Using AI when writing texts is controversial—some people fear that it will replace human creativity. In this guide, I will show you how to use artificial intelligence to create texts responsibly, without sacrificing quality. AI tools such as ChatGPT can be extremely helpful for proofreading or improving structure, but they should not replace our own thinking.

The basic principle: when creating a text, we must understand the subject we are writing about. More precisely, we must strive to understand the topic.

The second principle: how much time can be saved. With artificial intelligence, you can create in 4 hours a text that would take us, for example, 12 hours to write “by hand.” We can therefore spend, for example, two additional hours on more thorough research, and still be 6 hours ahead.

But we shouldn’t reduce the time spent writing such a long text to, for example, 10 minutes.

Update: texts written by artificial intelligence, 2025.

Changes in the field of artificial intelligence are happening very quickly. In principle, most of this text is still up to date, as it describes the general principles of creating texts with the help of AI, but a few things have changed since then.

AI simply works better; it can create longer texts.

This is quite a significant change. Two years ago, writing a text of more than 1,000 words was quite an achievement for language models. Their context windows have now increased, so longer texts—even those of 3,000–5,000 words—are within their reach. Of course, this refers to text that makes sense, since lengthening a text by “padding it out” and repeating the same thing has always been possible. Such text still requires refinement, and it is a good idea to write some passages manually and generate others separately, even for posts created in a hurry, but this is now easier than it was two years ago.

Claude AI Deserves Recognition.

Claude AI, although it is one of the leading language models, is much less well known than ChatGPT or Google’s Gemini. However, this model regularly surprises me, and despite having paid versions of ChatGPT and Gemini Pro, I sometimes use the free version of Claude. ai.

In short, texts written by Claude sound more human and also contain fewer artifacts typical of AI-generated content.

Deep Search is the foundation.

2025 is the year of deep research tools. This feature is offered by ChatGPT, Gemini, Perplexity, and Grok. In this mode, the model searches hundreds of websites, then analyzes them and writes the actual text. With this approach, generating a response takes several to a dozen minutes, but the result is impressive. This can also be used to create longer texts.

Quick prompt to get started.

Write a 2,500-word article about [insert topic here]. Use H2 and H3 headings; each heading may have a different number of paragraphs, and each paragraph may vary in length. Aim to write fluidly, but use lists when necessary; most of the text should be written in a flowing style. First think through the structure, then write the article.

Language models have a problem creating a good text structure. They like to use patterns, and many such texts have a formulaic layout, such as a heading and a paragraph, each of the same length. Simply asking for H2 and H3 headings, apart from making the work easier, forces the use of a more complex structure, but models still “like” patterns, and often every H2 heading has the same number of subheadings. This is quite a problem because, with good texts, the structure adapts to the topic: some issues can be described quickly, while others require elaboration.

ChatGPT as an Aid in Proofreading

To begin with: you won’t create useful text with a simple prompt. However, AI can significantly speed up the work – text that would take us 8–16 hours to write without using AI can be created in 2 hours. This benefit is particularly evident where AI helps us overcome creative blocks, making the writing itself more enjoyable.

The simplest use of ChatGPT when writing an article is proofreading. In this case, you can start working right away, without having to delve into how AI works. Simply write the text and ask ChatGPT to make corrections. Experimenting and testing which prompt works best is extremely easy here. Sometimes it is enough to ask ChatGPT to correct only the errors without changing the style. Other times, it is worth focusing on improving specific passages.

After such a correction, once the text is technically and grammatically correct, manual edits should be added to make it sound more human.

Of course, proofreading alone is of limited help, especially for people who already write well.

Another advantage of ChatGPT emerges here – language models are quite good at understanding text containing so many errors that it is virtually illegible. Deciphering such notes is time-consuming and tiring for a human, but for mo If we have a specific thought but lack the right words, we can simply write it down as a rough draft or “stream of consciousness” and then ask ChatGPT to refine it. This opens up a new way of thinking: we often do not feel like writing this way because we know that making corrections later will be time-consuming. In this case, AI does it for us. This allows us to put on paper thoughts that, generally speaking, because of writer’s block, we would not otherwise be able to write. In other words, in such cases, artificial intelligence supports our “natural” thinking instead of replacing it.

The faster we write, the greater the chance of inaccuracies, which may cause AI to misunderstand us. Therefore, the farther our text is from its final form, the more crucial prompt testing becomes.

Not just ChatGPT—it’s worth checking out other models, such as Claude, Google Gemini, or DeepSeek (which we can safely run using Perplexity).

Dictating text to a language model.

