How to Edit and Improve Your Writing with ChatGPT and Other Large Language Models
Originally published January 31, 2023
Substantially revised and updated August 11, 2026
Maurício Pinheiro
Editorial note: This article was originally published only a few weeks after ChatGPT became publicly available. At that time, using a conversational artificial intelligence to revise a text was still a novelty. Much has changed since then. ChatGPT is no longer the only practical option, and large language models have become significantly more capable at proofreading, editing, restructuring, criticizing, and rewriting text. This revised edition preserves the original purpose of the article while updating its methods, examples, limitations, and recommendations.
I. Introduction
ChatGPT text editing is one of the simplest and most useful applications of large language models.
A few years ago, automated writing tools were mostly limited to spelling, grammar, and simple stylistic suggestions. Modern large language models can work at several different levels. They can correct punctuation, improve sentence structure, eliminate repetition, clarify arguments, adjust tone, identify ambiguities, reorganize paragraphs, compare versions, and even explain why a particular passage is difficult to understand.
ChatGPT is perhaps the best-known example, but it is no longer the only possibility.
Today there are several large language models and AI assistants capable of performing similar tasks. Depending on the platform, language, document, and type of revision required, a user may work with ChatGPT or with alternatives based on other major language-model families such as Claude, Gemini, and others.
The specific model matters less than the underlying principle:
the quality of the revision depends not only on the artificial intelligence, but also on the instructions you give it.
A vague request such as:
“Improve this text.”
may produce a useful result, but it also gives the model considerable freedom to change your style, vocabulary, argument, and sometimes even your intended meaning.
A better approach is to specify what should be changed, what should be preserved, and how much editorial freedom the model is allowed to use.
II. What Can a Large Language Model Actually Do with a Text?
Text correction is not a single task.
It is useful to distinguish several levels of intervention.
Proofreading
At the most conservative level, an AI can check:
- spelling;
- grammar;
- punctuation;
- verb agreement;
- pronouns;
- articles;
- prepositions;
- typographical mistakes.
The goal is to correct errors without substantially changing the author’s language.
Copyediting
At the next level, the model can improve:
- awkward sentences;
- unnecessary repetition;
- excessive wordiness;
- unclear constructions;
- transitions;
- consistency of terminology;
- readability.
Here, the AI does more than correct mistakes. It improves the way the text communicates.
Stylistic Editing
A language model can also adjust the writing style.
For example, it can make a text:
- more formal;
- more concise;
- more academic;
- more conversational;
- more technical;
- more accessible to a general audience;
- more persuasive;
- less repetitive.
This is powerful, but it also requires greater caution because stylistic editing can gradually erase the author’s individual voice.
Structural Editing
For longer texts, AI can examine organization rather than individual sentences.
It can identify:
- paragraphs that appear in the wrong order;
- repeated arguments;
- missing transitions;
- weak introductions;
- conclusions that merely repeat earlier material;
- sections that interrupt the logical flow.
This is particularly useful for essays, reports, articles, papers, and long-form writing.
Critical Review
Perhaps the most interesting development is that large language models can be asked not to rewrite a text at all, but to criticize it first.
They can search for:
- contradictions;
- hidden assumptions;
- ambiguous statements;
- unsupported conclusions;
- logical gaps;
- unnecessary digressions;
- claims requiring references;
- passages that may confuse a reader.
In many situations, this is more valuable than simply asking the AI to rewrite everything.
III. The Most Important Rule: Tell the AI What Not to Change
Consider the following instruction:
Revise this text.
The request is technically valid, but it leaves several important questions unanswered.
Should the AI change only grammatical errors?
Can it rewrite entire sentences?
Can it shorten the text?
Can it change the author’s vocabulary?
Can it modify the argument?
Can it correct factual statements?
Should it preserve the author’s style?
The model has to infer the answer.
A much better instruction would be:
Copyedit the following text. Correct grammar, spelling, punctuation, repetition, and awkward syntax, but preserve my meaning, vocabulary, and writing style. Do not rewrite sentences that are already clear. After the revised text, briefly list the most important changes you made.
