Every AI draft has the same rhythm. Timbro rewrites it.
Even sentence lengths. Connectives announcing a relationship the sentences already show. Nouns standing where verbs belong. Timbro rebuilds the prose and then checks the result against your original, so every number, quotation, date and citation survives exactly as you wrote it.
You can rewrite something without an account. Timbro makes no promises about AI detectors and is not built for them — that is answered plainly in the questions below.
- Facts are checked, not assumed
- Numbers, percentages, dates, quotations, citations and links are pulled out of your draft before the rewrite and compared against it afterwards — in both directions, so an invented figure is caught as well as a dropped one.
- Structure, not synonyms
- Sentences get merged, split and rebuilt. Nominalisations go back to being verbs. Scaffolding connectives come out. A thesaurus pass would leave the rhythm exactly where it was.
- You see every change
- Each rewrite comes with a paragraph-level diff, and paragraphs the model left alone are marked as untouched. Nothing is accepted on your behalf.
Before and after
The same passage, with the machine cadence taken out of it
A draft that says something real, written the way a model writes it: even sentence lengths, a connective on every opening, actions buried inside nouns. Read the rewrite, then read what changed and why.
Natural mode, default controls. 195 words in, 182 out.
The implementation of a four-day working week at Perpetual Guardian in 2018 represents a significant development in the ongoing evolution of workplace policy. Moreover, the trial demonstrated that a reduction in scheduled hours does not necessarily result in a corresponding reduction in output. Additionally, the company reported that productivity remained stable while self-reported stress levels declined from 45% to 38% (Haar, 2018). Furthermore, these findings have proven influential in subsequent policy discussions across a range of jurisdictions.
It is important to note that the trial was conducted with 240 employees over a period of eight weeks, which represents a relatively limited sample. Nevertheless, the results provide valuable insight into the potential benefits of alternative scheduling arrangements. In addition, the methodology relied primarily on self-reported measures, which introduces the possibility of response bias. Consequently, further research utilising objective performance metrics would be beneficial in order to validate these preliminary conclusions.
In conclusion, the Perpetual Guardian trial offers a compelling illustration of the ways in which a reduction in working time can be implemented without an adverse impact on organisational performance. However, it is essential to acknowledge that the generalisability of these findings remains an open question.
Three of the changes, and why they are improvements
01
The connectives come out
Moreover, … Additionally, … Furthermore, …
nothing takes their place
Three consecutive sentences opened by announcing a relationship the sentences already had. Nothing replaces them: the order of the claims carries the argument, which is what the connectives were pretending to do.
02
A noun goes back to being a verb
a reduction in scheduled hours does not necessarily result in a corresponding reduction in output
Cutting scheduled hours did not cut output.
The original states one action twice as a noun, and needs seventeen words to do it. The claim is identical. Six words, two live verbs, and it is now short enough to land against the long sentence in front of it.
03
Two sentences become one
It is important to note that the trial was conducted with 240 employees over a period of eight weeks, which represents a relatively limited sample.
The trial was small, though: 240 employees over eight weeks.
“It is important to note that” instructs the reader to pay attention instead of telling them anything. The judgement and the figures now share a sentence, and both figures come through untouched.
What did not change
Every one of these was extracted from the draft before the rewrite and compared against the result afterwards. A missing figure is an omission; a figure in the rewrite that was never in the draft is a fabrication. Both are reported, and the paragraph responsible is regenerated rather than shipped.
- Perpetual Guardian
- 2018
- four-day
- 45%
- 38%
- (Haar, 2018)
- 240 employees
- eight weeks
An illustrative example written for this page, not a live rewrite. Open the editor to run this on your own draft.
Capabilities
What it actually does
Six things, each of which is a decision about how a rewriting tool should behave rather than a feature bullet.
Meaning is verified, not trusted
Numbers, percentages, currency, dates, quotations, citations, links and acronyms are pulled out of your draft with pattern matching rather than judged by a model, then compared against the rewrite in both directions. Something dropped is an omission. Something present that was never in your draft is a fabrication. Longer documents get a second pass that records the claims, conclusions and causal links the rewrite has to keep.
Structural rewriting, not synonym swapping
Sentences are merged, split, reordered and reopened. Nominalisations become verbs again. Connectives that only announce a relationship the sentences already carry are removed rather than replaced. Swapping words for longer words changes none of the measurements that make prose read as machine-written, which is why Timbro does not do it.
