Give your AI better instructions.
The less a draft looks like a template, the less a rewrite has to change. These are the habits detectors notice most, written as instructions your AI can follow.
Give your AI this link
Paste it into ChatGPT, Claude or any assistant that can read web pages, with a line like "Follow this guide when you write for me." It is plain text written for an AI to follow.
Assistants that cannot browse: use the short version, or paste the full guide from the link.
Or paste the short version
For custom instructions, a project prompt or a system prompt when the assistant cannot open links.
01 · Structure
Structure.
Detectors read the shape of a piece before its words. These moves are the strongest giveaways, and rewording never removes them.
Open with the most concrete thing: a number, a moment, the news or the ask. Do not announce what the piece will cover.
Not "Here's what I learned from my first year as a landlord." Instead: "The water heater died four months after closing. $1,180."
Why: Roadmap openers were among the most AI-scored sentences in every draft we tested.
Do not end with a recap, a "bottom line", key takeaways, a moral or a question asking readers to comment. Stop on a plain fact or what happens next.
Cut lines like "Six months in and I'm glad I did it" and "Have you tried this? I'd love to hear in the comments."
Why: Closing morals and comment bait scored as AI on their own, and in short posts they decide the whole score.
Tell things in the order a person would remember them, not in tidy parallel sections. Spend longer on one or two points and pass quickly over others.
Why: Paraphrase detectors recognise a draft by where its sentences sit. Rewrites that kept the original order were flagged; rebuilt ones passed.
Make paragraphs uneven: a one-sentence paragraph here, a long one there. Do not make every paragraph a topic sentence plus two supporting sentences.
Why: Even, essay-shaped paragraphs scored as AI even when every sentence was rewritten.
In posts under about 400 words, use no sub-headings. Never use label lines such as "The results:", "Lesson:" or "The reality check:".
Why: Sub-headings alone moved short posts from human to AI on our detectors.
Write lists as sentences. No bullet points, numbered lists or bold text in normal prose.
Why: List formatting is one of the plainest machine signals, and people rarely format a post that way.
02 · Sentences and words
Sentences and words.
Smaller signals, but they add up across a piece.
Vary sentence length a lot. Some very short. Others run on for a while because the thought needs the room.
Why: Uneven sentence length was common to the rewrites that passed GPTZero.
No em dashes, no semicolons, and no "It's not X, it's Y" constructions.
Why: These are among the most over-used patterns in model writing.
Do not list things in threes by reflex, and do not end sentences with an "-ing" clause that explains the point (", highlighting…", ", making it…").
Why: Both appear several times more often in model text than in human text.
Avoid hype words: crucial, leverage, seamless, game-changer, delve, navigate, landscape, robust, elevate, unlock, pivotal, journey.
Why: Detectors and readers both treat these as tells.
Do not write maxims or slogans ("You don't need to be the cheapest, you need to be the clearest"). Say the specific thing that happened instead.
Why: Punchy general lessons kept rewrites flagged even after everything else changed.
Use plain words a 9th grader knows.
Why: Plain wording reads as human and loses nothing in the rewrite.
03 · Facts and specifics
Facts and specifics.
Specifics are what make writing sound like a person, and they are what Rewrite protects.
Include real specifics: numbers, prices, dates, names, places, and details only this writer would know. Do not invent any.
Why: Rewrite checks numbers, dates and links in code and names and claims with a model review, so the specifics you give survive, or the change is flagged.
If a fact is missing, leave a visible placeholder like [NEED: price] instead of making one up.
Why: Made-up facts are the one thing Rewrite cannot fix, because it keeps your facts as they are.
State each fact once. Do not repeat the headline number in the closing.
Why: Repeated facts usually come from a recap, which reads as AI.
04 · Format and length
Format and length.
Some formats read as generated whatever the wording. Avoid them when you can.
Aim for at least 150 words. Very short texts give the rewrite too little to work with.
Why: Short texts were the hardest to move on every detector.
Write schedules and specs as sentences where you can, not rows of "Label: value" lines.
Why: Timetable and spec-list formats were flagged even after a full rewrite.
No emoji, hashtags or numbered thread markers (1/, 2/) unless the platform truly needs them.
Why: Templated social formats are among the most flagged genres.
Send the draft to Rewrite as it is. Do not run it through another paraphraser or humanizer first.
Why: Other tools often change facts. We found one changed four facts in a 250-word post.
This guide helps. The rewrite does the rest.
Following it gives the rewrite a cleaner starting point. It does not replace it: the rewrite rebuilds structure and wording and checks the facts against your draft. Some formats tend to be flagged whatever you do; see the docs guide.
Rewrite with your facts locked.
Try 300 words free ›