right answer instead of listing every option undifferentiated and hoping visitors self-sort.

Delta brief, execution, and SEO preservation

The delta brief covers only changes, not the whole page, and reads more like an engineering change request. Only Fix and Add sections are written—for example a persona chooser that routes existing information purposefully without inventing new facts. Then comes a fact-check including Keep sections: distances, prices, transit times, and terminal assignments must be current. Many AI refreshes only update surface wording and leave the fact base untouched—exactly what this process avoids.

Images are checked for accuracy, brand fit, and performance, such as WebP or AVIF, correct sizing, and lazy loading. SEO equity is preserved by default: URL slug unchanged, meta title kept if CTR is strong, schema extended rather than replaced, internal links checked in both directions. New UI components can be built via natural language and AI generation, server-rendered so LLMs can read the content without JavaScript.

Measurement against the baseline

Every update runs as a test with GSC and GA4 data. In the Antalya case, clicks per day rose by about 24%, position improved from 14.89 to 10.87, and CTR and query coverage moved with them. The key was the combination of diagnosis, angle, rewrites, and the new component—not a single lever. Only a locked baseline shows whether the publish truly worked or was luck.

Scaling with Claude Code

Done manually, the process takes about two focused days per page. Across thousands of URLs in multiple languages, that does not scale; the opportunity cost of a page at position 14 instead of 11 adds up quickly. So the 14 steps were mapped into Claude skills: Claude Code enforces the rules but does not replace human judgment. A dedicated project holds brand book, policies, and reference material so constraints stay in context.

Skills cover GSC baseline, audit tagging, competitor gaps, persona validation, delta brief, editorial writing, fact-checks, image audit, SEO preservation, and component generation. Hard gates block progress without validated personas, current local knowledge, and a defined angle—exactly what prevents generic AI copy. Deployment runs through a CMS connection via MCP: staging, diff against live, link and schema checks, human approval. Parallelizing multiple updates on the same agents reduced quality; therefore one update runs sequentially at a time.

Portfolio results and transferable principles

Across 59 completed tests, organic clicks rose on seven in ten pages; net gains exceeded 2,200 clicks on commercial transfer pages. Organic purchases finished about 25% and revenue about 20% above baseline. Content updates on commercial pages thus rank among the highest-ROI SEO actions—when treated as measurement projects, not maintenance.

What transfers is the system shape: humans set priorities and approve; automation enforces judgment. Preservation protects what already performs. Every change is measured against its own baseline. The starting point is the diagnostic, not the rewrite—and the next stage surfaces decay signals such as CTR softening and position drift before revenue already suffers.