Most articles about ChatGPT focus on describing the general capabilities of the language model and the basics of writing commands (prompting).

In 2025, most users are already aware of these basics. However, it is difficult to find detailed guides that discuss ChatGPT’s unique features, such as memory management in threads, real-time content search, or integrations with other tools. In this article, we will examine these more advanced options, which significantly improve the convenience and efficiency of using ChatGPT.

The article is being updated, and new features are being added to ChatGPT at a rapid pace.

Hidden ChatGPT Features.

ChatGPT debuted on November 30, 2022, and quickly became immensely popular. This means that two years have passed since its introduction. During this time, both the language model itself and the application providing access to it have undergone many modifications, and its capabilities have increased significantly. ChatGPT now has access to the internet and can also read lengthy PDFs, without the need for additional plugins.

The key to making full use of its capabilities is understanding:

How do the individual models work?
When should you use the more advanced GPT-4 model, and when is a simpler version sufficient?
How to efficiently manage memory between threads?
How to organize content (archiving, projects, searching)

Below, I will discuss the key features and practical tips for using them.

The Most Important Relatively New Features (PDF and Internet Access)

The ability to upload documents (PDFs and Word files, as well as text files) to ChatGPT significantly expands this tool’s capabilities. Many people, even those who use AI every day, are unaware of this feature and the possibilities it offers. It helps overcome many of the problems that arise when using language models.

Someone who uploads a PDF to the free version of ChatGPT can get better answers than someone who has the paid version but does not use this feature.

Many people are unaware of the possibilities offered by the function for uploading files to a language model, or even that such a function exists at all. I noticed this in my own case: I had been using ChatGPT from the very beginning, and when the option to use plugins became available, I immediately tried to test plugins that could read PDFs. However, at the time, I was unable to find a working, free plugin. I did not need it badly enough then to spend several or a dozen hours looking for a working option, although I thought about it regularly. Despite using ChatGPT practically every day, I only learned about this function about a month or two after it was introduced.

ChatGPT has gained the ability to process PDF, Word, and text files, significantly expanding the scope of its applications. With this feature, you can upload lengthy documents, books, or statutes and instantly obtain a summary, allowing you to quickly assess whether the material is valuable to you. In the case of more complex content, such as legal provisions, the tool will help you understand specific paragraphs, point out related regulations, and help you determine whether it is worth seeking legal advice.

However, what truly sets this feature apart is the ability to upload texts related to your company, organization, or situation. You can provide ChatGPT with a detailed description of your business, as well as introduce specialized knowledge, such as excerpts from books or industry reports. This way, the tool will not only better understand your context but also offer precise solutions tailored to your needs. It can generate marketing slogans, help develop strategies, or optimize processes within your company.

What’s more, uploading such data makes it possible to maintain continuity in the conversation, which is particularly useful during longer interactions. As a result, ChatGPT becomes an even more effective advisory tool that not only answers questions but also actively helps solve problems and achieve goals.

Internet access: most people are familiar with this feature because it is activated automatically for many commands. However, many people still do not use it consciously. A useful function is asking ChatGPT to search the internet, focusing on the aspects we select.

ChatGPT can display YouTube videos within a thread. It also displays other structured content, such as cinema listings.

ChatGPT Versions.

We have several versions of ChatGPT available. We can use ChatGPT for free; there are also paid versions costing $20, $25, and $200 per month.

Free version (Free)

Currently, the main limitation is the number of messages and files we can upload. Additionally, the only model we have access to is GPT-4o mini (although this model is sufficient for most tasks). We cannot create our own models, but we do have access to models created by others. We can create images with it, and we also have access to the internet (for a long time, these features were available only in the paid version). Additional features appear earlier in the paid versions.

Paid version (Plus $20 per month)

Main advantages: access to o1 models, the ability to create your own personalized models, and the ability to submit a larger number of queries.

Team ($25 and $30)

Even more inquiries, the ability to share personalized models among team members, and an administrator panel. The administrator does not have direct access to content created by the team.

