A plain-text version of this site for AI assistants is at https://contengi.com/llms.txt, with the full product documentation in one file at https://contengi.com/llms-full.txt. Every page is also available as markdown by adding .md to its URL.
These are the same rules Contengi runs on every draft in the app, lifted straight out of its anti-slop standard and turned into plain code you can run on anything.
Zero-width spaces and joiners, byte-order marks, soft hyphens, directional marks, and Unicode tag characters, which can carry hidden instructions to the next AI that reads your text. Narrow and non-breaking spaces become normal spaces.
Every em dash becomes a spaced hyphen and ranges get a plain hyphen. Emojis go. Curly quotes and markdown symbols are optional, for when you paste into a plain-text field.
Corporate verbs get swapped for plain ones, so utilise becomes use and in order to becomes to. Throat-clearing like “it’s worth noting that” is deleted. Stock AI words like delve, tapestry, and seamless get highlighted for you to rewrite.
The “not X, it’s Y” contrast, fake-suspense questions, “No X. No Y. Just Z.”, a pile-up of lists in threes, sentences that all start with the same word, and a flat rhythm where every sentence runs the same length.
Free Claude skill
The same checks as this page, packaged as a skill for Claude. It strips the hidden characters and then does the part a regex can't, rewriting the flagged sentences while keeping your facts and your voice.
A cleaner fixes characters and words. Structure is where AI writing gives itself away, and fixing a straw-man sentence means rewriting it in your voice.
Contengi does that before you see the draft. Every writer starts with the anti-slop standard and your brand's rules, Claude edits its own draft, plain code scans for the exact rules, and Jev checks for the patterns a word list misses.
A slop check from a live Contengi run. Only the flagged sentence gets rewritten, everything else stays yours.
FAQ
No. Claude’s watermark is statistical, carried in which words Claude chooses, and Anthropic says it adds no hidden characters. Nothing that edits characters can touch it. The cleaner removes the hidden characters that other tools and copy-pasting leave behind, and highlights the writing habits readers spot.
Sometimes. In April 2025 researchers caught o3 and o4-mini adding narrow no-break spaces to long answers. Copying out of any chat window, PDF, or Google Doc can also pick up zero-width and non-breaking spaces. Paste a draft in and the cleaner counts them for you.
No promises there. Detectors and watermarks read word choice and statistics, and the cleaner leaves those alone. It’s built to make your writing better to read.
No. The cleaner runs in your browser with plain code, no AI model and no server. Close the tab and it’s gone.
Chatbots use them far more than people do, and readers have noticed. One em dash won’t sink a post, ten in a LinkedIn update will. The cleaner swaps each one for a spaced hyphen, which reads the same.
A rough count of AI tells per 100 words on a scale of 0 to 100, weighted so a “not X, it’s Y” sentence counts for more than a stray emoji. Under 20 reads clean. Treat it as a pointer to where to look, a human editor still has the final say.
Download the zip. In Claude Code, unzip it into ~/.claude/skills/ and ask Claude to clean a draft. In the Claude apps, upload the zip in your Skills settings. The skill runs the same checks as this page and rewrites the flagged sentences too.
Also free: the LLM visibility check, for seeing whether AI engines mention your brand.