tesifyUK
Start writing free
How it worksFAQBlogHelpContactLog in

Tag: proofreading policy

  • Which Language-Editing Tool Actually Understands a Thesis? Four Compared (2026)

    Which Language-Editing Tool Actually Understands a Thesis? Four Compared (2026)

    Before you choose between these tools, be aware that at one UK university the correct answer may be none of them. Cambridge’s guidance on proofreading doctoral theses states that a proofreader may not “Make grammatical, syntactical or stylistic corrections” — which is precisely what all four of these products exist to do. Here is the comparison, and then the policy question that governs it.

    The comparison

    Tool Built for academic writing? Trains on your text? Where your text goes Institutional licence
    Writefull Yes — “Using language models trained on millions of journal articles, Writefull’s edits are tailored to academic writing.” No — “We train our models on published papers instead.” Servers “located in The Netherlands, EU”. Discloses that several features send text to OpenAI, “kept for up to 30 days and not used for training purposes”. Yes — grants “all students and staff access”
    Paperpal Yes — “The AI writing tool built for academic research”, with checks against “1,500+ top journals, indexed in Scopus and Web of Science” No — “never used for model training… on every plan including Free” Journal Fit uploads “automatically deleted after 90 days”. States ISO/IEC 42001:2023 certification. Not established from published pages
    Trinka Yes — for those “writing research papers, theses, reports” No — “never used to train Trinka’s models or any third-party AI. Your IP stays yours.” “Processed content is not stored on our servers.” Retention is tier-gated. Enterprise tier offered
    Grammarly General-purpose, with an education product No — text “never sold or used to train models” “Amazon Web Services data centers in the US East region” Yes, an education plan; includes Authorship reporting

    The recommendation

    If your institution permits grammatical correction at all, use Writefull or Paperpal rather than a general-purpose checker — both are trained on research writing, both state that they do not train on your text, and both understand that a hedge is a feature rather than a fault. Between them, Writefull is the more transparent about data flow; Paperpal is the stronger on journal submission requirements.

    Everything after this section is about the two questions that actually decide it: what your regulations allow, and what these tools do to a doctoral sentence.

    The policy question, and why it is not settled in the UK

    Three UK universities publish rules for this. They draw three different lines, and the differences are not cosmetic.

    Cambridge draws it at identify versus correct. Its statement covers “the proofreading of all written work up to and including doctoral theses”, and its reasoning is about authorship: “In no cases should a proofreader edit a student’s writing (that is, check or amend ideas, arguments or structure), since to do so is to compromise the authorship of the work”. The prohibition list is explicit that a proofreader may not “Make grammatical, syntactical or stylistic corrections” or “Translate any part of the work into English”. It closes off the obvious appeal too: “Students should note that the use of a proofreader will not be accepted in mitigation of any deficiencies in their work.”

    Manchester draws it at mechanics versus expression, and its guidance is the only one of the three that names this technology directly — it “explains the use of third-party proofreading services, including AI tools and applications.” It permits fixing “spelling, punctuation, grammar, syntax, and general word usage”. It then lists what a proofreader may “highlight but not fix”: “Clarity of expression. Changes to verb tense or switching from passive to active voice. Logical flow and connections between sentences and paragraphs. Ambiguities or repetition.” Rewriting that alters “clarity, tone, or expression” is out, and the consequence is stated: “Exceeding limits is misconduct: Using support beyond what’s acceptable is considered plagiarism and academic malpractice.”

    UCL draws it at proofreading versus copyediting, and makes it a disclosure rule rather than a prohibition. Help “improving your grammar and writing structure” does not require acknowledgement, with a bracketed exception that does the real work: “[N.B. If GenAI is used to draft any text including copyediting (rather than proofreading), this must be declared].” The governing standard is accountability: “You must be able to explain clearly what work you did and what work you received human or computational support with.”