Pro ($200 per month)

Even higher limits and access to o1pro mode, which provides more accurate answers. Additionally, this grants access to Sora AI, which generates videos.

Developer version. (OpenAI platform) Payment for each message.

This is a completely separate version, requiring the creation of a separate account. We access it via https://platform.openai.com, not https://chatgpt.com/.

This version is primarily intended for creating chatbots that interact with external customers who will communicate with our model, for example via Messenger (i.e., without using their ChatGPT account).

Here we have access to various parameters; for example, we can set the temperature (i.e., the level of creativity). When creating an assistant that, for instance, presents an offer to customers, we set creativity to the lowest level by default.

We can also include code with instructions here; we also have access to fine-tuning (i.e., more extensive customization of the chat’s responses to our needs). In its basic form, fine-tuning involves adding a table of questions and expected answers to the model. The process itself is fairly simple; the real challenge lies in creating an appropriate list of several dozen to several hundred questions.

Additionally, the OpenAI platform has well-prepared documentation, which is useful when we want to understand how it all works. https://platform.openai.com/docs/quickstart

https://help.openai.com/en/collections/3742473-chatgpt

ChatGPT Models – Differences and Applications

ChatGPT offers several models that differ in computing power, operating speed, and additional features. Here is an overview of their characteristics:

GPT-4o

The most popular model, with the most features, which:

Can use the Internet to search for up-to-date information.
Creates images
Creates more complex and natural texts.
Can create Canvases – that is, text that we can edit directly in ChatGPT.

GPT-4 Mini

It is a simplified version of GPT-4, but still powerful. It works faster, though sometimes at the expense of analysis detail. This model is worth using when time is important and the task does not require advanced data processing.

Can use the Internet to search for up-to-date information.
Creates images

No access to Canva.

GPT -o1

For a long time, this was the most powerful model. Text generation takes place in multiple stages; the number of stages depends on the complexity of the command.

Disadvantages: no internet access, cannot create graphics.

This model tends to create lists and write in bullet points. This is useful when creating an article outline / table of contents, whereas for creating the final text, especially when style matters, it is better to use other models.

The main applications of this model are research, text analysis, and, for example, organizing lists of concepts.

o3 mini and 03 mini high

The latest models from OpenAI. They were added to ChatGPT in response to the creation of DeepSeek’s model. In short: they are simply much more complex, effective, and faster than GPT o1.

GPT-4o with tasks (ChatGPT with tasks)

Experimental task mode. We can schedule simple tasks, such as searching for content on the Internet at a specific time or reminding us about things to do.

Most likely, this mode will be implemented in other models (similarly to what happened with the canvas model, which is now simply a feature in other models).

It does not allow us to create more complex step lists (for that, it is better to use other programs, such as Zapier, Make.com, or Relevance.ai). The main use of this mode is simply the ability to set reminders and send emails containing a link to a given thread to our email address.

OpenAI recently created an AI operator that allows users to plan a list of steps. However, similar to the developer studio, accessing the operator requires creating a new account on a new domain (and at the moment, it is unavailable in Poland).

GPT 4

Older version; it is currently not worth using.

Changing models within a single thread (be careful with anything that is not text, such as images, canvases, charts, and tables).

Currently, we can change the models we use within a single thread. This is very convenient, as we don’t need to copy text and switch between threads. However, caution is needed when generating graphics, charts, and tables, and when using the canvas feature—if we use it with the 4o model, we cannot later switch to the o1 or o1 mini model.

Key Features of ChatGPT (within a text thread)

Photo Analysis

ChatGPT is an advanced artificial intelligence model that can analyze and process natural language, responding to user questions in a contextual and logical manner. One of its useful features is the ability to analyze content contained in screenshots.

Simply upload an image of a document, spreadsheet, code snippet, or system error, and ChatGPT will analyze its contents and explain what it shows. This can be particularly useful in office work, programming, or data analysis, where complex formulas and messages require additional explanation. With this feature, you can quickly obtain explanations without having to manually transcribe the content, saving time and making your work easier.

