Best AI Proofreader: Engineering, CS, Physics, Chemistry, Math

The best AI proofreading tool for engineering, computer science, chemistry, physics and math papers edits your sentences and lets you reject any change to the notation. ProofreaderPro.ai fits that job. It shows every edit as a tracked change, keeps IEEE citations and notation in place and lets you pick a Light, Standard or Comprehensive editing level. You keep control of the technical content, and the tool handles the English around it.

The volume of technical research keeps growing. IEEE Xplore hosts over 6 million documents and adds 20,000 new ones every month. NeurIPS received 21,575 submissions in 2025, AAAI about 29,000 in 2026 and CVPR 13,008 in 2025. Submission counts at top conferences grew 128% to 345% in five years.

Computer science is also the only major field where conferences, and not journals, are the main publication venue. A conference paper gets one review round. If ICML rejects it, you can't revise it for the same reviewers, and the next conference is often six months away. The language has to be right at the first submission.

Most of these papers are written in a second language. China now produces 69% of submissions to AAAI, and computer science is India's top research field at 21% of its output. Over 70% of engineering submissions worldwide come from non-native English speakers. Chemistry, physics and math papers add their own layer of notation, units and formulas that a general grammar tool treats as errors.

Best AI proofreading tool for engineering, CS, chemistry, physics and math

ProofreaderPro.ai is an online AI proofreader for academic writing, with features that suit technical manuscripts. It keeps IEEE citations in their numbered square brackets, leaves mathematical notation and code snippets alone and handles the dense terminology of engineering, CS and the physical sciences. It has three editing levels, so you can choose how much of each sentence it touches.

General grammar checkers can flag LaTeX commands as errors, suggest shortening "convolutional neural network" to "a type of neural network" or break a numbered IEEE citation. ProofreaderPro.ai is built for technical writing. It treats "O(n log n)" as a complexity expression and "[1]-[3]" as a citation range.

Each of the tools technical authors use has a different job, and the right pick depends on where you write and what you need checked. The descriptions come from each vendor's own pages:

Writefull and Trinka are the closest alternatives for this audience. Writefull is the simpler choice if you write in Overleaf all day, and Trinka's personal dictionary stops the same terms being flagged again. Paperpal's readiness checks suit a final pass before submission. Grammarly suits general writing in email, documents and the browser.

ProofreaderPro.ai is built around the academic manuscript. Its Light level makes the smallest changes, which suits a paper where every other word is a term of art. The Standard and Comprehensive levels rewrite more, so read their tracked changes closely. For head-to-head detail, see our comparisons with Writefull, Trinka and Paperpal.

Why engineering and CS papers get rejected for language quality

Reviewers in engineering read under time pressure. A typical CVPR reviewer handles 5 to 8 papers in 2 to 3 weeks. When the first paragraph mixes tenses, the abstract uses undefined acronyms and nominalizations hide the contribution, the paper takes more effort to read. Reviewers then engage less with the technical content and tend to score the paper lower.

Elsevier reports that 30% to 50% of submissions are desk-rejected, with "poor English and grammar" listed as a top reason. IEEE editorial guidelines say manuscripts with "severe language deficiencies" will be returned to authors before review. ACM journals note in their author guidelines that "papers must be written in clear, grammatical English" and that "poorly written papers may be rejected regardless of technical merit."

A rejection rarely mentions your English. Reviewers write that the paper is hard to follow, the contribution is unclear or the methodology section is confusing. The cause is often the language, even when the content is sound. Our guide on how to respond to a desk rejection covers what to do next if it happens.

Why a general grammar tool gets chemistry, physics and math wrong

A general grammar tool is trained on everyday prose, so it flags anything that doesn't look like everyday prose. In a technical paper that includes most of the notation. The tool sees "pH" and suggests "PH", sees "x" next to "y" and suggests a word, or sees "5 mL" and tries to join the number and the unit.