    Set those against the table above and the problem is obvious. Manchester’s “highlight but not fix” list — verb tense, passive to active, flow, repetition — describes a substantial share of what a modern editing tool proposes by default. Under Cambridge’s wording, the grammar correction itself is out. The permission question is therefore not one question with one answer; it is your institution’s question, and you have to go and read it.

    The wider integrity argument, including where the line sits for researchers writing in a second language, is covered in our piece on the AI academic English editor for doctoral researchers. This page assumes you have settled that and are choosing a product.

    A cautious claim flattened into an overstated one after an automated editing suggestion was accepted
    The doctoral-specific failure: a tool optimising for confident prose removes the hedging your data requires.

    What these tools do to a doctoral sentence

    At doctoral scale the risk is not that a tool misses a comma. It is that it makes your claims stronger than your evidence supports.

    Every general-purpose checker is tuned towards concision and confidence, because most writing benefits from both. Academic writing at the point of a finding does not. A suggestion that turns a qualified statement into a direct one is, in a results or discussion chapter, a suggestion to overclaim — and overclaiming is a thing examiners are documented as noticing. Accept suggestions one sentence at a time wherever a claim is being made, and reject anything that removes a modal verb or a scope qualifier.

    The second doctoral problem is vocabulary. Your field’s technical terms will be flagged as errors, and a tool trained on research writing does this less often than one trained on general English — which is the strongest practical argument for the two academic-specialist products over a general-purpose one.

    The third is that none of these tools reads structure. They work at the sentence and the paragraph; they cannot tell you that chapter five belongs before chapter four, or that your review is organised by source rather than by claim.

    Confidentiality, which matters more here than in most tool choices

    You are pasting unpublished research, sometimes participant data, into a commercial service. Three things are worth checking before you do.

    • Training. All four state that they do not train on your text. Writefull is the most specific about the alternative: “We train our models on published papers instead.”
    • Jurisdiction. Writefull states EU servers; Grammarly states US East. If your data is covered by an ethics approval or an industrial agreement with a location clause, that difference is the whole decision.
    • Third-party routing. Writefull’s disclosure that certain features send text to OpenAI is, notably, a point in its favour — the others do not publish an equivalent statement, which is not the same as not doing it.

    Our fuller treatment of what UK law actually requires here is in is it safe to put unpublished thesis data into an AI tool, which reaches a less alarming conclusion than most researchers expect, for different reasons than they expect.

    A note on price, stated honestly

    We could not establish comparable sterling pricing for these four products, and would rather say so than publish figures we cannot stand behind. Writefull publishes its plans in euros; Trinka publishes in US dollars and lists a free Basic tier; Paperpal’s plan pages did not expose prices to us at all, stating only that “We offer billing in local currencies”; Grammarly’s figures varied by the region we were served.

    The practical advice does not depend on the numbers. Check whether your university already holds a licence before paying for anything — Writefull states that an institutional licence “grants all students and staff access to Writefull’s complete suite of tools”, and researchers routinely buy individual subscriptions to software their library already provides. Ask your graduate school or subject librarian first.

    A university library help desk where a researcher checks which writing tools the institution already licenses
    The cheapest step in this comparison: find out what your institution already pays for.

    How this fits the rest of your toolchain

    A language editor is one of four tool decisions a doctoral researcher makes, and they are genuinely separate. Reference management is covered in Zotero versus Mendeley versus EndNote for a PhD thesis; literature discovery and appraisal in the best AI research assistants for PhD students; and the document itself in LaTeX versus Word. None of those overlaps with what is on this page, and a tool that claims to do all four is worth more scepticism, not less.

    Where Tesify sits, and where it does not

    Tesify is not on the comparison table above, because it is not a language-editing engine and it would be dishonest to score it as one. It is a drafting workspace: you write the thesis in it, with your sources and structure in one place, and the bibliography formats itself.

    That difference maps onto the policies quoted above rather neatly. The regulations are anxious about tools that alter text you have written; a workspace where you compose the text yourself sits on the other side of that line. But the standard UCL sets is the one to hold yourself to whatever you use: you must “be able to explain clearly what work you did and what work you received human or computational support with.”