Seven modes that actually differ
Natural, Professional, Concise, Student, Academic, Casual, and Write Like Me. Each mode ships its own control baseline, its own generation guidance, and its own list of signals to tolerate. Academic keeps every hedge and qualification, because in academic writing those are content, not padding — stripping them would produce a worse piece of writing, not a better one.
Five controls, on a scale that means something
Formality, vocabulary, sentence variation, conciseness and voice, each from 0 to 100. Every dial has five bands with concrete, behavioural instructions behind them, so 30 and 70 produce genuinely different writing. A dial only reaches the model when it has moved meaningfully away from the mode's baseline.
Paragraph-level control
When one paragraph is the problem, fix one paragraph. Select it and ask for shorter, simpler, more natural, more professional, more casual, or a structurally different version. Only that paragraph is rewritten; the paragraphs on either side go along as context and come back untouched.
A diff you can actually read
Changes are shown per paragraph rather than as one document-wide smear, so a paragraph the model left alone reads as left alone. Additions and removals are separated by colour, background and typography together, never by colour alone.
Write Like Me
A voice profile is measured from your own writing
Paste in a few things you have already written — an email you were happy with, an essay, a post, anything at least 60 words long. Timbro runs them through the same measurement engine it runs your documents through, and the numbers that come out are the profile.
A profile has two halves, deliberately kept apart. One half is measured and reproducible: run the same samples again and you get the same figures. The other half is a description a model writes after reading the samples — how you join ideas, how you open a paragraph. The prompt uses the first as concrete targets and the second as guidance, and the app tells you which is which.
It describes writing, not people. There is no personality inference, no demographic guess, and no claim about who you are. And it never copies: sentences and signature phrases from your samples are explicitly out of bounds, as is their subject matter.
When the samples are thin, it says so
Confidence is named rather than scored, because a percentage would imply a precision the assessment does not have. Below 150 words in total there is not enough writing to measure and Timbro will not build a profile at all.
- Rough sketch150–400 words
- Enough to measure, not enough to be confident. The profile says so, and the rewrite leans on it more lightly.
- Decent match400–900 words
- The structural habits are stable across samples. The finer punctuation habits may not be.
- Strong match900+ words
- Enough writing for the measurements to hold steady from one sample to the next.
Measured from the text
Reproducible. Every figure comes back the same on a second run.
- Median sentence length, and how far your sentences spread around it
- How often you use contractions, first person and second person
- Your rate of em dashes, semicolons, colons and parentheses
- How often you open two sentences the same way
- Typical paragraph length, and how much it varies
- Hedges, intensifiers and nominalisations per thousand words
Described after reading
A model’s account of the samples, presented as judgement rather than fact.
- How you move between ideas
- How your paragraphs tend to open, and how pieces tend to end
- The character of your vocabulary, without listing phrases to copy
- What you conspicuously never do
Privacy
Your writing stays yours
A tool you paste unpublished work into has to be specific about what happens to it. Four commitments, each of which is a property of how the software is built.
- You can delete any of it
- A single rewrite from your history, one writing sample, a whole voice profile, or the account itself. Deleting the account takes the documents, samples and profiles with it rather than leaving them orphaned in a table somewhere.
- Nothing you write trains anything
- Your documents and your writing samples are not used to train or fine-tune a model, and they are not sold or shared. They exist to do the rewrite you asked for and to be there when you come back for it.
- Full documents are never logged
- Logs carry word counts, stage timings, signal scores, error codes and a request id — the things needed to tell whether the system is working. The text of your document is not among them, by rule and not by habit.
- Nothing sits at a public URL
- There are no share links and no public pages for your documents. Every read is scoped to your account in the query and again in the database's own row-level policies, so one failure is not enough to expose anything.
What that cannot cover: rewriting means sending your document to the model that rewrites it, and there is no version of this product where that does not happen. That is the one place your text goes. Timbro can also run with no database configured at all, in which case there are no accounts, nothing is stored server-side, and voice profiles live in your own browser.
Questions
The ones worth answering honestly
Will this beat an AI detector?
Timbro makes no such promise, and it is not built for that. It is not tested against detection tools, it does not tune anything toward their scores, and no rewrite it produces comes with a claim about how any of them will read it.