Overleaf's own documentation says Grammarly doesn't fully understand LaTeX markup, which sometimes causes wrong suggestions and words that aren't spell-checked. The same is true of most general tools, because they're built for emails and essays. A methods section with a buffer recipe is outside what they expect.

The errors fall into five groups:

Common English language errors in engineering and CS manuscripts

Engineering writing has its own error patterns, different from medical or social science writing. These are the ones reviewers see most often.

"Which" versus "that". This is the most common grammar error in engineering papers. "The algorithm which achieves the best performance" should be "The algorithm that achieves the best performance", because it's a restrictive clause with no comma. "The ResNet architecture, which was introduced in 2015, serves as our backbone" is non-restrictive and needs the comma. Unedited engineering drafts misuse "which" on almost every page.

Nominalizations that hide the action. Engineering authors often turn verbs into nouns, as in "The implementation of the algorithm was performed" instead of "We implemented the algorithm." Another example is "The optimization of the loss function was conducted using SGD" instead of "We optimized the loss function using SGD." This adds words without adding information. It can make a methods section 30% to 50% longer than it needs to be, and it hides who did what.

Articles with technical nouns. "We train model on ImageNet" is missing an article. "We train the model on ImageNet" refers to a specific model, and "We train a model on ImageNet" introduces one for the first time. For non-native speakers, articles with technical nouns are the most persistent error. Chinese and Japanese researchers, who write the largest share of CS papers, come from languages with no article system.

Tense in experimental sections. Use the past tense for what you did ("We trained the model for 100 epochs"). Use the present tense for what is generally true ("Batch normalization reduces internal covariate shift") and for your paper's claims ("Our method outperforms the baseline"). Mixing them makes it unclear what's established and what's new.

Dangling modifiers with the passive voice. In "Using a learning rate of 0.001, the model was trained for 200 epochs," the model didn't use the learning rate, the researchers did. "Compared to the baseline, our method achieves 3.2% higher accuracy" is correct. "Compared to the baseline, the accuracy is 3.2% higher" dangles, because the methods were compared and not the accuracy.

Undefined or inconsistent acronyms. CS papers are full of acronyms such as CNN, RNN, LSTM, GAN, LLM, ViT, MLP, SGD, BERT and GPT. Each needs a definition at first use. Authors often define an acronym in Section 3 but use it undefined in the abstract, or switch between "Transformer" and "transformer" or "self-attention" and "Self-Attention".

Run-on sentences. "We propose a novel framework that leverages attention mechanisms to capture long-range dependencies in sequential data and combine them with graph neural networks to model structural relationships between entities while maintaining computational efficiency through a sparse attention pattern that reduces the quadratic complexity to linear." That's one 52-word sentence, and it should be three.

Units and SI style: what to check in every sentence

Units are the first thing to check in a chemistry or physics paper, because the rules are short and every draft breaks them somewhere. NIST publishes a checklist of SI style rules that follow the BIPM SI Brochure, and it's the quickest reference for a paper that follows SI.

These rules come up most often in a draft:

• Put a space between the number and the unit symbol, so "75 cm" and not "75cm". The same space applies when the unit is used as an adjective, as in "a 25 kg sphere".
• Unit symbols never take a plural "s", so it's "75 cm" and not "75 cms".
• Don't add a period after a unit symbol unless it ends the sentence.
• Variables and quantity symbols are italic, and unit symbols are upright. The mass m and the metre m look the same in plain text, and the typeface tells the reader which one you mean.
• Don't attach information to a unit. NIST gives the example Vmax = 1000 V, and not "1000 Vmax".
• Don't use "ppm", "ppb" or "ppt" for a quantity value. NIST suggests forms such as 2.0 µL/L instead.
• Don't mix a unit symbol and a unit name, so write "kg/m³" or "kilogram per cubic meter", and not "kilogram/m³".

