I Let AI Audit My WordPress Site — Here's What It Found

The Turtle Hub is my oldest content site. 110 published articles about turtle care — species guides, diet information, habitat setup, health advice. It gets 18,200 monthly visitors and has been through multiple rounds of content pruning. I thought it was in great shape.

Then I connected Claude to it via Connect My Site to AI and asked it to find every SEO problem it could. The results were humbling.

23 issues. On a site I actively maintain. On a site that had been manually audited just two months earlier.

Here's exactly what the AI found, why I missed these issues manually, and how I fixed each one.

The Setup

I installed Connect My Site to AI on The Turtle Hub, which turns the site into an MCP server. Then in Claude Code, I connected to it with a single command:

claude mcp add theturtlehub --transport http https://theturtlehub.com/wp-json/cmsta/v1/mcp

After OAuth authorization (takes about 10 seconds), Claude had access to all 24 read-only audit tools. I started simple: "Audit this site. Find every SEO problem you can."

If you want to set this up yourself, I wrote a full step-by-step MCP guide.

Finding #1: 7 Orphan Pages

The tool: find_orphan_content — finds pages with zero incoming internal links.

Claude found 7 articles that had zero internal links pointing to them from anywhere else on the site. These pages were essentially invisible to both Google's crawler and to visitors browsing the site. Google discovers pages by following links. If nothing links to a page, Google may crawl it from the sitemap, but it assigns minimal authority because no other page on your site vouches for it.

The 7 orphans included 3 species care guides that I'd published recently and forgot to link from the parent category hubs, 2 health articles that had been disconnected when I reorganized the internal link structure, and 2 older posts that simply fell through the cracks.

Why I missed it manually: There's no visual indicator in the WordPress admin that a post has zero incoming links. You'd have to check every page's incoming links individually, which nobody does on a 110-post site.

The fix: For each orphan, Claude identified the most relevant existing articles and suggested where to add links. With the Pro version, it inserted the links directly — finding natural anchor text in existing paragraphs and adding the link inline. All 7 orphans were connected within 15 minutes.

Finding #2: 4 Posts Missing Featured Images

The tool: find_missing_seo_elements — audits for missing featured images, excerpts, alt text, categories, and tags.

Four articles had no featured image set. Two were posts I'd imported from a different CMS years ago — the migration preserved the content but dropped the featured image metadata. Two were posts where I'd accidentally removed the featured image while editing something else.

Featured images matter more than people think. They appear in social shares (Facebook, Twitter, LinkedIn), in Google Discover cards, in RSS feeds, in related post widgets, and in many theme archive pages. A post without a featured image looks broken in all of these contexts.

Why I missed it manually: I rarely look at my own articles from the outside. In the editor, you don't see the featured image unless you specifically check the sidebar panel. In the post list view, featured images are hidden by default in most WordPress themes.

The fix: Uploaded appropriate images and set them as featured images for each post. Took about 10 minutes.

Finding #3: 12 Posts With Missing or Thin Excerpts

The tool: find_missing_seo_elements again.

12 articles had no custom excerpt set, meaning WordPress was auto-generating excerpts from the first 55 words of the content. Auto-generated excerpts are almost always bad — they get cut off mid-sentence, they include formatting artifacts, and they don't represent the article well in search results or archive pages.

Why I missed it manually: Auto-generated excerpts look "fine" at a glance. You don't notice they're auto-generated unless you compare them to deliberately written ones. And in WordPress, the excerpt field is below the editor — easy to skip when publishing quickly.

The fix: I asked Claude to read each article and write a concise, compelling excerpt (1-2 sentences). With the Pro version's update_post tool, it applied all 12 excerpts in one batch. What would have been 30+ minutes of writing and copy-pasting took about 3 minutes.

Finding #4: Keyword Cannibalization on 3 Topics

The tool: find_duplicate_titles — detects posts targeting the same keywords.

Claude found 3 pairs of articles competing for the same keywords. The worst case: two separate articles about "red-eared slider diet" — one titled "What Do Red-Eared Sliders Eat?" and another titled "Red-Eared Slider Feeding Guide." Both targeting essentially the same search intent, splitting my authority between them.

Keyword cannibalization is when your own pages compete against each other in Google. Instead of having one strong page ranking #3, you have two weak pages ranking #12 and #15. Google can't decide which one to show, so it shows neither prominently.

