AI Content Auditor Reviewed My Blog: Average Score Was 55/100

My Blog’s AI Content Got a Reality Check, It Scored 55/100

Hey there! So, I always suspected there was something off with my writing, but actually seeing the score was pretty shocking. I decided to build a tool to see just how much AI was sneaking into my work.

Like many of us nowadays, I rely on AI assistants quite a bit - maybe too much if I’m being honest. DeepSeek, ChatGPT, Gemini - you name it, I’ve probably thrown my drafts at them. The result? I managed to publish 69 articles in just about a month! Impressive, right? Well, sort of. The quality wasn’t quite what I’d hoped for.

You know that telltale AI vibe? Too many em dashes, similar-sounding sentences, that unmistakable robotic tone. I needed some hard data to back up my hunch, so I put together a scoring tool.

How the Scoring Works

My little content grader checks each article across seven key areas:

1. Sentence rhythm - It looks at how much the length of sentences varies. Human writing tends to flow in a natural ebb and flow, while AI often feels more uniform.
2. Word variety - It counts how many unique words are used. AI can be quite repetitive in its vocabulary.
3. AI buzzwords - It flags those overused phrases, like “delve deeper,” that just scream machine-written content.
4. Transition words - It tracks unnecessary fillers like “however” and “moreover.”
5. Dash overload - It counts the number of em dashes per 1,000 words.
6. Readability - It uses a standard Flesch score to see how easy the text is to understand.
7. Sentence structure - It checks for variation in sentence patterns.

Each area gets a score from 0 to 100, which combines into an overall grade. The tool also sends the text to DeepSeek for line-by-line feedback - these suggestions don’t affect the score but provide specific fixes.

If you’re looking for some technical SEO help, I’ve got a separate technical SEO guide that might interest you. This tool is all about writing quality, though.

The Harsh Reality

After checking all 69 posts, the results were a bit painful to look at:

Overall grades:

• Average: 51/100
• Median: 55/100
• Best: 72
• Worst: 23

The patterns revealed my bad habits:

• 52 articles (76%) were repeating the same words way too often - terms like “guide,” “fix,” and “step” appeared more times than I’d like to admit.
• 51 posts (74%) had robotic sentence rhythms with hardly any variation.
• 50 articles contained a total of 253 unnecessary em dashes.
• 462 specific issues were found, ranging from filler sentences to entire paragraphs that felt AI-generated.

The feedback from the AI was pretty blunt, pointing out empty transitions and redundant explanations that just added fluff without real value.

My Fixing Process

To tackle these issues, I built a couple of helper scripts:

• apply_fixes.py - This script matches suggested edits to the original MDX files using five fallback methods (since the AI views rendered HTML, not raw markdown).
• clean_emdash.py - This one replaces em dashes with proper punctuation while ensuring code blocks and frontmatter stay intact.

After running both scripts:

• I made 350 improvements across all articles.
• I replaced 247 em dashes (mostly changing “Header, text” to “Header: text”).
• I fixed 9 meta descriptions that were missing punctuation.
• I ignored 112 suggestions that were either duplicates, mistakes, or related to frontmatter edits.

About 25% of the AI suggestions didn’t hit the mark, but the rest caught issues I’d missed through several rounds of editing. The tool really helped highlight repetitive phrasing that I’d started to overlook.

This isn’t my first time automating fixes - my server monitoring system and database audit tool went through similar processes.

My Em Dash Problem

And here’s the kicker: I discovered a whopping 253 unnecessary em dashes spread across 50 articles! One GA4 guide alone had 18.

The fixes were pretty straightforward:

• Bold headers now use colons instead of dashes.
• Side thoughts get actual parentheses.
• Sentence continuations receive proper punctuation.

The cleaned-up versions now read more intentionally - dashes have become deliberate choices rather than just filler.

What I Learned

This whole experiment taught me more about my writing than any editor ever could. Before seeing the scores, I would have guessed my articles averaged around 70-75. The actual average of 55 was quite the wake-up call.

The seven metrics revealed patterns that I just couldn’t see while writing. It’s hard to spot vocabulary repetition when you’re in the thick of it, but this tool tracks every word across every sentence.

Out of the 462 suggested fixes, I implemented 350. That 25% rejection rate reflects my desire to keep my unique voice while also accepting valid criticism.

“Write better” is a bit vague, right? But “boost your lexical diversity above 0.35” gives me something concrete to work towards. Now, I run every new post through the auditor before hitting publish.

Recommended Reading

• Technical SEO Checklist: Site optimizations beyond content.
• GA4 for Content Strategy: Tracking if these changes actually help engagement.
• Backlink Building Guide: Next steps after a content cleanup.
• Automated Server Monitoring: Another AI-powered tool I built (and fixed).
• API Cost Tracker: How I monitor expenses for these AI tools.

What do you think about using AI to audit writing? Have you tried something similar? I’d love to hear your thoughts!

References & Further Reading

• Google Search Central Documentation
• Wikipedia: Artificial Intelligence