That is partly a design choice and partly arithmetic. Those systems are unreliable in both directions — they flag writing done entirely by hand and they miss writing done entirely by a model — so any promise about them would be a promise nobody can keep.
If you are somewhere that requires you to disclose AI assistance, running a draft through Timbro does not change that obligation. It is a tool for making prose read better, and that is the whole of what it claims.
Is this just a paraphraser?
No. A paraphraser swaps words and leaves the sentences standing where they were, which is exactly the part that reads as machine-written. Uniform sentence lengths stay uniform. Connectives stay stacked. Every measurement the analysis stage cares about comes out unchanged.
Timbro works on structure: merging sentences that were making one point across two, splitting the ones carrying three, varying how sentences open, turning nominalisations back into verbs. Then it checks the result against your original, which a paraphraser has no reason to do.
Can it change my facts?
Every number, percentage, sum of money, date, quotation, citation, link and email address is pulled out of your draft before the rewrite and compared against the result afterwards — in both directions, so something invented is caught as well as something dropped. Those comparisons are string comparisons, not a model being asked whether it behaved.
When something fails, the paragraph responsible is regenerated with the specific problem named, then re-checked. On longer documents a second pass looks for the drift string comparison cannot see: a hedge quietly removed, a claim strengthened, a causal link reversed.
That catches the failure modes it can name, which is most of them. It is a check rather than a proof, so every rewrite arrives with a paragraph-level diff and your original still sitting beside it.
How long a document can it handle?
Your plan sets the ceiling: 5,000 source characters in a single rewrite on Free, up to 25,000 on Premium. Anything over it is refused outright, with the measurement and the limit both named, rather than quietly cut short and rewritten in part.
Documents long enough to need it are split at paragraph boundaries and rewritten in sections that each carry the surrounding context, then reassembled in order. Long pieces are usually better handled a section at a time anyway: you can actually read what changed.
How is a character counted?
By the length of the text you submit, spaces and punctuation included. That is the entire billing unit — source characters in, counted once — so you can measure a draft before you press the button and know what it will cost you.
What is deliberately not counted: everything the model writes back, the system prompts, the pass that measures your draft, the pass that verifies the rewrite against it, and any repair pass a paragraph needs. Those are our cost rather than yours. A paragraph that had to be regenerated twice costs you exactly what it would have cost had it come back right the first time.
Usage is recorded only once a rewrite has produced a result, so a request that fails costs nothing. The Free plan allows 20,000 characters a month, and every allowance resets at the start of the calendar month.
Do I need an account?
No. Without one you can rewrite up to 2,000 characters at a time, held in check by rate limits rather than by a monthly meter — enough to decide whether the output is any good. A Free account raises that to 5,000 characters per rewrite, adds a monthly allowance of 20,000 characters, and keeps your recent rewrites in a history you can go back to.
Timbro also runs with no database configured at all. In that mode there are no accounts, nothing is stored on a server, and voice profiles live in your browser.
How is Write Like Me different from a tone setting?
A tone setting is a label somebody chose. A voice profile is a set of measurements taken from writing you actually did — your median sentence length and how far your sentences spread around it, how often you use contractions, how you punctuate, how your paragraphs tend to open.
Those numbers go into the prompt as concrete targets rather than adjectives. And when there is not enough writing to measure confidently, the profile says so instead of pretending: it is labelled a rough sketch, and the rewrite leans on it more lightly.
A profile has to be stored somewhere, so it is one of the things the Free plan does not include: Pro keeps up to 3 profiles and Premium up to 10, for people who write in more than one register.
Can I use it for coursework?
Timbro rewrites text you already wrote. It does not write it for you, it does not do your research, and it will not add a claim your draft did not make.
Whether you may use a rewriting tool on an assignment is a question for whoever set the assignment. Many institutions have rules about AI assistance and some require disclosure; follow them. Timbro makes no claim about academic integrity systems and is not designed to interact with them.
What if the rewrite is worse than my draft?
Then keep your draft. Your original is never overwritten, and nothing is accepted on your behalf.
You can also fix one paragraph rather than accept or reject the whole document: select it and ask for something shorter, simpler, plainer, or just structurally different, and only that paragraph is rewritten. Moving the controls and running it again is cheap.
Try it on something you have already written
Paste in a draft you are not happy with and read the diff. That is the only honest way to judge a rewriting tool, and it takes about a minute.