Dictating text to ChatGPT is just as “fine” and safe as using a language model merely to proofread text, while in many cases it is much more convenient.

In many cases, writer’s block can be overcome simply by talking about the topic at hand. Thanks to artificial intelligence, we can save time; when we encounter writer’s block, we don’t have to alternate between speaking and writing—we can simply record what we want to say and then process it.

Although many models have a voice mode, unfortunately they do not support longer recordings by default. Therefore, this solution requires minimal preparation: you need to set up a convenient process for recording audio and then uploading it to language models that support voice-to-text transcription. In the case of the Polish language, this is somewhat tricky, because default programs, such as those from Google, have problems with it. For example, Google Meet can generate captions in Polish and can transcribe a recording in English, but not in Polish.

However, Google AI Studio handles such transcriptions without any problem, and Microsoft Word has no problem with it either. There are also paid options, such as Sonic.ai.

Such a transcript then needs to be processed; manual corrections are often necessary, for example when creating the text skeleton, but this is much easier than starting from scratch.

Changing the Length of Text Fragments

You started writing a text and noticed that a side thread takes up as much as 70% of its length? ChatGPT can easily shorten it while preserving its meaning and ensuring that it supports the main idea of the text instead of replacing it. If you run a blog, the longer passage in its unabridged version can serve as the basis for a new article.

It also happens that a given passage is too short and does not look good in the context of the whole. In such a case, language models can help expand it. However, this is somewhat more difficult if we want to ensure that the result is coherent and natural.

Changing the Text Style

If you notice that you have written something in a style that doesn’t sound quite right, ChatGPT can improve it and adjust the tone—for example, making the text sound more casual or more formal.

If we have trouble adapting the text using commands such as “write more formally” or “make it more official,” it works well to provide the example of a specific person known for their strong writing skills. However, copyright and any associated risks must of course be taken into account; the safest approach is to refer to people whose texts are already in the public domain.

Research Assistance

The basis of most articles is solid research. Even if we have our own experience or knowledge gained through regular learning, it is always worth verifying certain information when writing a text. This is where AI comes in handy—it allows us to explore a given topic more quickly, especially when we have difficulty understanding certain issues.

However, it should be remembered that AI can hallucinate—generate false information. Therefore, it is crucial to verify the most important data, although such errors do not occur often. If we have basic knowledge of a given subject, it will be easier for us to spot inaccuracies.

Much also depends on the type of questions we ask. If we suggest in the query that something happened or exists, the language model may fabricate false content.

In this case, AI works similarly to an internet search engine (which have long relied on certain forms of artificial intelligence).

Tips:

Perplexity – an AI tool based on various language models, specifically designed for research.
Useful prompts in ChatGPT: “Create a list of terms and organize it”, “create a table of the most important events.”

Deep Search Tools

Over the past few weeks, various Deep Search–type tools have been made available. They are a great help when writing texts based on research. Grok (available on X (Twitter)) and Google Gemini assist with rapid research, although Grok is somewhat more accurate and conducts its research a little longer. Perplexity’s Deep Search feature “understands” complex topics quite well, although in the free version we have access to only three queries per day.

ChatGPT’s Deep Search performs a powerful analysis and is capable of creating long, coherent texts. It understands context quite well, although Perplexity sometimes delivers better results. With Plus and Team subscriptions, we have access to 10 queries PER MONTH.

You can see the capabilities of the deep search function here. Texts written by artificial intelligence are nothing new, but the quality of text created by deep research functions deserves special attention.

Where doesn’t deep search work?

The Deep Search function does not handle searching for specific events in the current year very well, for example. This is because most of the articles that rank highly in the search results concern earlier years. Unfortunately, ChatGPT’s Deep Search is no exception here. Therefore, in the text below, I had to do everything manually. https://blog.zgloszenia24.pl/majowka-2025-w-polsce-przewodnik-po-dlugim-weekendzie-gitarowy-rekord-swiata/

Feeding AI with Specific Knowledge

Using prompts, you can improve the style of a text relatively easily. It is more difficult, however, to convey specific knowledge in a particular way.

A great help is uploading longer texts to the language model, such as PDF files containing books, Wikipedia articles, or other sources. ChatGPT probably contains most of the information from Wikipedia, but providing it with articles allows you to add context and a “way of thinking.”

Thanks to this:

The risk of generating false information is lower.
We can immediately verify the content in a trusted source.

The best results come from a mix of your own texts and texts containing data. It is very convenient to create folders such as “AI dump,” where you simply put texts that may be useful in the future. The materials collected this way can then be added directly to specific threads, and if you intend to use them more often, it is convenient to add them to Google’s NotebookLM or to a project in ChatGPT.

