How Our Paraphraser Performs - A Hands-On Look
July 6, 2026 · Khasan Parpiev · 5 min read
Alright, greetings everybody. This is Khasan from PT Technologies, and in this post I'd like to show you how our Paraphraser performs. I'm not a writer, so I've put together a YouTube video along with a walkthrough of what we saw. In the video I take real text through our Paraphraser and compare the result against a few other tools. Watch it and draw your own conclusion.
The point isn't to put down other paraphrasing tools - they do the job. It's that people using them should know there's an option that produces genuinely natural, easy-to-read writing. We built our engine (KapeVeronica) from the ground up, on a set of algorithmic patterns that are ours. The name is inspired by Resident Evil: Code Veronica - I'm a big fan of the franchise, and so is my brother Khusan, the co-founder. He might write something here himself one day.
How we set the test up
It mattered to us to keep this as unbiased as we could, so the starting passage was chosen to favour nobody.
We began with a plain 200-word passage on a biomedical topic, generated from a non-specific prompt - one that leaves out the detailed requirements a writer would normally work from: a particular structure, a set of points to cover, a target reader, a tone. Generic prompts produce flatter, more predictable prose, which makes a fair neutral baseline. If you start from text that is already well written, every tool looks good; if you start from something deliberately tortured, every tool looks bad.
Then the same passage went through several popular paraphrasing tools, ours included, with default settings on each. Same input, no cherry-picking between runs, no hand-editing afterwards.
In our reading, the rewrite that came back most natural while keeping the original meaning intact was ours. That's our judgement on one passage, and I'd rather you didn't take it on faith - which is what the rest of this post is about.
How to judge a paraphraser yourself
This is more useful than any claim I can make, because you can run it on your own writing in ten minutes.
Start from your own flat draft, not from a sample. Tools are tuned on generic prose. Your writing has habits - a subject you keep returning to, sentence shapes you favour - and how a rewriter handles those is what you actually care about.
Read the output aloud. This catches more than any metric. Machine-smoothed prose has a particular sound: every sentence lands at about the same length, every clause resolves neatly, and nothing ever interrupts itself. Human writing is lumpier. If the rewrite sounds like a brochure, it has flattened you out.
Check what happened to the specifics. This is where rewriters fail most often and most quietly. Take a sentence with concrete detail - three named factors, a figure, a date - and see whether they survived. Generalising "trade, cultural exchange and regional cooperation" into "these factors" reads more smoothly and has thrown away the content. Smoother is not the same as better.
Check the citations. If your paragraph had references, count them, then read each one against the sentence it now follows. A reference that survived but moved is worse than one that vanished, because nothing looks wrong. We wrote that up in full here.
Ask whether it still sounds like you. A rewrite you would not be comfortable putting your name to is a failed rewrite regardless of how it reads. The goal is your point, expressed more clearly - not an anonymous house style.
Run the same paragraph twice. A tool that returns something meaningfully different each time is making choices rather than applying a fixed transformation. Whether you want that depends on the job, but it is worth knowing which kind you are using.
What ours does and does not do well
It works best on one paragraph at a time, 50 to 150 words. That isn't an arbitrary limit. Hand over a whole section and the output averages toward a single register, which is exactly the flatness you were trying to get rid of.
It is built to leave in-text citations where they are while the language around them changes, rather than treating a reference as text to vary.
Where it will let you down: rewriting for flow can generalise a specific claim, and heavily technical passages with defined terminology are better left alone - a rewriter that swaps a defined term for a synonym has damaged the paragraph, however well it reads afterwards. Rerunning a paragraph usually returns a different rewrite, so if the first result isn't right, a second pass often is.
And it is a rewriting tool, not a writing one. It changes how your text reads. It does not know whether your argument works, whether your evidence supports it, or whether the paragraph belongs in the essay at all. Those are still yours.
One correction
In the video I mention the subscription costs $5.99. That's out of date - we lowered it to $3.99 a month. Why? Because, why not. Enjoy a tool that genuinely works, for even less.
You can try the Paraphraser on a paragraph now - the free plan covers 1,000 words a month, no card needed. If your draft feels flat and you want to know why before rewriting it, the Lexical Diversity checker points at the repeated words, the repeated sentence openers, and the runs of sentences that are all the same length.