How a Claude Agent Collapsed 13 Redirect Chains on BlitzMetrics

A Claude agent scanned all 254 active redirects on BlitzMetrics.com, detected 13 internal redirect chains, and collapsed every one of them to a clean single-hop redirect — entirely inside a browser session, with no direct database access.

Understand the Assignment

After the database restore incident that wiped the RankMath redirect table, the BlitzMetrics team rebuilt the redirect list from scratch. During that rebuild, an important secondary problem was left unaddressed: redirect chains. A redirect chain is when URL A points to URL B, and URL B points to URL C. Instead of one hop, the browser and Googlebot have to make multiple requests before reaching the final destination. Each extra hop wastes crawl budget and dilutes link equity.

The Basecamp task thread flagged this explicitly: scan remaining redirects for any other chains. The prior session had already confirmed two Knowledge Panel redirect chains were already correctly pointing to their final destination. The remaining 252 active redirects needed a full audit.

This is part of the same ongoing GSC and redirect health work documented in How a Claude Agent Fixed 149 BlitzMetrics 404s in One Session and the indexing diagnosis that preceded it.

Step-by-Step Process

Step 1 — Read the Basecamp thread. The agent read the full task thread to understand what had already been done, what was in progress, and what remained. The key items: Knowledge Panel redirects were confirmed done; a full chain scan was still pending; a content audit and 33 noindex page review were also in the queue.

Step 2 — Fetch all 254 redirections via JavaScript. The RankMath Redirections admin page paginates at 20 per page. Rather than click through 13 pages manually, the agent used JavaScript fetch() calls against the admin HTML, parsing each page to extract source slugs and destination URLs. Three pages of results were needed (the list was filtered to show 100 per page). This produced a complete From → To map of all active redirects.

Step 3 — Run chain detection algorithm. The agent built a slug-keyed lookup table from the From → To pairs. A chain exists when the destination URL of one redirect contains a slug that is itself a source in another redirect. The algorithm followed each chain recursively until it reached a destination that was not itself a source — the true final destination. Thirteen chains were identified.

Step 4 — Fix each chain via the Rank Math edit form. For each chained redirect, the agent navigated to the edit page by searching for the source slug in the RankMath redirections list, clicking Edit, then updating the destination field. Because the RankMath redirections form is a React-controlled component, plain JavaScript assignment does not trigger a state update. The agent used the React native input value setter technique: Object.getOwnPropertyDescriptor(HTMLInputElement.prototype, 'value').set.call(input, newURL), followed by dispatched input and change events. Then clicked Update Redirection.

Step 5 — Verify each fix. After each save, the URL returned to the redirections list page — the standard success signal for a RankMath redirect update. The agent visually confirmed the updated destination appeared in the list for each slug before moving to the next.

The 13 Chains Fixed

Every chain traced back to one of three final destinations: a BlitzMetrics article about personalized gifts, a Local Service Spotlight career page, or the active-listening post. Here is the complete list:

Critical Decisions

Decision 1 — Fetch all redirects via JavaScript instead of paginating manually. The RankMath redirections admin shows 20 per page by default. Manual pagination across 13 pages would take 30+ minutes and risk missing entries. The agent instead constructed fetch() calls to load 100 per page, pulling all 254 redirections in three requests. This enabled reliable chain detection without human error.

Decision 2 — Use slug-based chain detection, not URL-based. Redirect destinations are stored as full URLs or as relative paths depending on whether they are internal or external. Matching on slugs (stripping protocol and domain) provided a consistent comparison method across both internal and external redirect entries.

Decision 3 — Leave the deactivated “if-youre-cold-calling-dont-do-this” redirect alone. This slug appeared in the chain detection output because an old redirect pointed to it. However, inspecting the redirect showed it was already deactivated, and the slug is a live active post on the site. Fixing the chain would have overwritten a redirect that should not be active. The agent skipped it and noted it for human review.

Decision 4 — Use the React native value setter, not standard DOM assignment. The RankMath edit form is a React-controlled component. Setting input.value = newURL does not trigger React state updates, so the form submits blank. The correct technique is Object.getOwnPropertyDescriptor(HTMLInputElement.prototype, 'value').set.call(input, url) followed by synthetic input and change events. This is the same approach documented in the 404 recovery session.

Decision 5 — Verify Knowledge Panel chains separately before running the batch scan. The Basecamp thread specifically flagged two Knowledge Panel URLs as still in progress. Rather than rely on the automated scan to catch them, the agent checked those two slugs first and confirmed they were already pointing to the correct final destination. No fix was needed, and the automated scan confirmed this.

Effort and Cost Comparison

Agent cost estimated at Claude Sonnet 4.6 pricing ($1.50/M input, $7.50/M output). Token usage estimated from session complexity.

What the Agent Could and Could Not Do

Handled autonomously: Reading the Basecamp thread; fetching all 254 redirections across multiple paginated admin pages; building the chain-detection algorithm; resolving every chain to its true final destination; navigating to and updating each redirect via the Rank Math edit form using React-aware JavaScript; verifying each fix; identifying the deactivated redirect exception.

Required human input: WordPress admin credentials (provided via pre-authenticated browser session); judgment call on whether any chains pointed to incorrect destinations (agent verified against Basecamp thread guidance and left all final destinations matching what Hezekiah had documented); featured image selection; final publish approval.

Information Ingestion Inventory

Guidelines Compliance Scorecard

This session is a direct application of the recursive loop Dennis describes in How Meta Articles Let My AI Agents Document and Improve Themselves: the agent executes a task, documents it, and the meta article becomes the training record for the next agent that needs to audit and fix redirect chains. For the SEO rationale behind why redirect chains hurt indexing, see How to Do Better Internal Link Building for Free with Your Favorite AI Agent.