A grammar tool can break a few of these. It may add a plural, a period or a hyphen, because "25-kg" looks right in everyday English. It rarely catches the opposite problem, such as a missing space or a "ppm" a journal wants rewritten. Those are reading jobs for you, or for a human editor with a style guide open.

The BIPM SI Brochure is the primary source for the system itself, and the ninth edition is the current one. Use it when a reviewer questions a unit or a prefix. Most chemistry and physics journals add their own house rules, so check the author guidelines before you decide a rule applies.

Chemical names, formulas and IUPAC nomenclature

Chemical names follow IUPAC rules, and no grammar tool should change one without a check. IUPAC publishes organic nomenclature in the Blue Book and inorganic nomenclature in the Red Book. A name such as 2-methylpropan-1-ol is built from locants, prefixes and a parent name, and each piece has a fixed place.

Three things go wrong most often in a draft.

Capitalization. Chemical names are lower case in running text, except at the start of a sentence. A grammar tool tends to capitalize them after a hyphen or a number. If your journal follows the ACS Style Guide, check the current edition when a case or a hyphen is in doubt. The guide covers compound names, numbers and units.

Formulas and charges. Subscripts and superscripts carry the meaning. CO₂ and CO are different compounds, and Fe²⁺ and Fe³⁺ are different ions. When you paste text from a Word file into any tool, check that subscripts and superscripts survived. Copying from a PDF often turns them into normal characters on the baseline, so "H2SO4" and "Fe3+" appear as plain text.

Italic prefixes. Stereochemical and positional prefixes such as cis, trans, N- and tert- are italic in most journal styles. A tool that rewrites the sentence may drop the italics without telling you. A tracked change in Word shows a formatting change as well as a word change, so you can catch it before you accept.

Our biology and life sciences post covers gene and species names, which follow different italic rules from chemical names. If your paper spans chemistry and biology, read both.

Variables, subscripts and italics in physics

Physics papers rely on the typeface to carry meaning, which is the main reason an editor should work on the formatted file and not on plain text. IUPAC and NIST give the same rule for quantity symbols: they're italic, and unit symbols and labels are upright. NIST adds a rule for subscripts. A subscript that is itself a variable or a running index is italic, as in cp for heat capacity at constant pressure p. A subscript that is a descriptive label is upright, as in mp for the mass of a proton.

Authors often miss that second rule. Both forms look similar on a page, and a typesetter fixes them quietly, so many drafts reach review with inconsistent subscripts. If you paste your paper into a proofreader, check that the output kept the italic and upright forms. A plain-text editing box removes them, and you then have to restore them by hand.

Other notation points come up in physics:

• Mathematical operators and functions such as sin, log and d (in an integral) are upright. The AMS sets multi-letter abbreviations in regular type and single-letter variables in italic.
• Vectors and matrices follow a journal convention, bold or an arrow. Pick one and use it all the way through.
• An equation is part of a sentence, so it takes the punctuation the sentence needs. A displayed equation that ends a sentence ends with a period, and one that continues the sentence ends with a comma or nothing.
• Define every symbol the first time it appears, and don't use one letter for two quantities.

AIP and APS journals and the AMS journals each publish an author guide with their own details. Read the one for your target journal before the last proofreading pass, because it settles the open questions about notation for that submission.

Numbers, significant figures and uncertainty

A number in a science paper has three parts: the digits, the uncertainty and the unit. A proofreader that changes any of them is changing a result. Two rules from the NIST checklist matter here.

The first is about uncertainty. Write "(123 ± 2) g" and not "123 ± 2 g", because the second form leaves it unclear whether the unit applies to both numbers. A range gets the same treatment, so "1 MHz to 10 MHz" and not "1 MHz-10 MHz". The second is about long numbers. NIST says to group long numbers in threes with thin spaces and not commas, as in 15 739.012 53. That applies when there are more than four digits on either side of the decimal marker. Journals differ on this one, and many accept commas, so match the journal.

