Scribbr or Statistical Review: What Each One Actually Corrects

Your manuscript is finished, you know your English is not a native speaker's, and you have had Scribbr's price calculator open in a tab for a while now. The question you are actually asking is not whether Scribbr is any good — it is very good at what it does, and this is not a hit piece — but whether paying for it removes the whole problem or only part of it. The short answer is that it removes a large and highly visible part, and leaves untouched exactly the part a methods reviewer is going to look at. What follows is the split: what Scribbr says it does, checked on its own site, and which question is still unanswered when the document comes back immaculate.

What Scribbr does, according to Scribbr

Scribbr sells academic proofreading and editing done by human editors. Its standard service, included in every order, combines the two classic stages: correcting surface errors (typos, spelling, punctuation, inconsistencies) and editing for style (grammar, syntax, tone and the conventions of your field). The service page lists what gets cleaned up: spelling and grammar errors, dialect inconsistencies, overuse of the passive voice, and subjective or inflated language. They take theses, PhD dissertations, manuscripts and academic articles, proposals, literature reviews and term papers, and return the edited document using Track Changes in Word with comments in the text. The calculator's turnaround options run from same-day delivery to seven days. All of this was checked on their site on 4 September 2026; prices and deadlines change, so if you are reading this much later, confirm them there.

On top of the standard service you can buy four add-ons, described here in their own terms. Structure Check ensures sections and chapters are structured and focused and the writing is free of redundancies. Clarity Check ensures ideas are presented clearly, arguments are consistent and the reader can follow along. Paper Formatting makes the document presentable: automatic tables of contents and other lists, page and heading numbering, style sheets for headings and body text, and the formatting and labeling of tables and figures; it is advertised at $1.95 per page, in an APA version or a house version. Citation Editing makes sure citations and references are consistent and meet the style guide's requirements: a human expert at $2.75 per source, or an APA-only automated checker at $9.95 that tests your citations against more than a hundred APA guidelines.

Let us state the obvious: this is a serious, well-run service. They say they have more than 800 editors, all of whom passed their in-house academy, which their site puts at a 2% acceptance rate; they match the editor to your field of study, their APA specialists default to the seventh edition, and for anyone not writing in native English that solves a real problem. A manuscript that takes effort to read reaches the reviewer with the mood already spoiled, and desk rejections on language do happen. Making sure language is not the first thing the report talks about is worth paying for. The only thing to be clear about is what exactly you are buying when you buy it.

Their APA content is a genuine reference, and that deserves saying

The other half of Scribbr is its knowledge base, and there they are excellent. Their APA citation guide is current with the seventh edition, distinguishes parenthetical from narrative citations properly, handles multiple authors and missing information (no author, no date, no page number), and comes with a free citation generator that supports APA 6 and 7, MLA and Harvard, plus its own pages for Chicago and other styles. They themselves say their knowledge base is used by millions of students and educators every year, and that is easy to believe: if you have ever searched for how to cite something awkward, you have ended up on one of those pages.

Their tables and figures guide is just as solid. It asks for the table number in bold above the table, a brief title in italics and title case below the number, no vertical rules, horizontal rules only where necessary for clarity, clear and concise column and row headings, numbers formatted consistently, and any relevant notes below. Seven rules, exactly what the manual says, well explained. If what you need is the version applied to psychology, with the specific case of the correlation matrix and its asterisks, that is in the APA 7 correlation table.

They even have a free statistics section, headed by a five-step guide to statistical analysis: write your hypotheses and plan the design, collect data from a sample, summarize with descriptive statistics, test with inferential statistics, and interpret. And it does not stop at the introduction: there are entries on degrees of freedom, effect size, choosing the right test, assumptions, regression and chi-square. It is honest, well-written material, and it helps a lot of people who arrive not knowing what a null hypothesis is. What it does is teach you what a p value is. What it does not do, and does not claim to do, is tell you whether yours has been calculated correctly.