Graphic Generation

ChatGPT can create graphics based on text descriptions. Additionally, you can paste other images into the conversation, for example, if you want to quickly extract text from an image.

Such graphics are useful wherever a quick illustration is needed. They also help when creating more complex graphics, when we need to test a color scheme or create conceptual sketches. However, in many cases, a traditional graphic designer is still irreplaceable, especially when we want to achieve a specific effect or modify a graphic.

Advantage of ChatGPT when creating graphics – we do not need to create complex commands; ChatGPT can create them for us. At the same time, since ChatGPT remembers the conversation context, we can request revisions.

Tips:

Having trouble achieving a specific style? Simply paste an image showing the effect we want into the conversation.

Precise editing: if we simply use a command such as “make it more minimalist,” “improve the background,” etc., ChatGPT generates a new image. However, if we want to preserve the main part of the image, we can click on the image, and a brush tool (mask) will appear, allowing us to select which parts of the image the new command should apply to.

Creating and Editing Tables. Can ChatGPT Create Tables?

Creating tables is another very useful, but relatively rarely used, function. It has two main practical applications and three possible forms. The first application is, of course, assistance with editing tables when we have data collected in tabular form. For example, if we have a column containing first and last names, ChatGPT can easily split it into two separate columns. It can also provide us with the appropriate formula, allowing us to do this ourselves in Excel.

Additionally, we can install a ChatGPT plugin for Excel or Google Sheets.

Another important use of tables is to support research. Creating content in table form is often more effective than writing a lengthy text describing a given topic. Therefore, a useful command is to ask for a table to be designed for a particular subject and to add new columns.

As for the possibilities offered by using tables, ChatGPT is a good analytical aid. We can therefore ask it, for example, to create a segmentation of our customers or select the main target groups (and its advice is very useful).

Table display options:

The table may be displayed:

In text form
In text form, but in a canvas (we can edit it manually later)
In the dedicated table window.

It is good to know this, because it is not always intuitive, so sometimes it is necessary to clearly specify the form it should take.

Uploading tables: at the moment, it is possible to upload files saved in CSV format to ChatGPT, but this is not convenient. It handles the .xlsx format much better. However, if we want to work comfortably with larger tables, it is necessary to use plugins.

An intermediate solution is to upload the table as a PDF or text file.

Often the simplest way to upload a table is simply to paste a screenshot.

Uploading files to the language model.

A powerful feature that not only makes using artificial intelligence easier but also significantly enhances its capabilities strongly influences how we think when using a language model.

Okay, ChatGPT can read PDFs—how can you use this?

The main use is summarizing long books and documents, allowing us to quickly assess whether a given text will be useful to us. In addition to summarizing, we can use it to “talk” to the book, while ChatGPT finds the most useful passages for us.

This is particularly useful when reading statutes. In this way, ChatGPT provides us with specific answers that we can then verify. For example, we have found a specific paragraph and want to make sure that we have understood it correctly, especially considering that other provisions may affect its meaning. In such situations, ChatGPT is an enormous help and can point us to related articles and paragraphs.

This is an enormous help in legal situations that are not serious enough to require a lawyer, as well as when we want to get a preliminary understanding of the situation before speaking with a lawyer.

In many situations, we are unsure whether a given problem requires consulting a lawyer; ChatGPT helps us assess whether such a conversation will be worthwhile for us.

ChatGPT as a specialist in any field.

Anyone who has taken even a slight interest in AI has heard of the command: behave like a specialist in a given field. This allowed ChatGPT to act as, for example, a psychologist, financial advisor, nutrition advisor, coach, mentor, and so on. A useful command was: “Ask me questions so that you can explore the subject.” In this way, ChatGPT helps solve the problems or overcome the obstacles we encounter.

However, by default, ChatGPT’s capabilities in this regard were quite limited. ChatGPT’s ability to solve everyday problems often did not result from the AI’s incredible abilities, but from the fact that someone, under the guise of experimenting with AI and discovering the future, found time to think about problems.