Significant figures are a content decision, so they stay with you. If your instrument gives three significant figures, the paper reports three. An AI editor that tidies 0.250 into 0.25 has dropped a significant figure. When you accept tracked changes, check every digit in the results section against your data table.

Two more habits help. Spell out a number that opens a sentence, since most styles ask for that. Keep the form consistent, so a quantity is "5 mL" every time or "five milliliters" every time, and never both on the same page.

Math papers: the prose around the proofs

A math paper is mostly prose between formulas, and that prose has conventions a general tool treats as mistakes. Authors write "Let G be a finite group. Then G has a subgroup of index 2 if and only if..." and expect the editor to leave the structure alone. A grammar checker may flag "Let" as a fragment or ask to rewrite "if and only if" in plain English.

An editor should check these points in a math paper:

1. Consistent use of "we" for the author's steps and the imperative for the reader's ("Consider the map...").
2. Tense, since proofs run in the present tense and results from the literature use the past or the present perfect.
3. Definitions placed before first use, and no symbol reused for something else.
4. Punctuation around displayed equations, the most common error in math drafts.
5. Terms of art used the same way in the abstract, the statements and the proofs.

The AMS publishes a style guide for authors and sets variables in italic and operator names in regular type, the pattern most math journals follow. A math paper's source is usually LaTeX, so the next section matters most here.

LaTeX source or PDF: what to send to the proofreader

Send the prose, and keep the markup and the equations out of the proofreader. A .tex file mixes sentences with commands such as \cite, \ref and $...$, and a grammar tool can treat those as words. A PDF has the right appearance, but copying text out of it loses italics and subscripts and breaks lines at the page width.

Our post on proofreading a LaTeX paper in Overleaf has the full workflow for splitting the source, editing the prose and putting the changes back. In short, paste whole paragraphs of prose, include inline math only when you need the context and compile again after each batch.

Writefull works inside Overleaf and says its feedback is tailored to LaTeX, so it's the simplest choice when the source stays in the editor. For a team that reviews in Word, a file with tracked changes is easier to share. Both routes work, and the choice depends on where your co-authors read.

Engineering and CS terminology our AI proofreader preserves

General grammar checkers often flag code snippets, mathematical notation and domain terms as errors. ProofreaderPro.ai preserves:

• Mathematical notation: O(n²), ∀x ∈ X, argmin_θ L(θ), ||x||₂
• Code and pseudocode: function names, variable names, API references
• ML and AI terms: backpropagation, softmax, cross-entropy loss, batch normalization, dropout, learning rate decay, gradient clipping
• Hardware specs: NVIDIA A100, TPU v4, 256GB RAM, 8×H100
• Dataset names: ImageNet, COCO, CIFAR-10, SQuAD, GLUE, SuperGLUE
• Metrics: mAP, F1-score, BLEU, ROUGE-L, perplexity, FID, IS
• IEEE citation format: [1], [2]-[5], [1, Theorem 3]
• Conference names: NeurIPS, ICML, CVPR, ICCV, AAAI, ACL, EMNLP

How to proofread an engineering or CS paper with AI

Match the editing level to the stage of the paper, and review every tracked change.

Step 1: Comprehensive editing on your first complete draft. This pass catches structural problems. These include nominalizations, a passive voice that hides who did what, run-on sentences, tense shifts and article errors. Do it 1 to 2 weeks before a conference deadline.

Step 2: Standard editing after co-author feedback. Your collaborators asked you to restructure Section 4, and you rewrote the experimental setup. The new text needs a grammar pass that leaves the sections you already cleaned up alone.

Step 3: A Light proofread 24 hours before submission. This final pass catches typos, inconsistent figure references ("Fig. 3" and "Figure 3") and formatting slips from last-minute edits.

Here's a comprehensive edit on a CS results section.