Additionally, we can also feed artificial intelligence texts created using artificial intelligence—for example, if we consider a particular answer to be very good and have previously verified its accuracy.

AI writing specific sections of text.

Another way to use AI when writing texts is to create so-called “more boring passages” or “fillers.”

It is not about artificially extending the content itself. Often, when writing, we have passages that come naturally to us and are enjoyable, but at the same time we see that it is necessary to add:

Explanations of more difficult concepts.
Clarifications that are not essential to the main thesis.

For example: we have used a specialized term and want to explain it, but it is not the main topic of the article. If, for some reason, we have difficulty writing such a passage, ChatGPT can prove invaluable. In this case, it does not limit our creativity—we know exactly what should be included in the passage, while AI simply relieves us of the monotonous work.

The same applies to abstracts and summaries. Sometimes their inclusion is required, and AI can quickly generate a first draft. For more complex texts, summaries usually require refinement, but they are still useful.

Artificial intelligence writing an article based on a table of contents or a list of bullet points

At this stage, it is already worth knowing the basics of how commands (prompts) work in ChatGPT and being able to test them.

If we have the entire article plan in mind, we can provide AI with a table of contents or a list of key points. Based on these, ChatGPT will generate the full text. This way:

We avoid the blank-page problem – AI helps quickly create a first draft of the article.
We can focus on editing and adapting the text instead of starting from scratch.
AI may suggest additional expansions that complement our material.

It is important to formulate instructions precisely—the more accurately we define the structure, style, and expectations for the content, the better the result we will achieve.

ChatGPT creating a table of contents for us

In this mode of operation, we allow artificial intelligence to write the entire text. However, this does not mean that we lose control over the outcome. There is a huge difference between text generated based on a short prompt, e.g., “Write an article about cars,” and content that we have precisely tailored to our needs using more detailed instructions.

Workflow with ChatGPT

Provide a list of topics related to the given subject.
Generate a table of contents for the selected topic.
Polish individual sections, focusing on the key sections.

After generating the table of contents, it is often worth adjusting it manually to better tailor the article’s structure to our needs.

Creating a complete article

After refining the table of contents, we ask ChatGPT to write the entire article. The first version is rarely immediately usable—revisions are usually necessary. In practice, most often:

We are introducing several changes throughout the entire text.
We ask AI to regenerate specific fragments.
We combine the best versions into a cohesive whole.

Example base command

“Write a 2,000-word blog article. Use H2 and H3 headings, write in a casual style, provide real-life examples, and solve specific problems.”

Adapting the Writing Style

One popular strategy is to add the instruction: “Act as a specialist in [topic].” If the AI still does not produce a satisfactory style, it is worth referring to specific people, e.g.:

“Write like Mariusz Zaruski, adding elements of the styles of Sun Tzu and Ayn Rand.”

When writing about living people, care should be taken not to create content resembling plagiarism. For example, if we are writing about finance, it is better not to model the content on contemporary financial bloggers, but, for instance, on experts from other fields. Combining the styles of several people often produces the best results.

“Feeding” AI with Your Own Materials

As with research, it can be very helpful to provide the model with preselected content—both your own texts and articles from trusted sources.

When it comes to originality, it is useful to think of AI in terms similar to collages, remixes, or reaction videos on YouTube. Our contribution is crucial—in selecting materials, building the structure, and adapting the content to our needs.

Creating Text Based on a Conversation with ChatGPT

The foundation of working with language models is making revisions, analyzing the model’s responses, and adapting the next query accordingly. In other words, it is much easier to write a short prompt and then specify what needs to be improved (a few times, if necessary) than to write one long, polished prompt that produces a satisfactory result right away.

When we ask a language model to write a text, the entire process takes quite a long time. We can speed it up by talking to the language model as if we were trying to learn something. Once we have gathered the necessary knowledge in the thread, we can ask the language model to write the text based on the conversation.

General Information About Advice on Writing Texts

Who are writing tips for?

Most advice on writing articles is aimed at bloggers and journalists—people who earn money from their writing, mainly through advertisements and paid collaborations.

However, in recent years, content marketing has been playing an increasingly important role—blogs used to support sales. Their purpose is not only to attract customers, but also:

Building a positive company image – professional content demonstrates competence and experience.
Positioning yourself as an industry expert – valuable articles strengthen authority.
Creating a knowledge base for customers – the blog helps readers better understand the topic and find answers to frequently asked questions, saving time on individual consultations.

Writing to Order vs. Writing for Our Own Needs: About What We Know

Writing commissioned texts requires greater skill than creating content about topics we know from our own experience.