    Draft your thesis in Tesify — free tier, your scholarship, the clerical layer automated. And whichever editor you pair it with, read your own institution’s proofreading policy first.

    Frequently asked questions

    Which language-editing tool is best for a PhD thesis?

    For thesis-scale academic English, Writefull and Paperpal are the two built specifically for research writing, and both state that they do not train on your text. But the tool matters less than your institution’s rule — check that first, because at some UK universities the answer is that none of them may touch your grammar.

    Do these tools train their models on my thesis?

    All four state that they do not. Writefull: “We never store your text, nor do we use your text or interaction with Writefull to train our AI. We train our models on published papers instead.” Paperpal says documents are “never used for model training… on every plan including Free.” Trinka says “Your IP stays yours.” Grammarly says customer text is “never sold or used to train models.”

    Where is my text actually processed?

    It varies, and for unpublished research it matters. Writefull states that its “servers are located in The Netherlands, EU”, while Grammarly “hosts data in Amazon Web Services data centers in the US East region”. Trinka states that “Processed content is not stored on our servers.”

    Does any of them send my text to a third party?

    Writefull is the most explicit about it. It discloses that several features “also use GPT”, and that for those “the text you select or enter to use these features is sent to the OpenAI servers (where it is kept for up to 30 days and not used for training purposes).” Read that before pasting confidential material into a paraphrase or summarise function.

    Is using one of these tools allowed on a UK doctoral thesis?

    It depends entirely on your university, and UK policies are not compatible with each other. Cambridge, Manchester and UCL draw three different lines. Read your own institution’s proofreading policy before you install anything.

    What does Cambridge actually forbid?

    More than most researchers expect. Its guidance, which covers “all written work up to and including doctoral theses”, states that a proofreader may not “Make grammatical, syntactical or stylistic corrections”. Read literally, that rules out the core function of every tool on this page.

    What does Manchester allow?

    A middle position, and the most workable one. Manchester says it is acceptable for a proofreader to “Fix issues like spelling, punctuation, grammar, syntax, and general word usage”, but only to “highlight but not fix” clarity of expression, verb tense, voice, logical flow and repetition. Its guide explicitly covers “AI tools and applications”.

    Do I have to declare that I used one?

    At UCL, it depends which side of a line you are on: help with “grammar and writing structure” needs no acknowledgement, but the guidance adds that if generative AI is used “to draft any text including copyediting (rather than proofreading), this must be declared.” The underlying standard is that you can explain what was yours.

    Does my university already pay for one of these?

    Possibly, and it is worth asking before you buy. Writefull states that “An institutional license grants all students and staff access to Writefull’s complete suite of tools”, and several UK institutions hold licences. Your library or graduate school will know.

    Will an editing tool weaken my hedging?

    It can, and this is the specific risk at doctoral level. A general-purpose checker optimising for concision and confidence will happily turn a carefully qualified finding into an unqualified one. Accept suggestions one at a time in any sentence that carries a claim.

    Can these tools replace a human proofreader?

    Not for the judgement calls. They are reliable on mechanics and unreliable on discipline-specific terminology, where a legitimate technical term is often flagged as an error. Whether either is permitted at all is a separate question your regulations answer.

    Is a tool that rewrites for me safe to use?

    That is where the risk concentrates. Manchester classes rewriting that alters “clarity, tone, or expression” as beyond acceptable support, and states plainly that exceeding the limits “is considered plagiarism and academic malpractice.” A tool that drafts sentences for you is a different category from one that flags a comma.