Original: "The proposed method achieves a top-1 accuracy of 78.3% on the ImageNet validation set which is 2.1% higher compared to the baseline ResNet-50 model and the inference time was measured to be 4.2ms per image on a single NVIDIA A100 GPU which represents a 15% reduction compared to the previous state-of-the-art approach."

After AI proofreading: "The proposed method achieves a top-1 accuracy of 78.3% on the ImageNet validation set, 2.1% higher than the baseline ResNet-50. Inference time is 4.2 ms per image on a single NVIDIA A100 GPU, representing a 15% reduction compared to the previous best method."

The edit split one 54-word run-on into two sentences and turned the "which" clause into a shorter phrase. It also tightened "compared to", removed the extra "model" and "approach" and simplified the passive "was measured to be". Every number stayed the same.

Separate passes for notation, numbers and references

In a chemistry, physics or math paper, run the editor on the prose first and check the notation afterwards. Mixing the two makes it easy to accept a change to a unit while you're reading for grammar.

1. Use the Light level on the abstract and introduction. These sections are mostly prose, and a light pass fixes articles, agreement and tense without touching terms.
2. Use the Standard level on the discussion, where long sentences and weak transitions collect. Read each tracked change.
3. Check notation only. Search the document for each unit, symbol and formula, and compare it with your last good version in the Word comparison view or the tracked changes list.
4. Check the numbers. Open the results table next to the text and check each digit, uncertainty and unit.
5. Check the references. The AI Citation Checker checks each reference in a pasted list against Crossref and Semantic Scholar. It finds references that don't exist, and it doesn't compare your in-text citations with the list.

The Academic plan costs US$19 a month or US$169 a year and covers 25,000 words a month at 500 words per pass. Academic Plus costs US$29 a month or US$259 a year and covers 50,000 words at 1,000 words per pass, with tracked-changes export. A 6,000-word paper is twelve passes on Academic and six on Academic Plus.

Hedged claims in physics, chemistry and math

Hedging is a place where an editor can do harm, so read those changes slowly. A hedge such as "suggests" or "is consistent with" tells the reader how strong your evidence is. A tool that turns it into "shows" or "proves" has made a stronger claim than your data supports.

The reverse also happens. An author who writes "it may possibly be the case that the value might be slightly higher" has stacked four hedges on one claim. A good edit leaves one: "the value may be higher". Use this table as a guide:

In math, a proved statement takes no hedge at all. A proved lemma is stated plainly, and a conjecture is called a conjecture. A tool that adds "it appears that" to a theorem has changed its status.

How to paraphrase related work in CS without plagiarism

A related work section has its own paraphrasing problem. You need to describe other methods accurately and in words different enough from the source. Technical terms can't change: "convolutional neural network" stays "convolutional neural network", and "gradient descent" can't become "slope reduction". The math is fixed too, so only the framing language can change.

Our academic paraphrasing tool restructures the sentence and keeps every technical term, method name, dataset name and numerical result.

Source: "Zhang et al. (2023) proposed a multi-scale feature pyramid network that extracts features at four different resolutions and fuses them using learned attention weights, achieving a mAP of 45.2 on COCO val2017."

Paraphrased: "A multi-scale feature pyramid network with learned attention-based fusion across four resolution levels was introduced by Zhang et al. (2023), reporting 45.2 mAP on the COCO val2017 benchmark."

The technical terms, the numbers and the citation are the same, and the sentence structure is different.

How to humanize AI-assisted drafts for engineering papers

Many CS researchers use ChatGPT or Claude to help draft related work sections, describe standard methods or structure an introduction. AI-drafted engineering text has recognizable patterns. Paragraphs have the same length and open with a topic sentence and three supporting sentences, and "Moreover", "Furthermore" and "It is worth noting that" come up often.

Reviewers notice these patterns, and some conferences, including NeurIPS and ICLR, are discussing policies on AI-generated content in submissions.