The ability to upload larger documents significantly enhances ChatGPT’s advisory capabilities.

We operate on two levels:

we can provide AI with expert knowledge in a specific field
We can also create a document about our situation.

Combining these two actions produces the best results. Adding PDFs containing books or lengthy articles not only provides the ChatGPT thread with additional knowledge, but above all gives it important context.

Describing our situation significantly increases the usefulness of ChatGPT’s responses. Useful responses often emerge only after a longer conversation with ChatGPT. However, limitations resulting from the context window mean that after obtaining a dozen or so useful messages, their quality often declines again, and we have to repeat the process of “fine-tuning” the model (ChatGPT “forgets” the beginning of a longer conversation). This problem can be solved by regularly asking ChatGPT to summarize the conversation, asking it to create a prompt based on many of our messages, or using the cross-thread memory function (which I describe later in the article).

However, creating a larger document into which we can put more information (content from conversations with ChatGPT, other texts, etc.) is a huge help.

Practical application:

We can upload our website, marketing materials, and other things that are not closely guarded secrets to ChatGPT. The chat can help us come up with slogans, website copy and leads, or create strategies. Additionally, we can also describe our legal situation in greater detail.

AI journaling (AI diary)

This involves using AI as a journal in which we describe our activities. Thanks to the large amount of information not necessarily directly related to a specific problem, Chat allows us to optimize our activities. The conversation feature is useful here. We can also record text and then upload it to Google AI Studio to have it transcribed.

Google AI Studio handles searching for specific texts, or characters in a text, better, while ChatGPT understands texts better.

Creating code (code window)

The model is ideal for programmers because:

Helps write, debug, and optimize code.
Supports many programming languages.
Creates comments and code documentation.

💡 Example use: Automating tasks in Python or creating simple applications. This function is useful when learning to program.

If you don't know programming, the simplest way to check whether what ChatGPT created works is to paste the code into online compilers, such as this one: https://www.programiz.com/python-programming/online-compiler/

Recently, ChatGPT introduced the ability to run a program within a thread; however, from a layperson’s perspective, it works worse than a compiler.

Charts in ChatGPT

Canvas in ChatGPT.

A common problem when using ChatGPT: after several prompts, we receive satisfactory text, but one section needs improvement. We ask for an edit, but the new version turns out to be worse than the previous one. We can always manually assemble the final text, but instead of manually copying and pasting the content, we can use the Canva feature.

How does Canva mode work?

Simply put, ChatGPT’s response in the canvas becomes editable. We can manually revise the text or select a passage that we want ChatGPT to rewrite. It’s a convenient solution for refining content.

Built-in web search engine

The new search feature allows you to quickly find information on the internet without leaving the application. This is particularly useful for:

Information verification. Writing articles requiring up-to-date data.

Voice Conversation Mode

ChatGPT has a conversation mode that allows us to communicate with the model without having to type. It is useful for more casual tasks, such as when we are walking and want to learn something. This mode works best in the app.

Additionally, we can simply dictate messages, just as in most applications supporting speech.

Main drawback: ChatGPT cannot process audio files. Therefore, for example, we cannot upload a longer recording to the model, and dictating longer messages also causes problems.

Deep Search

The most powerful ChatGPT feature. The Team and Plus plans include 10 “deep searches per month.”

Carrying out such a search takes 10–20 minutes. During this time, ChatGPT browses a huge number of websites and, most importantly, analyzes them. The main advantage of this feature is the ability to create long texts—4,000 or even 6,000 words or more—that are coherent. Sam Altman claims that the Deep Search feature produces texts at an academic level, and while the text may appear that way at first glance, it still lacks deeper insights. Nevertheless, it remains an enormous help with research. Thanks to the Deep Search feature, we can easily save a couple of hours of preliminary research.

Sora (video generator) in ChatGPT

At the beginning of March 2025, we gained access to Sora in ChatGPT. This feature had previously been available elsewhere in the world, but in March it was introduced in the EU.