Many people have unnecessary blocks related to writing, causing them to give up on many opportunities. Meanwhile, a blog can become a powerful tool—it helps attract customers, build expert status, and at the same time motivates further learning and development.

How Is AI Changing the Approach to Writing?

Artificial intelligence enables people who previously did not have time to write or could not afford to hire a copywriter to create valuable content independently.

Thanks to AI, people with knowledge, experience, and unique insights—such as business owners and craftspeople—can effectively build their reputation as experts and grow their brand without the need for major investments in content creation.

Useful Commands in ChatGPT

General Information About Commands in Language Models

Right after ChatGPT’s debut, many guides and even courses were created in which the main selling point was a list of useful prompts. However, this approach makes little sense, because one of the main advantages of language models is precisely that we do not have to remember specific phrases.

Much more important is the ability to converse with the chat and adapt to its limitations (and then work around them). Sometimes language models have trouble understanding a fairly simple instruction; often, using a particular word or phrase causes ChatGPT to go in a completely wrong direction. Guessing which word is causing the issue is a key skill. On the other hand, using the right word often makes the task easier.

Useful commands come in handy when we want to speed up content creation, whereas we should be able to figure everything out through experimentation.

A Few Useful Commands

Correct this text, fix errors and grammar
Preserve more of the text I sent you.
Write the text half as long
Write this text in two paragraphs.
This text is intended to be the introduction to the article.
Analyze the ideas in this text, and then write a better text while preserving the key ideas.
Write the text, then analyze it and improve it.

Step List – Powerful Meta Prompt

A powerful meta-prompt that makes creating other prompts much easier: a list of steps. I had intuitively used this many times, but only recently realized that it is a useful tool (it’s a thing).

Before the introduction of o1, it was possible to obtain a simple chain of reasoning this way, using the command: “write this text, analyze it, and then write it again.” This is a solution that often works where other things do not.

I rediscovered this prompt while testing deep search and comparing it with a regular language model that had access to the internet: a command like “create a research plan, then execute it, analyze it, and execute it once more” produces good results.

Sometimes the operation of such a list of steps is immediately apparent; in that case, a change in behavior can be observed, while at other times it simply affects the quality of the response.

Quasi-philosophical reflections on the use of AI in writing texts

AI as a modern version of dictating text/artificial intelligence as an assistant.

Before language models became widespread, many texts were created through dictation—one person spoke while another recorded the content. However, the writer’s role often extended beyond faithfully transcribing the speaker’s words. Sometimes, the text was created based on a casual conversation, with the writer giving it its final form.

In some cases, the specific person responsible for the recording was crucial to the final result; in others, they could easily be replaced by someone else.

Currently, using AI to write texts can be compared to such a process—with the difference that instead of engaging another person to do the tedious work, we entrust it to a language model.

Mechanical Texts – Do They Make Sense?

In most cases, generating texts with AI without any form of oversight defeats the purpose. If someone can generate a similar article themselves, why would they read it?

There are, however, situations in which such an approach can be justified. If the text is created in a short time (e.g. half an hour) with the support of AI, and then reviewed, corrected, and properly formatted—with graphics, links, or a well-chosen structure added—adding such content to some websites may be justified. Sometimes, the difference between having a ready-made text and having to generate one is significant.

At present, tools that automatically create articles from start to finish are not widely used, and those available on the market often do not work well enough.

Where Does the Value of a Text Lie?

Sometimes the idea for the text itself is what matters most – it is where all the creative work lies. Of course, a manually written article would be better, but if the idea is original, even text generated by AI based on a few prompts can hold its own.

Another approach is automated data processing and the creation of valuable content from large sets of information. For example, if we have the rights to books or articles, we can transform them into shorter, condensed materials.

In such a case, the creative value lies not in the writing itself, but in designing a system that intelligently processes and organizes information. This approach not only saves time, but also enables the creation of content that would be impossible or not cost-effective to prepare manually.

Summary

Artificial intelligence is becoming a powerful tool supporting the writing process—from proofreading and editing, through research, to generating complete texts based on an outline. However, it does not replace human creativity and analysis.

AI brings the greatest value where it helps organize knowledge and streamlines the creative process. Mechanical text generation without supervision is often worthless, but thoughtful use of AI makes it possible to save time and increase efficiency.

Thanks to artificial intelligence, people who previously did not have time to write can create valuable content, build their image, and use blogs in content marketing. However, it is worth remembering that the human as editor, content curator, and strategist plays a key role, giving texts their final form and unique value.

How to Create Texts with Artificial Intelligence (ChatGPT Texts)

Table of contents