Our AI text humanizer for academic papers changes these patterns and keeps the technical content as written. It varies sentence length, removes formulaic transitions and gives the text the rhythm of experienced technical writing.

AI-generated: "Deep learning has achieved remarkable success in computer vision tasks. Moreover, recent advances in transformer architectures have further improved performance on various benchmarks. Furthermore, the integration of self-supervised learning has reduced the dependence on labeled data. It is worth noting that these developments have significant implications for real-world applications."

After humanization: "Deep learning has done well on computer vision tasks, and transformer architectures have pushed benchmark results further. Self-supervised learning has also cut the need for labeled data. Together, these changes matter for systems used outside the lab."

The humanized version makes the same claims with fewer stock transitions and more varied sentences. To make it sound like a researcher in the field, add the specifics yourself, such as the architecture you mean, the paper you cite and the benchmark numbers.

Conference culture: why deadline pressure shapes proofreading

CS runs on conference deadlines. CVPR, ICML, NeurIPS and AAAI each have one main submission deadline a year, and some now have two. Miss it by a day and you wait 6 to 12 months for the next one, so the last week before submission is under heavy time pressure.

Researchers often write and revise until hours before the deadline, and the camera-ready version after acceptance also has a fixed deadline. A human editor usually needs 3 to 5 days to return a manuscript, which doesn't fit that schedule. An AI proofreader returns its edits in seconds, so you can still edit on deadline day.

The growth numbers show the demand:

• NeurIPS submissions grew 128% in 5 years, from 9,467 in 2020 to 21,575 in 2025.
• AAAI grew 194% in 2 years, from 14,823 in 2024 to about 29,000 in 2026.
• ICLR grew 345% in 5 years, from 2,594 in 2020 to 11,530 in 2025.

Each of those submissions needed publication-ready English by a fixed date.

Top engineering and CS venues where language quality matters

Conferences (acceptance rates):

• NeurIPS 2025: 24.5% (21,575 submissions)
• CVPR 2025: 22% (13,008 submissions)
• ICML 2024: 27.5% (9,473 submissions)
• AAAI 2026: 17.6% (about 29,000 submissions)
• ICLR 2025: 32% (11,530 submissions)
• ACL 2024: 24% (NLP)
• EMNLP, ICCV, ECCV, SIGKDD, WWW

Journals:

• IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI), IF 20.8
• IEEE Transactions on Neural Networks and Learning Systems, IF 14.3
• Nature Electronics, IF 33.7
• Nature Machine Intelligence, IF 18.8
• ACM Computing Surveys, IF 16.6
• Proceedings of the IEEE, IF 20.6

All of them expect clear, grammatical English, and all of them desk-reject papers with serious language problems.

Which style guide applies to your paper

The journal's own author guidelines decide every open question, and a general rule gives way to them. These are the main references:

• The BIPM SI Brochure and the NIST Guide to the SI for units and quantity style.
• The IUPAC Blue Book and Red Book for the names of organic and inorganic compounds.
• The ACS Style Guide for papers in ACS journals.
• The AIP and APS author guidelines for physics journals.
• The AMS author guidelines for mathematics and atmospheric science journals.
• The IEEE and ACM author guidelines for engineering and computing venues.

Before you submit, run through our journal submission checklist and the points above. If the journal wants British or American spelling, the free US and UK English converter handles the spelling in prose. Check units afterwards, because "metre" and "meter" are each correct in their own spelling system. For medical papers, our medical research proofreading guide covers the conventions that differ.

Frequently asked questions

Q: What is the best AI proofreading tool for engineering and computer science papers?

ProofreaderPro.ai is a good fit, because it keeps IEEE citations, notation and code in place and shows every edit as a tracked change. Its three editing levels let you choose a deep edit on a first draft and a light pass before the deadline.

Q: Can the AI proofreading tool handle mathematical notation and code?