Features outside the text conversation thread

Cross-Thread Memory

One of the biggest improvements in 2025 is the ability to manage memory across threads. This means you can:

Save data from previous conversations, allowing work to continue without needing to recall the context.
Configure memory, deciding what information should be stored.

How does inter-thread memory work?

Using commands such as “remember this text,” the chat adds items to its memory. We can then review what has been remembered and delete unnecessary items. ChatGPT sometimes decides on its own that something is worth remembering and informs us about it. This happens when our command includes words associated with memory or defining something (e.g., “The chair should be green”).

This memory affects all threads.

If we want convenient access to specific threads from memory, we can use the command: “remember this, call it Rothbardo3000.”

If you are managing a project, you can store threads related to individual stages of the work, making it easier to revisit them later.

💡 Usage example: We want to achieve a specific style, but simple commands do not produce a good result. So we use a mini fine-tuning process, refining the way ChatGPT should respond through a dozen or so messages. During this process, we pay attention to what it should improve, which topics it should address, and how it should select them. Once we achieve a satisfactory result, we use the command: “remember this thread and name it Misess 3000”. This allows us to immediately use the developed style in other threads. This is especially useful in more complex threads that cannot easily be summarized in a short command—one that can be conveniently pasted into new conversations.

2. Using Project Mode

If you are working on a long-term project, you can create a thread dedicated exclusively to that task. This allows you to:

Organizing notes.
Creating draft versions of texts.
Easy access to previous conversations.

This is a very useful solution when we work with a larger number of materials and do not want to upload the files again every time we start a new chat.

Drawback – we cannot conveniently set which materials we are using at a given moment. Google’s NotebookLM has such a feature for disabling and enabling files.

Search Within Threads

Another very practical feature that makes it easier to navigate threads. It is especially useful when we have many threads that we want to return to later.

It’s best to save the most important things outside ChatGPT anyway, for example in a note-taking app; you can also save links to individual threads. However, the search function is useful when we’re not sure whether something is important.

Plugins

One of ChatGPT’s main advantages over other language models is the ability to use plugins created by external companies. Some plugins are appropriately adapted ChatGPT models, while others enable connections to their services. This way, for example, with the Whimsical AI plugin, you can easily create simple mind maps.

Personalized threads (assistants, custom models, mini fine-tuning).

If we use the paid version of ChatGPT, we can create simple plugins ourselves. Through conversation, we can develop an appropriate response style, and most importantly, we can also provide the model with the knowledge it should use.

Additional Features Worth Knowing

Custom Instructions

ChatGPT allows you to define custom instructions for threads, enabling you to tailor how responses are generated. For example:

You can set your preferred writing style.
Determine the level of detail of the response.

At the moment, personalization applies to all threads; we cannot have different supercommands set for different threads.

However, this can be worked around. If we use the paid version, we can create our own assistants/mini-models, with each model having different instructions.

Keyboard Shortcuts

Keyboard shortcuts have been introduced for users of the desktop version of ChatGPT to speed up their work. However, they are fairly basic.

If we forget the Ctrl+/ shortcut, which shows us the list of shortcuts, we can always click the question mark in the bottom-right corner.

Summary

Using ChatGPT in 2025 is no longer just a matter of the quality of the generated text, but above all of how we interact with the model and provide it with the right data. The conversation context, personalization, and materials we upload ourselves—from documents to situation descriptions and specialized content—are becoming increasingly important. These are precisely the elements that determine the effectiveness of the responses and the model’s real-world usefulness.

Thanks to features such as cross-chat memory, project mode, file analysis, and integration with external tools, ChatGPT is becoming an increasingly personalized assistant. Its value lies not only in its ability to generate text, but also in its capacity to organize information, support decision-making processes, and adapt to the user. We, as users, have increasingly greater control over how the model operates and what data it uses—and this means that consciously managing this interaction is becoming crucial.

ChatGPT 2025 – A Practical Guide to Using the AI Model

Table of contents