How to make humanised text sound like you
Give humanisation sample texts you wrote yourself, in the same language and ideally of the same type as the text you are rewriting. From them, humanisation works out the length of your sentences and words, how much you vary sentence length, your punctuation and your most frequent connecting words, and it adds short excerpts to the instructions. Every humanisation that passes the checks becomes a sample by itself; you can upload your own earlier work with the Pro plan. Without samples, the rewrite aims at a generic human style, not yours.
What humanisation reads from the samples
From sample texts in the same language it works out the average length of a sentence and a word, how much you vary sentence length, how many commas, dashes and question marks you use per hundred words and which connecting words you use most often. The language model receives these figures as part of its instructions.
Short excerpts from three samples are added to that, so the model sees how you actually write, not just the numbers. Samples in another language are not used — an English text is rewritten according to English samples.
Which texts to use as samples
The best sample is a text you wrote yourself, without help from AI — a term paper, an article, a longer e-mail. Each one must have at least 250 characters, and a longer text gives more stable figures. If a model wrote the sample, the rewrite will move closer to the model again.
Samples of the same type as the text you will be rewriting work best. An academic paper has a different rhythm from a blog, and an average of the two fits neither. There can be at most twenty samples, and uploading your own takes the Pro plan.
How the samples fill up on their own
Every finished humanisation that passes the checks is added to the samples automatically. Even without uploading anything, the rewrite therefore settles over time on the style of your earlier rewrites. Once there are twenty samples, a new humanisation replaces the oldest of the humanisations, and uploaded work stays.
A humanisation used as a sample also carries the style of the rewrite, not just yours. If you want the text to sound as much like you as possible, upload your own work and delete the samples from humanisations that do not suit you.
FAQ
How many samples are enough?
The style can be worked out from a single one, but the figures from a single text fluctuate. Three to five of your own texts of the same type give a more stable picture.
Do you store the sample texts?
Yes, in your account, because the style is worked out from them. They are used only for your detection and humanisation, we do not train any shared model on them, and you can delete them at any time.
Do the samples help detection too?
Yes. The Style deviation detector compares the text with your earlier checks and uploaded samples, so uploaded work makes both more accurate.
Try it on your own text
Paste a text or upload a file and look at the score and at the specific sentences that came out suspicious.
Run a detection