  • AI Academic English Editor for Doctoral Researchers: Polish the Thesis Without Losing Authorship (2026)

    AI Academic English Editor for Doctoral Researchers: Polish the Thesis Without Losing Authorship (2026)

    You have run the study, built the argument, and now the last barrier between you and submission is the sentence-level state of 80,000 words of academic English — which, if English is your second or third language, has been a silent double workload for the entire doctorate. The instinct to hand the manuscript to someone who will “fix the English” is understandable, and it is exactly where careful researchers get into trouble, because universities regulate that hand-off tightly. The legitimate version of the fix exists: an AI editor working inside your own draft, suggestion by suggestion, with you deciding every change. Here is the whole picture, rules included.

    The stakes are real in both directions

    Under-edited, a thesis pays twice: examiners distracted from the contribution by surface errors, and corrections lists padded with typographical items that a systematic pass would have removed — avoidable weeks, when minor corrections deadlines are typically measured in months. Over-edited, it pays worse: a thesis whose prose outruns the candidate’s demonstrated voice invites exactly the authorship questions a viva exists to probe, and “a service rewrote it” is not an answer any examination outcome improves on. The target is the middle: your sentences, systematically corrected, in a process you can describe out loud without discomfort.

    What the rules actually say

    Most UK universities publish proofreading guidance, and the architecture is remarkably consistent — the University of Edinburgh’s guidance on proofreading of student assessments is a usefully explicit example. Students are “normally permitted” to engage a proofreader — a friend, family member, paid professional or an online proofreading tool — to suggest minor changes that improve the readability of written English. What a proofreader may do: correct minor, localised issues of spelling, punctuation, grammar and syntax, with edits tracked; and comment on larger problems without fixing them. What a proofreader must never do: make untracked changes, rewrite text, or produce content on the student’s behalf — that is editing or ghost-writing, and it is academic misconduct. Students remain responsible for every change they accept, must keep before-and-after copies, and — the clause written for this decade — must acknowledge the use of generative AI where it operates inside a proofreading tool.

    Read those rules again and notice something: they describe a workflow, and it is precisely the workflow a suggestion-based AI editor enforces. Tracked, localised corrections; comments rather than rewrites for structural issues; the author deciding each change; a record of what happened. The policy anxiety about AI editing dissolves when the tool is built to operate inside the rules rather than around them — though your own institution’s wording, and your school’s, is the version that governs you, so read it before your final pass.

    Margin corrections on a printed thesis draft
    The permitted register everywhere: minor, localised, tracked — with the author deciding every change.

    What an AI academic editor does well at thesis scale

    Four things, all inside the permitted register. Consistency at volume: an 80,000-word document written over three years drifts — hyphenation, capitalisation of your own key terms, tense conventions between chapters — and a machine finds every instance where a human proofreader finds most. Academic register: flagging conversational phrasing, hedging that collapses (“may possibly suggest”), and the noun-stacked sentences that grow in second-language academic writing like ivy. Grammar in the long tail: article usage and preposition choice — the classic persistent errors for speakers of languages without articles — corrected suggestion by suggestion, which incidentally teaches the pattern as you accept or reject. And error classes humans fatigue on: duplicated words across line breaks, mismatched brackets, citation punctuation.

    What it must not do for you — and what you should not accept even when offered — is generate your argument’s prose: new paragraphs, restructured sections, “improved” versions of whole passages. That is the rewrite line every university’s guidance draws. Structural problems belong in a different, equally legitimate workflow: diagnose them, then rewrite them yourself — our guides to the discussion chapter and the notes-to-chapter writing-up method cover the two places structure most often fails.

    The workflow that keeps authorship yours

    First, finish the thinking before polishing the surface — editing a chapter whose argument will change is double work. Second, run the AI editor chapter by chapter, reviewing suggestions individually: accept the corrections, reject the rewrites, and notice the patterns in what it keeps flagging, because that list is your personal grammar curriculum. Third, keep the before-and-after versions, exactly as proofreading policies require of human proofreading — with a suggestion-based tool inside your own workspace this is automatic, and it is your evidence of a compliant process. Fourth, disclose according to your institution’s formula. A one-line acknowledgement — language-editing suggestions from an AI tool were reviewed and accepted or rejected individually by the author — costs nothing and survives any scrutiny. If a disclosure sentence feels uncomfortable to write, treat that as diagnostic: the process, not the sentence, is what needs changing.