Why I missed it manually: When you're deep in a niche, similar titles feel different to you because you know the nuance. "What do they eat" and "feeding guide" feel like different articles when you're writing them. But from Google's perspective, they're targeting the same search intent.

The fix: For each pair, I merged the best content from both articles into the stronger one (the one with more backlinks and better rankings), then redirected the weaker URL to the merged article. This is exactly the merge step from my content pruning strategy.

Finding #5: Heading Hierarchy Issues in 8 Articles

The tool: analyze_content_structure — checks heading hierarchy and readability.

8 articles had heading hierarchy problems — jumping from H2 to H4 (skipping H3), using multiple H1 tags, or having H3 headings that should have been H2s. Proper heading hierarchy matters for both accessibility (screen readers use headings to navigate) and SEO (Google uses headings to understand content structure).

Why I missed it manually: Heading hierarchy is invisible when you're reading the content. Everything looks fine visually because CSS can style any heading to look like any other heading. The only way to catch this is to inspect the actual HTML structure.

The fix: Simple Gutenberg editor fixes — changing heading levels to maintain proper H2 → H3 → H4 nesting. About 20 minutes for all 8 articles.

What the AI Caught That I Couldn't

The pattern across all 23 issues is the same: these are problems that are invisible when you're looking at individual posts but obvious when you analyze the entire site programmatically.

No human is going to open 110 articles, check the incoming links for each one, verify the featured image exists, read the excerpt, compare titles for keyword overlap, and inspect the heading structure. You might do this for 10 posts during a focused audit session, but you'll never cover the full site.

The AI did it in about 90 seconds.

That's not a dig at manual auditing — I've been doing it for years and wrote a whole content audit guide based on the manual process. But the AI-connected approach catches things that manual auditing structurally cannot, because it can hold the entire site in context simultaneously.

The Fix Workflow

Here's how the full session went, start to finish:

• 0:00 — Connect. Install plugin, run the MCP add command, authorize via OAuth. 60 seconds.
• 1:00 — Audit. Ask Claude to run all available audit tools. It runs orphan check, missing SEO elements, duplicate titles, content structure, thin content, and stale content scans. 90 seconds.
• 2:30 — Review. Claude presents all 23 issues organized by type. I review each one and confirm which ones to fix. 10 minutes.
• 12:30 — Fix orphans. Claude finds appropriate anchor text in existing articles and adds internal links to all 7 orphan pages. 5 minutes.
• 17:30 — Fix excerpts. Claude reads each of the 12 articles and writes custom excerpts, then applies them. 3 minutes.
• 20:30 — Fix duplicates. I handle the content merges manually (these require editorial judgment), with Claude identifying which article in each pair is stronger. 20 minutes.
• 40:30 — Fix headings. Manual fixes in the Gutenberg editor for 8 articles. 20 minutes.
• 60:30 — Fix featured images. Upload and assign images for 4 posts. 10 minutes.

Total time: about 70 minutes to find and fix 23 issues across 110 articles. A manual audit of the same scope would have taken 6-8 hours, and I'd probably still miss the orphan pages and keyword cannibalization because those require cross-referencing the entire site.

What Changed After the Fixes

Within 3 weeks of fixing all 23 issues:

• The 3 cannibalized keyword pairs saw the merged articles climb an average of 6 positions in Google.
• The 7 formerly-orphan pages started appearing in Google Search Console's indexed pages (previously they were in the "discovered but not indexed" bucket).
• Crawl stats in Search Console showed Google was crawling fewer pages but spending more time on each one — a sign of improved crawl efficiency.
• Average time on site ticked up slightly, likely because the improved internal linking kept visitors moving between articles.

None of these were dramatic overnight wins. They were the kind of incremental improvements that compound over months. But that's exactly how site health works — it's the accumulation of many small fixes, not one magic bullet.

Try It On Your Site

The free version of Connect My Site to AI includes all 24 read-only audit tools. You can run every scan I described in this post without paying anything. Install the plugin, connect Claude, and ask it to audit your site. You'll probably be surprised by what it finds — I was, and I do this for a living.

If you want the write tools to actually fix the issues (adding internal links, updating excerpts, managing redirects), that's the Pro version at $49/year. But even with just the free read-only tools, you'll have a complete picture of your site's health in under 5 minutes.

For the full setup walkthrough, check my MCP setup guide. For broader audit strategy beyond AI tools, see the content audit guide.