Correcting a table's style is not checking that the table adds up

Go back to the table rules in the previous section and look at the only one that comes near the numbers: it asks that figures be formatted consistently, with the same number of decimal places, say. That is a rule about how the number is written, not about what the number is. And that holds across the whole system: APA style specifies what goes in italics, what carries a leading zero, how many decimals, where the note sits. A table can satisfy all seven rules perfectly, land in the journal looking exactly right, and still be wrong from top to bottom.

The standard example fits on one line. Your Table 2 says t(58) = 1.75, p = .04. Flawless formatting: the t italicized, the degrees of freedom in brackets, the p value with no leading zero and two decimals, aligned with the rest of the column. A copy editor has nothing to object to, and quite rightly, because nothing in the brief asks for a recalculation. The problem is that the two-tailed p for that t on 58 degrees of freedom is .085, and that asterisk should not be there. A methods reviewer works it out in ten seconds with any calculator, and reads the rest of the manuscript differently from then on.

The same goes for the rest of the arithmetic in a results section: the Cohen's d that does not follow from the F and the group sizes you published yourself, the partial eta squared above 1 that is impossible by construction, the N in the Method that does not match the degrees of freedom in the next table, the percentage that will not fit its denominator. These are calculation and transcription errors, almost always innocent, and they are invisible to any language review however good. They are also what produces the line in the rejection letter about the reported statistics not being consistent. The full list of what to check before you send anything is in check your statistics before submitting.

Consistent is not the same as real: the reference problem

Here Scribbr does get close. Their citation editing add-on commits to citations and references being consistent and meeting the style guide's requirements, and their automated APA checker detects missing references in your document and flags misuse of et al., punctuation and capitalization errors, missing information, and inconsistencies between citations and references. That is genuinely useful work: most people have three or four of those mismatches in any long manuscript.

But look at what a missing reference means there: you cite someone in the text and they do not appear in the final list. It is a check of your document against itself. The other question — whether that entry corresponds to an article that exists, with those authors, that year, that journal and that DOI — cannot be answered inside the document. You have to go out and ask the bibliographic databases, one reference after another. And a perfectly punctuated entry for a paper nobody ever wrote passes every style check in existence, because style is precisely the part it gets right.

This stopped being a theoretical problem the moment people started drafting with language models. A fabricated reference comes out immaculately formatted by construction: formatting is exactly what the model knows how to imitate. That is why the AI paper reviewer checks references one by one against seven separate databases (Crossref, OpenAlex, Semantic Scholar, PubMed, DOAJ, arXiv and OpenLibrary) rather than looking at how they are written. The manual procedure, if you would rather do it yourself, is in check your references before submitting.

Where language editing stops, in their own words

Scribbr publishes an honest table of the kinds of editing that exist and which ones they offer. Surface proofreading, yes. Copy editing, yes: grammar, syntax, style, tone and the conventions of the field, and — this is the interesting part — they say the editor also considers the internal logic of the text and flags any obvious contradictions. Line editing, yes, by combining the structure and clarity add-ons. And developmental editing, the kind that reorders content and decides what story the paper tells, no: they say plainly that it involves heavy rewriting and restructuring and that their editors cannot help with it.

That business of flagging obvious contradictions is as close as a language editor gets to your numbers, and it is not nothing. An editor who reads in the Results that the effect was significant and in the Discussion that it was not will mark it, and that saves manuscripts. But it operates on prose that contradicts itself when read. A p value that contradicts its own degrees of freedom cannot be seen by reading, however carefully: it has to be computed. Those are two different operations, and only one of them is editing.

None of this is a complaint. They define their remit and they deliver on it; the problem starts on the buyer's side, when you assume that reviewed by an expert covers everything a reviewer will look at. It does not, it never said it did, and the assumption only turns expensive when the journal replies.

What statistical review does, and what it also fails to do

The AI paper reviewer plays on the other half of the board. It recomputes p values from the degrees of freedom in code, not with a model, and tells you which ones do not come out. It checks effect sizes against the F or the t you wrote yourself. It verifies references against the eight databases from the previous section. The prose of the report is written by a model, but around numbers it cannot change, and that is the whole difference between a report that holds up and one that merely sounds good.