This creates a much clearer editorial contract between the author and the AI.
For more conservative editing, try:
Edit this text conservatively. Correct clear linguistic problems, but do not substantially rewrite it. If a sentence may require a change that affects its meaning, flag the sentence and explain the problem instead of changing it automatically.
For a stronger intervention:
Rewrite this passage for clarity and concision. Preserve all substantive ideas, but remove redundancy, simplify unnecessarily complicated sentences, and improve the logical progression between paragraphs.
The difference between these prompts is not cosmetic. They describe three fundamentally different editing tasks.
IV. Ten Better Prompts for Revising Text
The following prompts illustrate different levels of AI-assisted revision.
1. Basic proofreading
Proofread the following text. Correct spelling, grammar, punctuation, and typographical errors. Do not change the author’s style or sentence structure unless necessary.
2. Conservative copyediting
Copyedit the following text for clarity, grammar, punctuation, and readability. Preserve my vocabulary, meaning, tone, and writing style. Avoid unnecessary rewriting.
3. Remove repetition
Identify unnecessary repetitions and redundant arguments in the following text. Suggest a more concise version without removing important information.
4. Improve clarity
Identify sentences that are unnecessarily difficult to understand. Rewrite only those sentences and explain briefly why each change improves clarity.
5. Improve paragraph structure
Analyze the organization of the following paragraphs. Identify weak transitions, repeated ideas, or paragraphs that may be in the wrong order. Suggest a better structure before rewriting anything.
6. Preserve authorial voice
Improve this text while preserving my authorial voice. Avoid replacing distinctive vocabulary or stylistic choices merely because a more conventional formulation exists.
This instruction is especially important for essays, literary writing, opinion pieces, and personal reflections.
7. Academic revision
Review the following text as an academic editor. Improve clarity, precision, terminology, and logical flow. Identify claims that appear unsupported or require references, but do not invent citations.
8. Check argumentation
Do not rewrite the text yet. First identify logical gaps, contradictions, hidden assumptions, unsupported claims, and places where a critical reader could challenge the argument.
This is often one of the most powerful uses of a language model.
9. Compare two versions
Compare the original text and the revised version. Identify every substantive change in meaning, tone, emphasis, or factual content. Ignore purely grammatical changes.
This helps detect situations in which an apparently harmless revision has changed the author’s argument.
10. Final editorial pass
Perform a final editorial review of this text. Look for grammar problems, repetition, unclear passages, inconsistent terminology, weak transitions, factual claims requiring verification, and unnecessary verbosity. Do not modify the text yet. Give me a prioritized list of problems first.
Notice an important pattern in several of these prompts:
analysis comes before rewriting.
That is often preferable to asking an AI to immediately produce a new version.
V. Grammar Correction Is Not Fact Correction
This distinction is essential.
Consider the sentence:
“Paris is the capital of England.”
There is nothing grammatically wrong with this sentence.
The problem is factual.
Changing it to:
“London is the capital of England.”
is therefore not grammatical correction. It is factual correction.
Large language models can perform both tasks, but they should not be confused.
A useful instruction is:
Separate linguistic problems from factual claims. Correct grammar and style independently. List factual claims that may require verification rather than silently changing them.
This matters because a model may occasionally replace a correct statement with an incorrect one, especially when dealing with specialized, ambiguous, obscure, or recent information.
For important factual material, verification remains necessary.
VI. Ask the AI to Explain Its Changes
One of the advantages of working with a conversational language model is that revision does not have to be a black box.
Instead of simply accepting a rewritten paragraph, ask:
Why did you change this sentence?
or:
Show me which changes were grammatical and which were stylistic.
or:
Which sentences did you substantially rewrite, and why?
You can also ask:
Give me the original sentence, the revised sentence, and a short explanation of the change.
This transforms the AI from an automatic rewriting machine into something closer to an interactive editor.
It is also extremely useful for learning.
