Your session has been flagged for unusual activity.
Please confirm you're human to continue.
AI detectors flag writing that sounds predictable, and they flag plenty of human writing too. The reliable fix isn't a trick. It's making AI-assisted text genuinely yours: your ideas, your details, your voice.
What this guide won't promise
No tool can guarantee a result on every detector, and we don't claim to. If your school, journal or employer doesn't allow AI for a piece of work, rewriting the text doesn't change that. Where disclosure is required, disclose.
Detectors don't know who wrote something. They estimate how likely it is that a language model produced it, mostly from these four signals.
Models pick the most likely next word. Text where every word is the obvious choice has low "perplexity" and scores as AI-like.
People mix short and long sentences. AI drafts often repeat one length and shape, which detectors call low "burstiness".
"Delve into", "it is important to note", "in today's fast-paced world". Phrases models overuse are an easy tell.
General claims with no examples, numbers, names or opinions read like they could have been written about anything.
A detector score is a probability, not proof. The same text can score very differently in two tools, and careful human writing is flagged more often than most people expect.
That's why "beating the detector" is the wrong goal. Writing that carries your own thinking holds up when someone actually reads it, and when someone asks how you wrote it.
How our AI detector works and how to read a score61%
of TOEFL essays by non-native English writers were labelled AI-generated by common detectors in a 2023 Stanford study (Liang et al.).
July 2023
OpenAI withdrew its own AI text classifier, citing its low rate of accuracy.
~4%
sentence-level false positive rate reported by Turnitin for its own AI writing detection.
These steps make text read like a person wrote it, because a person did the thinking. They also happen to address every signal above.
Write your main points and your angle first, even as rough notes. Ask the AI to help with structure or wording, not with what to think.
A real example, a number from your work, a customer's words, a mistake you made. Specifics are the fastest way to sound like a person.
Follow a long sentence with a short one. Like this. Break up paragraphs that all have the same shape.
Delete openers and fillers that add nothing (see the list below). Say the point directly.
AI hedges by default ("can", "may", "it depends"). Where you have a view, state it, and give your reason.
If a line sounds like nobody would say it, rewrite it. Verify every number, quote and citation against its source.
Notes, outlines and version history show how the work came together. If you're ever questioned, they're better evidence than any score. And where your school or employer asks you to say you used AI, say so.
| Instead of | Try |
|---|---|
| In today's fast-paced world… | Start with your actual point. |
| It is important to note that… | Delete it and state the fact. |
| Delve into / dive deep into | Look at, explain, test |
| Leverage / utilize | Use |
| Plays a crucial role in | Matters because… (say why) |
| Furthermore / moreover | Also, and, or nothing at all |
| In conclusion… | End on your strongest line. |
Product update email
AI draft
"We are thrilled to announce that our latest update leverages cutting-edge technology to significantly enhance your overall experience. Furthermore, it is important to note that these improvements play a crucial role in streamlining your workflow."
Rewritten
"Exports are faster now. A report that took a minute to build now takes a few seconds, so you can stop waiting and get back to work."
The writer added the one detail that matters (export time) and cut the filler.
Opinion paragraph
AI draft
"Remote work offers numerous benefits and challenges. On one hand, it can increase flexibility. On the other hand, it may reduce collaboration. Ultimately, the right approach depends on various factors."
Rewritten
"Our team went remote two years ago and I wouldn't go back. Meetings got shorter. What we lost was the quick question over a desk, so now we keep one open video room every afternoon."
A clear view, a real experience and one concrete fix replace the "on one hand" hedge.
Humanize mode does the mechanical part of steps 3 and 4: one model takes your draft and a second model rewrites it with varied rhythm and without stock phrases, keeping your meaning and facts. Steps 1, 2 and 5 are still yours.
Click the model selector next to the send button and choose AI Team.
Fill slot 1 and slot 2 with models, select Humanize and click Go Live.
Send the text itself as your message. Add one line about audience or tone if you want.
Open Responses to compare what each model did, check the facts, then add your own details.
For text that matters, use Conversation mode instead: slot 1 writes, the other models push back on tone and weak wording, and slot 1 writes a final version with that feedback. Humanize mode is part of MultipleChat Pro.
Stay calm. A score alone isn't proof, and detector makers, Turnitin included, say it shouldn't be the only evidence.
Ask which tool and which passages were flagged, and what score it gave.
Show your process: notes, outline, earlier drafts and the document's version history.
Offer to talk it through. Explaining your argument and sources in person is strong evidence you wrote it.
Be honest about any AI help you did use, and point to the rules you followed.
Let two models clean up the draft, then add the details only you know.
Try MultipleChat
