Skip to content
Back to all articles
By TJ SoftUpdated:

Google is finally making good AI: Google Gemini, NotebookLM and Google AI Studio – Google’s artificial intelligence in practice.

Google w końcu robi dobre AI:  Google Gemini, Notebook LM i Google Ai studio – Sztuczna inteligencja od Google w praktyce.

When ChatGPT-3 was unveiled to the world in November 2022, the reactions were a mixture of fascination and… surprise. Fascination—because AI had begun speaking in human language. Surprise—because Google, a company that had dominated artificial intelligence research for years, was suddenly overtaken by a relatively young company: OpenAI. For many months, Google gave the impression that it did not know what to do. But today? Today, Google is back on the podium. And it is making quite an entrance.

In this article, I’ll tell you about three solutions from Google that do a really good job: Gemini, Notebook LM, and Google AI Studio. I’ll show you their practical applications, share my thoughts, and point out where they are better (and where they are worse) than ChatGPT and other tools.

Table of contents

When Google Was Behind.

Before we get into the specifics, a bit of context. As recently as 2016, Google dominated the world of AI: the introduction of neural machine translation in Google Translate and the launch of Google Assistant showed that the company had the know-how to create great things. But then something went wrong.

Google’s first response to ChatGPT was Bard. Unfortunately, the tool was inconsistent, less accurate, and clearly worse than the competition. Then Gemini appeared, which for many months was regarded as “the inferior chatbot.” Google—the former leader—found itself on the defensive.

But as in life, a crisis can be the beginning of a rebirth.

."""

2025: When Google Started Making Proper AI

Although Google had been signaling its AI potential since the beginning of 2024, we only began to see the real results of this work at the start of 2025. The culmination of these changes is undoubtedly Google’s advanced deep-analysis tool, Deep Research.

However, earlier, during this groundbreaking period of 2024/2025, Google had already begun presenting products that not only matched the competition but surpassed it in many respects. What was behind this transformation?

Above all, integration proved to be crucial. Combining the power of Gemini and other Google AI tools with widely used services such as Google Drive, YouTube, Maps, Gmail, and Docs created a synergy that is difficult to replicate.

Secondly, Google abandoned its relentless pursuit of the ChatGPT model and focused on innovation, forging its own paths. It was in this spirit that NotebookLM was born—a tool with a unique character, distinct from traditional chatbots.

Thirdly, the dynamic development of Google’s language models themselves cannot be overlooked. Models such as Gemini Pro, especially the latest iterations, regularly achieve top positions in industry benchmarks, demonstrating their advanced ability to understand language, generate text, code, and solve complex problems. This continuous progress in the architecture and capabilities of the models forms the foundation for all innovative Google AI applications.

And fourth, one cannot overlook the launch of Google AI Studio, a free (within its limits) development platform that opened the door to experimenting with and creating personalized AI solutions.

In this context, NotebookLM, which made its public debut in December 2023, is one of the first clear examples of this new Google strategy, involving integration and the pursuit of unique AI solutions based on increasingly powerful language models—an approach that began to yield such promising results at the turn of 2024 and 2025.

Notebook LM: the note-taking and research assistant we needed

At first glance, NotebookLM is simply a “chat for notes.” But the longer you use it, the more you notice what makes it unique.

How Does NotebookLM Work?

You can upload sources: they may include Google Docs, PDFs, website links, texts, and even YouTube videos (provided they have captions). You can add up to 50 different sources to a single project.

Then you ask questions—and converse not with a “chat,” but with your own notes. Importantly, you can control which sources the answer should be generated from. Want AI to respond based on just two documents out of fifty? One click—and it’s done.

.

A major advantage of Notebook LM is the ability to add a link to a website as a source. This means you don’t have to copy the entire content manually — simply paste the address, and the model will automatically retrieve the page’s content (as long as it is public and properly formatted). This significantly speeds up research and allows you to quickly analyze articles, blog posts, or online documentation.

An even greater convenience is the ability to add a link to a YouTube video. If the recording has captions enabled (automatic or custom), NotebookLM can retrieve the transcript and use it as a source. This allows you to analyze lectures, presentations, or tutorials without having to transcribe them manually—just paste the link, and the model will do the rest. It’s a huge time-saver when working with video materials.

Podcast Function

This is a gem. Based on your notes, NotebookLM can create… a podcast. An audio overview of the content you’ve collected yourself. It’s not a tool for creating recordings worth sharing on Spotify or YouTube, but it’s enough to go for a walk or clean the house while learning new things at the same time.

An alternative to NotebookLM in this regard could be the ElevenLabs app, which does a good job of reading text aloud. For example, Piotr Fronczewski’s voice is available there. This solution is better if we want to listen to the original text rather than a podcast.

Mind Maps in NotebookLM

Notebook LM offers a range of helpful features that streamline working with information. Mind maps allow you to quickly visualize the structure of a topic, which works great for planning projects or studying. Notebook LM creates a large map with expandable elements. Clicking on a given topic opens a text thread in which the concept is explained.

