Writefull or Statistical Review: What Each One Actually Fixes

You are about to submit, you have budget for one tool and two tabs open: an academic English editor in one, something to review the statistics in the other. The question that looks like it is in front of you is which one is better, and it is not that question, because they do not measure the same thing. The real question is which of the two reasons will send your manuscript back, because the one you leave uncovered is the one that costs you months. What follows separates the two territories using what Writefull's own site says, checked on 4 September 2026, and the order in which the two are worth using.

What Writefull does, in the words on its own site

So that none of this is second-hand: it was all checked on writefull.com and in their user guide on 4 September 2026, and if you are reading this months later, go and look again, because these tools move. Writefull presents itself as something to write, paraphrase and copyedit academic text, and its institutions page groups four pieces as the complete suite: Writefull for Word, Writefull for Overleaf, Writefull Cite and Writefull Revise.

In Word it gives you language suggestions, synonyms that fit the context of the sentence, paraphrasing, style transformations —more scientific, more concise, punchier—, splitting and joining sentences, summarizing, explaining a passage, and generating titles and abstracts from your own text. In Overleaf it does the same inside the LaTeX source and adds TeXGPT, which helps you write your LaTeX in there, plus table and equation generators you feed a prompt or an image and that hand you back the code. Writefull Revise takes the whole document and hands it back copyedited in Track Changes, ready to open in Word and accept or reject edit by edit.

What makes this work better than a general-purpose checker sits in one line on their institutions page: they use proprietary language models trained on research papers. That is not decoration. A general checker may underline significant and offer you important as a synonym; one trained on academic literature knows that in a Results section those are not synonyms. If you write in English and it is not your first language, that difference is measured in hours.

And there is a detail that in your situation weighs more than half the feature list: their front page says none of the texts or searches you process are stored or used for training. With an unpublished manuscript on the table, that sentence is part of the product, not small print.

The section feature exists, but it is a phrase bank, not an examination

I went looking for this one specifically, because it is the part people most often misread. Writefull has the Sentence Palette: per their user guide, over six hundred commonly used academic words and phrases, organized by sections and goals, that you can browse and paste straight into your text. Their own blog describes it as laid out across the sections a paper usually has —Introduction, Literature and Methods among them— plus a set of connectors.

It is a feature organized by section, yes. But notice which way it runs: from the tool toward you. It offers you how people usually write this study was the first to have explored when you are drafting a discussion. It does not read your discussion and tell you that you are calling a trend a contrast that came out at p = .09, or that in Results you reported an F without its degrees of freedom, or that the effect size in the table does not match the statistic two lines above it.

The Word add-in also has two buttons, Check Document and Check Paragraph, which run the language suggestions over the whole text or over the paragraphs you have selected. They check language. I read the full user guide looking for the opposite: there is no feature described that evaluates the content of a section, or that looks at statistics, p values or reference formatting.

None of that is a complaint. It is the product category. Writefull does not promise to review your statistics anywhere on its site, and you should not demand it, any more than you demand that a microscope weigh things.

Writefull Cite touches your bibliography, but from the opposite end

Here there is a feature that gets close to what worries you, and it deserves to be described properly because it is good. Writefull Cite, according to its own page, flags sentences that have no citation but need one, so you do not plagiarize by accident. It uses a model that evaluates the meaning of all the sentences in a text and assesses, one by one, whether each presents non-general knowledge that should be supported by a reference.

The problem it solves is real. Half of every discussion written at two in the morning asserts things nobody has supported, and that is exactly the material a reviewer opens the first round with.

But look at which side of the problem it covers. Cite finds where a citation is missing. It does not check that the citations you did include exist. These are opposite failures, and the second one is what sinks manuscripts now: a DOI that leads nowhere, a year that drifted, co-authors who never signed that paper, or plainly a reference a language model invented and you pasted without ever opening it.

Checking that is not a matter of reading more carefully: it is a matter of going to a database for each entry and seeing whether anything comes back. That is what Previum does, one entry at a time, against Crossref, OpenAlex, Semantic Scholar, PubMed, DOAJ, arXiv and OpenLibrary. If you would rather do it by hand, the whole procedure is in checking your references before you submit and it costs nothing to follow.

No language editor looks at the arithmetic of your results

Take a concrete case. You write t(48) = 2.10, p = .004. An English editor will leave that sentence spotless: comma in place, italics correct, tense in line with the rest of the paragraph. And the number is wrong. With 48 degrees of freedom, a t of 2.10 does not give .004; it gives .041 two-tailed. A zero slipped in while you were typing and it stays there until somebody looks at it.

The second case is worse because it is invisible. In that same comparison, with two independent groups of equal size, d can be recovered from the t and its degrees of freedom: twice the t divided by the square root of the degrees of freedom, which here comes to 0.61. If the table says d = 0.21 because you copied it from the output of a different analysis, you have a small effect printed next to a contrast that says medium. Nobody reading for style is going to catch that.