Now the other half of the honesty. It will not fix your English, format your document, build your table of contents or rewrite your Discussion. Nor is it peer review: it is the author's own assessment of their own manuscript before submitting it, and no journal editor is going to treat it as anything else. If you want to see exactly what the report looks like before deciding anything, there is a complete one published as the sample report.

On the practical side, for the record: there is a free Basic version and a paid Professional one, the data processing agreement is published with the sub-processors named, and the manuscript is not stored: it is read for the review and discarded. That last point matters more than it sounds when what you are uploading is an unpublished paper with clinical data behind it.

How to decide when you can only afford one

The order is not neutral, and this is the most useful advice in the article. If a check changes a number, it changes the sentence around it; and if that number crossed the significance threshold, it changes two paragraphs of the Discussion and sometimes the abstract too. Paying to polish prose you are about to rewrite is money burned. Numbers first, language second. If you are doing both, do them in that order even if you buy them on the same day.

The exception is clear enough: if your real problem is the language — a reviewer said the manuscript was hard to follow, or you were desk rejected on the writing — then editing is the right purchase and the numbers can wait for the next round. These are not rival services; they answer different questions and most manuscripts need both. What is not free is the order.

And there is a third case that neither one solves. If what you are missing is not checking the analysis but deciding it — which model applies, how to handle the missing data, whether that mediation survives your sample size — that is human work and you ask for it as such at statistical consulting. Neither a copy editor nor an automated checker is going to make that decision for you, and just as well.

Frequently asked questions

Does Scribbr review the statistics in my manuscript?

No, and it does not claim to. According to their own service description (checked on 4 September 2026), standard editing covers spelling, grammar, dialect inconsistencies, passive voice and inflated language, and the add-ons cover structure, clarity, formatting and citation consistency. Their copy editing does flag obvious contradictions in the text, which is a good thing, but that is reading. Recomputing a p value from its degrees of freedom, or checking a d against the author's own F, is not an editing task: it is a calculation, and nobody performs it by reading.

How much does Scribbr cost?

It depends on the service and the deadline, and their site shows prices in the currency of wherever you are browsing from, so the same rate may reach you in dollars or euros. The fixed prices shown on their site on 4 September 2026 are $1.95 per page for document formatting (both the APA and the house version), $2.75 per source for citation editing by a human expert, and $9.95 for the automated citation checker, which works with APA only. Text proofreading and editing is quoted through a calculator on their site based on word count, education level and the turnaround you pick, so there is no point in my giving you a figure: check it yourself, and note that it moves.

Does Scribbr check that my references really exist?

It checks that they are consistent with the style guide and that the in-text citations match the final list; the automated checker will even detect references you cite but never listed. All of that is your document against itself. Whether an entry corresponds to a real article, with those authors and that year, is checked outside the document, by asking the bibliographic databases. That is the check that matters now that people draft with language models, because an invented reference is perfectly formatted by definition.

Can I use both services?

Yes, and for most health sciences manuscripts that is the sensible answer: one makes sure the text reads well, the other makes sure the numbers hold. The only caveat is the order. Run the numbers review first, fix what it turns up, and only then send the language out for editing. Do it the other way round and, if a check knocks out a result, you will pay twice to edit the same paragraphs.

Is either of them peer review?

Neither. Scribbr is an editing and proofreading service and presents itself as one. The AI paper reviewer is the author's own assessment of their own manuscript before submitting it: it hands you the report a methods reviewer would have written so you can fix things before one writes it for real, but it does not replace the editorial process and cannot be presented as if it did. Peer review is organized by the journal and is not for sale.

Can I trust Scribbr's APA guides?

Yes. Their seventh-edition citation guide and their tables and figures guide are well made, current and free, and their citation generator handles the awkward formats without a fight. As a way of learning to write a correct table they are a perfectly good reference. What you will not find there, because it is not their business, is how to verify that the numbers you put inside that table are the ones your analysis produced.

The whole thing fits in one sentence: one review looks at how the manuscript is written and the other looks at whether what it says is internally true, and you need both because reviewers do both. If you want to know which failures on the numbers side come up most often, they are counted one by one in the most common manuscript errors.