I Had 110 Broken Blog Posts. AI Couldn't Fix Them Alone.

110 posts. That’s how many I had across my blog at one point. And I hadn’t touched most of them in months — not because I didn’t care, but because the task felt overwhelming.

Every time I logged into the CMS, I saw the list growing. Broken internal links, missing OG images, outdated meta descriptions, alt text that was never filled in. Individually, each issue was small. Collectively, they made the entire site feel like it was falling apart.

The Problem With Manual Audits

I tried the manual approach. I opened each post, checked the links, looked at the images, verified the meta tags. It was slow, tedious, and after about 20 posts my brain stopped caring. I was just clicking through, not actually reviewing.

The result? I fixed maybe 15 posts in a week and then stopped. The remaining 95 posts sat there, broken and embarrassing. Every time someone visited the blog, they saw the same neglected look — missing OG images when posts shared on social media, dead links in the navigation, and images with no alt text.

Enter AI — But Not the Way You Think

I didn’t just ask ChatGPT to “fix my blog.” That approach doesn’t work. The real breakthrough was treating AI as an audit tool, not a magic fixer.

Here’s the workflow that actually worked:

Step 1: Paste the URL into Claude and ask for an audit. No MCP connection needed at first. Just the URL and a clear instruction: “Audit this site for dead links, missing OG images, CMS gaps, and layout issues.”

Claude crawls the site and returns a structured report. For 110 posts, it categorized everything: what’s broken, what’s missing, what’s broken but intentional, and what’s just noise.

Step 2: Review the audit report. This is where my judgment comes in. Claude might flag every single missing alt text, but I know which posts drive traffic and which are dusty old pages nobody reads. I triage the list — fix the high-traffic ones first, batch the low-traffic ones, ignore the ones that don’t matter.

Step 3: Connect Claude to Webflow via MCP and apply the fixes. Once I have the prioritized list, I connect Claude to the Webflow CMS and hand it the task. It opens each CMS item, updates the OG image, fixes the alt text, corrects the meta description. It works through the list systematically.

Step 4: Spot-check before shipping. Every fix gets reviewed. I verify that the OG images are properly sized, that the links actually go somewhere, that the meta descriptions make sense. Claude does the work — I make the decisions.

What Actually Changed

The difference between the old approach and the AI-assisted workflow isn’t just speed — it’s sustainability. Before, I’d fix 15 posts in a week and then burn out. Now, the audit takes 20 minutes and the fixes take about 2 hours. I can maintain a clean site without spending days on it.

But here’s the part nobody talks about: the real bottleneck was never the tools. It was the decision fatigue of triaging 110 pages one by one. AI doesn’t get tired. It doesn’t click through 110 posts with glazed eyes. It does the grunt work, and I focus on the parts that require actual thinking.

The real bottleneck in fixing old content isn’t the tools — it’s the decision fatigue of triaging 110 pages one by one. AI handles the clicking; you handle the judgment.

What AI Can’t Do

I want to be clear about the limits because I hate blog posts that pretend AI is magic.

AI can’t decide which posts matter. It can suggest adding an OG image to every CMS post, but it can’t tell you that post #47 has 12 pageviews a month and doesn’t need the work. That’s your call.

AI can’t make design decisions. It can suggest a layout fix, but it can’t tell you whether that card should be two columns or three.

AI can miss context. A broken link might be intentional. An OG image might look wrong at the wrong aspect ratio. Your judgment is the final filter.

The Bottom Line

110 broken posts didn’t get fixed by AI alone. They got fixed by a workflow where AI handled the tedious parts and I handled the meaningful parts. The combination turned a weeks-long chore into a weekend project.

If you’ve got a neglected site, don’t try to fix it all at once. Audit it first, triage it second, fix it third. AI makes the audit and the fixes fast — but the triage is still yours to do.

AI suggests fixes, but you decide which ones matter and which are noise. That’s the partnership that actually works.