10% most likely human - I guess my bad writing makes me a human

by: Artur Dziedziczak

December 7, 2025

AI detector on my blog post

A couple of microblog entries ago I started to use LLMs to parse my thoughts and check grammar issues within content I created. After like 2 entries, I gave up on that as I noticed that it does not reassemble who I am.

Every time I used duckduckgo AI (“DuckDuckGo AI Features.”) feature to check grammar, it always parsed the whole text and changed my unique, awkward and fresh way of writing.

So now, instead of using LLMs to parse grammar, I use LLMs to check if my posts look like they are written by AI. Here is the summarization of one of the tools (“Free Grammar Checker.”) I used:

10% most likely human

  • Informal, personal details and emotional content suggest human authorship.
  • Colloquial language and expletive (“shitshow”) feel spontaneous and not templated.
  • Minor grammar and word-choice quirks (e.g., “would like to use,” awkward phrasing) indicate natural writing.
  • No signs of uniform polish, overly generic phrasing, or repetitive scaffolding typical of AI output.

Sources

“DuckDuckGo AI Features,” n.d. https://duckduckgo.com/#chat

DuckDuckGo AI chat which I used to fix my writing. Sadly I could not handle how generic and sloppy newly created content was.
.

“Free Grammar Checker,” n.d. https://andreasjhkarlsson.github.io/jekyll/update/2023/12/27/4-billion-if-statements.html

Free tool which I now use to check grammar errors. It does not replace my writing and focus on finding grammar and spelling errors.
.