    Why in-draft beats copy-paste editors

    The practical difference between an editor inside your writing workspace and a paste-into-a-box tool is bigger than it looks at thesis scale. In Tesify, the editor works on the draft where it lives — no fragmenting your thesis into pasted chunks, no reassembly errors, no wondering which version is current. The same workspace holds your chapter structure and your automatic bibliography, so the clerical layer — formatting, references, consistency — is handled by machinery while every sentence of argument remains yours: 100% written by you. And because your thesis is unpublished intellectual property, it matters that your draft is not being scattered across tools with retention policies you have never read; in Tesify your work stays yours and exports cleanly. Start free and run your next chapter through it — the suggestion-review loop takes an evening and teaches you your own error patterns faster than any style guide.

    A word to EAL researchers specifically

    Two things are true at once. Your English does not need to be native — examiners assess the contribution, and the QAA-style doctoral criteria nowhere require idiomatic elegance; clear, correct academic English is the standard, and it is reachable. And the workload asymmetry is real: you are writing at doctoral level in your second language while your anglophone peers write in their first, which is precisely why systematic tooling for the mechanical layer is fair rather than suspect. Spending your finite attention on argument instead of article placement is not a shortcut; it is correct resource allocation, and the disclosure line above makes it transparent.

    Frequently asked questions

    Is using an AI editor on my thesis considered misconduct?

    Not where it operates within your university’s proofreading rules: minor, localised language corrections, reviewed and decided by you, with the process documented and disclosed as required. Misconduct starts where generation starts — text produced on your behalf.

    Do I have to tell my examiners I used an AI editor?

    Follow your institution’s disclosure requirements — guidance increasingly asks for acknowledgement of generative AI even within proofreading tools, and Edinburgh’s does explicitly. A one-line acknowledgement is cheap insurance and reads as professionalism, not confession.

    Can I use a paid human proofreader instead?

    Usually yes, within the same rules that bind any proofreader: tracked minor corrections, no rewriting, and often a declaration. Human proofreading and AI editing sit under the same policy architecture; neither may author your text.

    What is the difference between proofreading and editing in these policies?

    Proofreading corrects surface errors — spelling, punctuation, grammar, syntax — locally and visibly. Editing changes substance: structure, argument, meaning, whole sentences. Policies permit the first with conditions and treat the second, done by any third party, human or machine, as misconduct territory.

    Will an AI editor make my thesis sound like it was written by AI?

    A suggestion-based editor correcting your sentences leaves your voice intact — that is the point of accepting changes individually rather than accepting a rewrite. If a tool’s output stops sounding like you, you have crossed from correction into generation; step back.

    Is my unpublished research safe in an online editor?

    Check any tool’s data handling before pasting in unpublished work. In Tesify your thesis is yours — not shared or published — and exportable; keeping your own backups remains good practice with every tool.

    My supervisor says my English is “not the problem”. Should I still edit?

    Probably — supervisors read for argument and forgive surface noise; examiners see the document cold. A systematic consistency pass before submission catches the layer supervisors stopped noticing in year two.

    Can AI editing fix my discussion chapter?

    No — if the discussion is weak, the problem is structural, not grammatical, and language polish will make well-written weakness. Fix the argument first with a proper chapter method, then edit the surface.

    How long does an editing pass on a full thesis take?

    Reviewing suggestions individually, budget an evening or two per chapter — front-loaded on the first chapter while you learn your own patterns, faster thereafter. It compresses dramatically compared with proofreading on paper precisely because the finding is automated and only the deciding is yours.

    Does good English actually change examination outcomes?

    Clean language does not pass a weak thesis, but noisy language taxes examiner attention and inflates corrections lists with fixable items. Removing that tax is one of the few pre-viva improvements entirely within your control in the final month.