A student can discover why a particular construction is awkward. A researcher can inspect whether technical meaning has been preserved. A non-native speaker can understand recurrent grammatical problems rather than merely receiving corrected text.
VII. Beware of Overediting
Large language models are often surprisingly good at making prose smoother.
That creates another problem: they can make everything sound the same.
Distinctive expressions disappear.
Unusual sentence structures become conventional.
Personal rhythm becomes standardized.
Strong stylistic choices may be interpreted as errors.
After several rounds of unrestricted rewriting, a text may become grammatically polished but increasingly anonymous.
For this reason, instructions such as:
Preserve my voice.
and
Do not rewrite sentences that are already clear.
are more important than they may initially appear.
The objective should not always be to produce the most polished sentence statistically available to the model.
Sometimes the slightly unusual sentence is precisely what makes the writing yours.
VIII. ChatGPT Is a Tool, Not the Tool
When this article was first written in 2023, ChatGPT dominated public discussion about conversational artificial intelligence.
The ecosystem is now much broader.
There are multiple large language models and AI assistants capable of reviewing, rewriting, summarizing, translating, and analyzing text. Their strengths differ. One may preserve style better, another may handle very long documents more effectively, another may be particularly useful for integration with a particular software environment.
For ordinary text revision, therefore, there is no reason to think in terms of a single mandatory platform.
The prompts in this article are intentionally model-independent.
You can often use essentially the same instructions with different large language models and compare the results.
In fact, for important texts, comparison can be useful.
Ask two different models to identify problems without rewriting the text. If both independently identify the same ambiguity, repetition, or logical weakness, that deserves attention.
The purpose is not to determine which artificial intelligence is universally “best.”
The better question is:
Which model, instructions, and workflow produce the best result for this particular text?
IX. What Large Language Models Still Should Not Decide for You
AI can make writing faster.
It can make editing more systematic.
It can notice problems that the author has stopped seeing after reading the same paragraph twenty times.
But the final editorial decisions still belong to the author.
A language model does not fully know:
- what you intended to imply;
- which unusual phrase was deliberate;
- which technical nuance must remain untouched;
- what your audience already knows;
- which uncertainty you intentionally preserved;
- which stylistic imperfection is part of your voice.
And fluent language should never be mistaken for guaranteed truth.
A beautifully written sentence can still be wrong.
A confidently presented correction can still distort the original meaning.
A perfectly grammatical paragraph can still contain a weak argument.
Human judgment therefore remains essential, especially in academic, scientific, legal, technical, journalistic, and professional writing.
X. A Better Workflow
For serious writing, instead of using a single command such as “improve this text,” consider a sequence.
First: ask the AI to identify problems without rewriting anything.
Second: decide which problems you agree with.
Third: request corrections while defining how much freedom the model has.
Fourth: compare the revised version with the original.
Fifth: verify factual changes separately.
Finally: read the entire text yourself.
This workflow may seem slower than pressing a button and accepting an automatically rewritten version.
In practice, it is often much better.
The objective of AI-assisted writing should not be to remove the author from the process.
It should be to give the author a more powerful editor.
XI. Conclusion
When this article was originally published in early 2023, asking an artificial intelligence to correct a paragraph still felt remarkable.
Today, simple grammatical correction is almost the least interesting thing a large language model can do with a text.
ChatGPT and other language models can act as proofreaders, copyeditors, stylistic reviewers, structural critics, argumentative adversaries, and interactive writing assistants.
But their usefulness depends heavily on how they are used.
The best results usually come not from saying:
“Rewrite this.”
but from specifying:
what is wrong, what may be changed, what must be preserved, and what the AI should question rather than silently correct.
The technology has changed considerably since 2023.
One principle has not:
Artificial intelligence can improve a text, but the author must remain responsible for what the text ultimately says.
Editorial transparency note: This article, as with all articles published on this site, was conceived, directed, written, and reviewed by Prof. Maurício Veloso Brant Pinheiro. Artificial intelligence was used as an assistant for editorial refinement, formatting, image generation, SEO metadata, and publication workflow.
Copyright 2026 AI-Talks.org