You do not fix this by reading better, you fix it by recomputing. Previum rebuilds the p values from the degrees of freedom with code, not with a language model, and checks the effect sizes against the F or the t you reported yourself; the prose of the report is written by a model, but around numbers it cannot touch. The list of what is worth checking before you press submit is in checking your analyses before you submit.

And the obvious thing, before this sounds like more than it is: this is not peer review. It is you evaluating your own manuscript before you send it, which is precisely the moment when these errors are still free.

The two rejections are nothing alike

Having your manuscript sent back over the English is annoying and cheap. The editor writes that the text needs review by a native speaker, you pay for an edit, you return it a week later and nobody has doubted your science. It is a toll, not a verdict.

Having it sent back because a reviewer found a p value that does not agree with its degrees of freedom is something else. From that moment on, everything else you wrote becomes suspect: if that number is wrong, how does he know the others are right? That rejection is not fixed with a copyedit; it is fixed by redoing the analysis and waiting out another journal's full cycle. Why Q1 journals reject you breaks down where manuscripts actually fall.

That is why the honest comparison between Writefull and a statistical review is not about quality, it is about category. One covers the toll, the other covers the verdict. The expensive mistake is believing you have covered both because you paid for one.

The order they are used in, and why it matters

Numbers first, language second. If you send a Results section off to be copyedited and then have to rewrite it because an effect size did not add up, you have paid twice for the same paragraph. And when an analysis changes it is never only the sentence that reports it: the abstract changes, the summary changes and sometimes half the discussion changes with it.

Second: before you reach for the card on the language side, check whether your university already has it. Writefull states that an institutional license gives all students and staff access to their complete suite —Word, Overleaf, Cite and Revise. If your library already pays, that half is free to you and the whole budget can go to the other half, which is the one nobody is covering right now.

Third, a way to work out which of the two is yours without theorizing about it: take the last reviewer report you received and count the comments. If most of them are about how something is phrased, your problem is language and Writefull solves it. If most of them start with the authors report and continue with a number, no paraphraser is going to help.

Frequently asked questions

Does Writefull check my p values or my effect sizes?

No. Checked on writefull.com and in their user guide on 4 September 2026, none of their product pages describes a feature that recomputes a p value from the degrees of freedom or that checks an effect size against the statistic it should come from. Their features are language features: correction, paraphrasing, style changes, summarizing, title and abstract generation, and phrases organized by section. If you need somebody to look at the arithmetic, that work has to be found elsewhere.

Does Writefull have a section-checking feature?

It has a feature organized by section, which is not the same thing. The Sentence Palette is a browsable bank of over six hundred academic words and phrases, arranged by sections and goals —Introduction, Literature and Methods among them— plus a block of connectors. It gives you language for writing the section; it does not evaluate the one you have already written. The Check Document and Check Paragraph buttons in the Word add-in run the language suggestions over the whole text or over the paragraphs you have selected.

Will Writefull Cite tell me a reference is fabricated?

No, it does the opposite, and that is also needed. According to its page, Cite flags sentences that have no citation but need one, assessing for each whether it presents non-general knowledge. It catches the gap, not the fabrication. Checking that the references you did write actually exist —live DOI, correct year, those co-authors and not others— is a different job, it is done against bibliographic databases, and that is where the ones a language model invented turn up.

How much does Writefull cost?

I am not giving you a figure, and that is deliberate. On 4 September 2026 their plans page loads the price in the browser and from outside all you get is a "Loading Pricing..." placeholder, and prices for these tools change several times a year. An invented price about a real company is not a slip. Look it up on writefull.com before you decide, and check first whether your institution already holds a license, because many do and do not advertise it.

If I use Writefull, do I also need a statistical review?

It depends on why your manuscripts come back. If the comments you get are about wording, clarity and grammar, Writefull covers you and you need nothing else. If they get into the numbers —a contrast that does not add up, an effect size that does not match, a reference the reviewer cannot find— no language tool touches that part, however good it is. They are two costs covering different things, not two versions of the same cost.

Does Previum also correct my English?

That is not what it does. It recomputes the p values from the degrees of freedom with code, checks the effect sizes against the F or the t you reported, and verifies the references one by one against eight bibliographic databases. The prose of the report is written by a language model, but around numbers it cannot change. For the English, Writefull or another language tool is still the right answer, and both can be used on the same manuscript.

Writefull and a statistical review are not competing: they are dividing up your manuscript. One handles how it sounds and the other handles whether it adds up, and there is an uncomfortable asymmetry between them, because imperfect English delays a publication while a number that does not add up can end up retracting one. Run the cheap test this week: take your last reviewer report and count how many comments are about how something is phrased and how many are about what the numbers say. That count tells you which of the two tools to buy better than any article will.