Short answer: proofread a thesis last, after the argument, structure and references are settled, and do it in separate passes: consistency, then grammar and punctuation, then academic style, then numbers and cross-references, then one slow read of a clean copy. Before you use any tool or outside help, read your university’s rules, because they often limit what a proofreader may change, and some of them cover AI.
Check your university’s rules first
“Proofreading” means different things at different institutions, so don’t assume. Three UK examples show how far the rules can vary:
- Swansea University’s proofreading policy has proofreaders of research students’ work identify spelling and typographical errors and highlight overly complex or unclear wording. They may not change any words beyond correcting those errors, rearrange text or alter an argument. It also urges extreme caution with more sophisticated grammar and paraphrasing tools.
- The University of Birmingham’s guidance on dissertations lets AI advise on spelling, punctuation, grammar and syntax, and on shortening long sentences without changing content. It can’t be used to develop ideas or arguments, or to correct factual accuracy. Students must tell their supervisor and include a statement of that use.
- The Open University’s doctoral guidance says “substantive and substantial” generative AI use must be declared in the thesis, and its section on declarations asks for generative AI use to be declared more generally. It gives a sample declaration for proofreading, and says output must be reviewed to ensure the meaning and accuracy haven’t changed.
These are examples, and your own department’s policy is the one that counts. Look for rules on three things: what a proofreader may change, whether AI tools are allowed for it, and how to declare them. If the answer isn’t clear, ask your supervisor in writing. Our guide on how to disclose AI use in an essay or thesis has statements you can adapt.
When to proofread
Editing and proofreading are different jobs. Editing asks whether the chapter works: is the argument sound, is the evidence enough, is anything missing? Proofreading asks whether the language is correct and consistent. Do it too early and you’ll polish paragraphs you later delete.
Leave a gap between finishing the draft and starting. Even a day or two makes errors easier to see, and a thesis deadline should allow for more than one round, along with time to act on your supervisor’s comments.
Five passes, one job each
You can’t check commas and terminology in the same reading. Give each pass one job and use the search function wherever it helps.
- Consistency. Build a short style sheet: UK or US spelling, hyphenation (“data set” or “dataset”), capitalisation of key terms, abbreviations and how they’re introduced. Then search for the variants. Check that one concept keeps one name.
- Grammar and punctuation. Watch for comma splices, subject-verb agreement after long noun phrases, and pairs like “affect/effect” and “principle/principal”. Spell-checkers often miss these, because the wrong word is still a real word.
- Academic style. Replace colloquial words and contractions. Split sentences a reader can’t follow in one go. Check first-person use against your field’s conventions. Keep your hedging exactly as strong as your evidence.
- Numbers and cross-references. Compare figures in the text with your tables. Check figure and table numbering, “see Chapter 3” references, the contents page, and that every in-text citation has a reference-list entry and vice versa.
- Final read. Read a clean copy in a different format, such as a PDF or a printout, or use text-to-speech. Your eyes correct mistakes they’ve seen many times, and a change of format helps break that habit.
Before and after: what corrections look like
Here’s an invented passage with typical problems, followed by a corrected version and a record of the changes.
Before: “There is a lot of differences between the groups, this suggests that the intervention may effect outcomes. I don’t think motivation was the only reason.”
After: “There are many differences between the groups; this suggests that the intervention may affect outcomes. I do not think motivation was the only reason.”
| Before | After | Category | Why |
|---|---|---|---|
| There is a lot of differences | There are many differences | Grammar | The verb must agree with a plural, and “a lot of” is colloquial. |
| groups, this | groups; this | Punctuation | A comma alone can’t join two independent clauses. |
| may effect | may affect | Spelling | “Affect” is the verb needed here. |
| don’t | do not | Style | Contractions are too informal for most theses. |
Note what stayed the same. “May” is still there, because “affects” would claim more than the writer did. “I do not think” also stays in the first person. Changing it to “it is unlikely that” might look more formal, but it’s a different claim, so that decision belongs to the author.
A long sentence. Length alone isn’t an error, but this one has 47 words and makes the reader hold too much at once:
Before: “The results of the survey, which was distributed to teachers in three districts during the spring term and completed by most of those who received it, indicate that workload was the factor most often mentioned by respondents, although it was not the only factor that they reported.”
After: “The survey was distributed to teachers in three districts during the spring term and completed by most of those who received it. The results indicate that workload was the factor most often mentioned by respondents, although it was not the only factor they reported.”
Every piece of information survives, including “most of” and “most often”, and the qualification at the end still qualifies the same claim.
Terminology needs your judgement. If a chapter moves between “participants” and “respondents”, a proofreader might make them uniform. That’s right only if they’re the same people. If “respondents” are those who answered a survey and “participants” are those who also joined interviews, the difference is real and has to stay.
The Thesis / academic mode in humai
The proofreader has a Thesis / academic mode. It fixes spelling, punctuation and grammar, and also deals with academic style issues such as colloquial wording, overused first person, inconsistent terminology and hard-to-follow long sentences. It’s set up to correct language and style without rewriting your content. Here’s what to expect:
- Corrected text plus a list of changes. Each item shows the before and after with a reason and a category: spelling, punctuation, grammar, style or terminology.
- The diff is the complete record. The list can be partial and shows at most 60 changes per section of text, so use the highlighted changes view to see every edit.
- Over-rewriting warning. If the result differs from your original more than proofreading should, humai warns you to check that the meaning and content haven’t changed.
- Limits. Free handles 3,000 characters per request, and Pro up to 20,000 per text, so work section by section. A section at a time is also easier to review.
- Protected phrases. Add terms that must stay exact, within the limits described in our guide to keeping keywords unchanged.
Your text goes to an AI model provider to be processed, so check your university’s rules and any data agreements before you paste unpublished results or interview material. Remember too that this mode may change more than your rules allow. Under a policy like Swansea’s, you’d treat its output as suggestions at most, if you can use it at all.
What a proofreader won’t do
It won’t improve your argument, fill a gap in your evidence or check whether a source says what you claim. It can also make mistakes, so read every change and reject any that alters your meaning. Our guide on checking an AI rewrite against your facts covers numbers, names and quotes.
A proofreader isn’t the right step for large rewrites either. If a passage needs new wording, that’s the job of a paraphraser or the humanizer, and policies like those above treat it as more than proofreading, so check first.
Finally, keep your drafts, notes and supervision records. They’re a better account of how the thesis came together than an AI detector score, which is an estimate, not a verdict.
When the content is settled, start with one section in humai’s proofreader, in the Thesis / academic mode, and go through every change against your university’s rules.