Yes. ProofreaderPro.ai preserves mathematical expressions such as O(n log n), argmin and norm notation, along with code snippets, function names and LaTeX-style formatting. It doesn't flag these as errors or suggest simplifying them, and it edits the English prose around your technical content.

Q: Is using an AI proofreading tool allowed for conference submissions?

Generally, yes. AI-assisted copy editing that fixes grammar and readability is widely accepted, and it's different from using AI to generate research content. NeurIPS, ICML and CVPR policies focus on AI-generated text. Proofreading your own writing with an AI tool is similar to using a grammar checker or hiring a copy editor, but check your venue's current policy.

Q: Can the paraphrasing tool handle related work sections without changing technical terms?

Yes. The academic paraphrasing tool restructures sentences while it keeps method names, dataset names, numerical results and citations. "ResNet-50 achieves 76.1% top-1 accuracy on ImageNet" stays exact, and only the sentence structure around it changes.

Q: How fast does it work for a conference deadline?

The proofreader returns tracked changes within seconds of pasting a section. Reviewing the edits for a full paper usually takes 10 to 15 minutes, so you don't have to wait days for a human editor.

Q: What is the best AI proofreading tool for chemistry papers?

ProofreaderPro.ai is a good fit, because it shows every edit as a tracked change and you can reject any edit to a formula, charge or compound name. Use the Light level for papers full of nomenclature. Check subscripts, superscripts and italic prefixes after each pass, because formatting can change when text is copied between files.

Q: Can an AI proofreader handle LaTeX equations?

No proofreader should edit equations, so the safe workflow is to proofread the prose and leave the math alone. Writefull and Trinka say they support LaTeX, so they're worth trying if you work in the source. With other tools, paste paragraphs without the math and compile after each batch. Our Overleaf post has the full steps.

Q: Does Grammarly work for physics and math papers?

Grammarly works for the prose, but Overleaf says it doesn't fully understand LaTeX markup and may give wrong suggestions. In Word it avoids that problem, though it can still flag unit symbols, variables and formulas as errors. Check each of those flags against your style guide.

Q: How do I check units and SI style in a manuscript?

Use the NIST SI style checklist, which follows the BIPM SI Brochure. Check the space between number and unit, no plural on unit symbols, italic quantity symbols, upright unit symbols and no "ppm" for a quantity value. A grammar tool can't check all of these, so do it as a separate pass.

Q: Should variables be italic in a paper?

Yes, quantity symbols and variables are italic, and unit symbols are upright. IUPAC and NIST give the same rule. A subscript is italic when it's a variable or an index and upright when it's a descriptive label, such as the "p" in the mass of a proton.

Q: Do chemical names need capital letters?

Chemical names are lower case in running text, except at the start of a sentence. Italic prefixes such as cis and tert stay italic. Check the IUPAC nomenclature books for the name itself, and the ACS Style Guide or your journal's guide for how it's set.

Q: Can I use AI to proofread a math paper?

Yes, for the prose between formulas. Ask for a light pass, and reject any change to a definition, a theorem statement or a hedge. Math prose has set phrases such as "if and only if" and "let", and a general tool may try to rewrite them, so a co-author should check any edited proof.

Q: How do I proofread numbers and significant figures?

Compare every digit in the edited text with your data table, and check the uncertainty and the unit together. NIST says to write "(123 ± 2) g", with the unit applying to both numbers. Reject any edit that rounds a value or drops a trailing zero.

Q: Does ProofreaderPro.ai keep my formatting and tracked changes?

ProofreaderPro.ai takes Word uploads and exports to Word on the Academic plan, and Academic Plus exports the edits as tracked changes. Check subscripts, superscripts and italics in the exported file before you accept the changes. The free plan is 250 words a month of editing, without uploads or Word export.

Try the AI Proofreader for Technical Papers

Proofread engineering, CS, chemistry, physics and math papers with IEEE citations, notation and technical terms kept in place, and every edit shown as a tracked change.