How to use De-slop
Slop is the writing that gives a language model away: em dashes as the default pause, "delve", "tapestry" and "testament", "not just X but Y", every list padded to three items, sentences that end on "highlighting its significance", a closer that restates the paragraph. De-slop sends your text to a language model with an editing brief built from three catalogues of AI tells, Wikipedia's Signs of AI writing, Simon Willison's LLM cliché highlighter and the Claudisms banlist, and returns the same text with those patterns removed: the vocabulary, the staged reveals (“here's the thing”, “turns out”), the “no X, no Y, no Z” chains, the totalizing “the only thing that matters”, the faux-thoughtful “sit with that” and “that's not nothing”, the performative “let's be honest”, stacked rhetorical questions, repeated sentence openers, and the “despite these challenges” outlook. It is an edit, not a rewrite: every fact, number, name, quotation and URL stays as written, the language of the input is kept, length stays within about fifteen percent, and code and Markdown that carry real structure are left alone. Decorative formatting goes: emoji bullets, random bold, headings on a short note, horizontal rules. Tick Show what changed to see a word-level diff of the edit. Because each run costs money, text is capped at 8,000 characters and a visitor gets a handful of runs per hour; the mechanical tools on this site remain unlimited.
- Copy the text or table you want to fix.
- Paste it above and press De-slop. The edit takes a few seconds.
- Copy, download, share, or embed the finished result.
Frequently asked questions
What counts as slop, and what does De-slop change?
The patterns editors use to spot machine-written text, drawn from Wikipedia's Signs of AI writing, Simon Willison's LLM cliché highlighter (38 patterns) and the Claudisms banlist. Vocabulary such as delve, tapestry, testament, pivotal, robust and seamless. Negative parallelisms like "not just X but Y" and "no fluff, no filler, no jargon" chains. Staged reveals: "here's the thing", "the punchline is", "turns out". Totalizing claims: "that's the whole point", "the only metric that matters". The faux-thoughtful voice: "sit with that", "worth naming", "that's not nothing", "I keep coming back to". Performative honesty: "let's be honest", "to be clear". Lists padded to three; "serves as" for "is"; "experts agree"; trailing "highlighting its significance"; "despite these challenges, time will tell"; stacked rhetorical questions and repeated sentence openers; em dashes as the default pause; emoji bullets, random bold, headings on short text; chatbot openers and sign-offs. Each is replaced with the plain statement or cut.
Does it change the meaning or the facts?
No. The editing brief tells the model to keep every fact, number, name, date, URL and quotation exactly as written, to keep the language of the input, to keep the author's register and point of view, and to stay within about fifteen percent of the original length. It is an edit, not a summary and not a rewrite. If the text has no tells, it comes back unchanged.
Will the result pass an AI detector?
That is not what the tool is for and no promise is made. De-slop removes the patterns human readers notice, which are also the patterns most detectors weight heavily, so scores usually move; but detectors are unreliable in both directions and this tool is an editor, not an evasion tool. Read the result and make it yours.
Is my text uploaded?
Yes, for this tool only. The text is sent to OpenAI's API (GPT-5.6 Terra with reasoning off) for the single edit and the result is returned to your browser. Fix My Formatting does not store the input or the output, and OpenAI handles API requests under its API data-usage terms rather than the consumer ChatGPT terms. Every other tool on this site still runs entirely in your browser.
Why is there a limit?
Every run costs real money for the model call, so the tool takes up to 8,000 characters per run and allows a small number of runs per hour per visitor, with a site-wide daily cap. When a limit is reached you get a clear message and can come back later. There is no account and no payment; the mechanical tools on this site remain unlimited.
How is this different from the AI Formatting Humanizer?
The AI Formatting Humanizer is mechanical: it normalizes punctuation, quotes, invisible characters and pseudo-fonts and never touches a word. De-slop rewrites wording and sentence shape, which is what actually reads as machine-written. Run De-slop first, then the humanizer if any smart quotes or hidden characters remain.