The drawback is that this map cannot be edited. It can only be exported as a PNG file.

For creating AI-powered mind maps, tools such as Mapify or Whimsical are much better. However, if we want to fully leverage the potential of mind maps, paper and pencil remain irreplaceable, while in the case of digital products, graphic design programs in which we have complete control over the layout are the best choice, e.g. Canva, Draw.io, or ClickUp.

Despite these limitations, NotebookLM’s mind map feature is genuinely helpful for research, as it is available immediately in an environment where we have more sources and notes.

.

Default text instructions

Summaries automatically extract key insights from longer texts—ideal for quickly reviewing content. Timeline helps organize events chronologically, which is useful, for example, when analyzing historical processes or the course of a project. FAQ generates questions and answers based on the source content—making it a great starting point, for example, for an article or presentation. All this information can easily be saved as notes with a single click, so there is no need to copy answers manually. These notes can then be viewed separately, without searching through the entire conversation. All of this makes Notebook LM not just an AI chat, but a true knowledge assistant.

Disadvantages of NotebookLM

You cannot save the entire conversation—only selected responses. This can be irritating, but it also forces you to actively choose what you want to keep. It can be regarded as an intentional feature, not a bug. It is also not possible to upload Word documents—although this is not really a major problem in practice.

."""

NotebookLM feature in paid ChatGPT

In ChatGPT (Plus), you can use projects, which partly resemble NotebookLM. They allow you to add files (PDF, DOCX, TXT) to a project and conduct multiple chats within it—the model knows the contents of these files and can refer to them. Unlike NotebookLM, however, you cannot choose which sources should be active for a given question. ChatGPT also does not offer automatic summarization, mind map creation, timelines, or podcasts. There is no feature for saving responses as notes—you have to do this manually. NotebookLM provides more tools for organizing knowledge, while ChatGPT is better suited to writing longer, polished texts.

Google AI Studio: you don’t have to be a developer to use it.

Google AI Studio is something like OpenAI’s Playground—only free (at least to a limited extent). What does it offer?

What is an API?

API (or Application Programming Interface) is the way different programs communicate with one another. It can be compared to a waiter in a restaurant—you (the application) place an order, the waiter (the API) passes it to the kitchen (another program), and then brings you the finished dish. Thanks to APIs, programs can exchange data and work together, even if they were created by different developers. This happens in the background, and the user simply sees that everything “magically” works together.

Using an API, you can, for example, connect accounting software to property management software. You can also connect artificial intelligence with other applications, such as customer service chatbots.

You can build your own application that communicates with users through Gemini. Or connect the language model to another tool, such as a booking system, contact form, or even your own website.

This solution is intended for more technically minded users—but also for those who enjoy learning and experimenting.

Main Goal of Google AI Studio

Google AI Studio was created primarily for testing a language model before deploying it, for example, to check whether it describes your company accurately and does not “make things up” about it, as well as whether it conducts conversations appropriately, so that users cannot use it, for instance, to write articles or joke around in other ways. The development studio also allows you to analyze how much specific responses cost—which is important when integrating it with your own application. However, because Google AI Studio is free for many applications (unlike OpenAI’s paid Playground), and is also very easy to use, it can also be used like a regular chatbot—as a lightweight, convenient alternative to ChatGPT or Gemini.

The only noticeable limitation is the length of a single conversation, although one million tokens allows you to create a really large one.

Transcriptions and audio processing

You can upload audio and video recordings, and AI will process them into text. For videos longer than an hour, you need to split them up, but 30–60-minute podcasts go through smoothly.

Quick workflow:

  • record voice notes on your phone,
  • upload them to Google Drive,
  • import them into AI Studio,
  • get a ready-made transcript that you can edit, summarize, or send to Gemini/ChatGPT for further processing.

.

Other ways to “feed your voice into AI”

Many tools, including ChatGPT, allow you to communicate with Chat in some way. However, most of them do not enable the creation of a transcript from a recording, which in my opinion is a key function when it comes to voice processing.

AI-powered solutions designed to process speech include, for example, Firefly Notes—a great application that creates transcriptions and meeting summaries, and can be easily integrated with, for example, Google Meet. If we simply want to dictate text, we can schedule a meeting just for ourselves. The Firefly Notes app is also useful in its free version.

Another solution is Voice notes, a dedicated solution for taking voice notes; however, it is paid and costs $13 per month.

Building a Google AI Studio program.

At the top of the thread, we can set instructions for a given conversation. The bar on the left side works the same way as in ChatGPT or other models, giving us access to the conversation history and additional features:

  • Stream – enables conversations with Google AI Studio; we can also share our screen with it
  • Video Gen – enables the generation of short, eight-second videos (an equivalent of Sora AI)
  • Starter apps – a collection of over a dozen simple applications, such as a trip-planning program, a flashcard creation tool, or a Windows 95 simulator.

.

The bar on the right is less intuitive; it is typically a developer tool. We can select the model used in the responses there (this is still fairly obvious to everyone). Below that, we have a counter for the Tokens used. This is useful when we want to know how much a given response costs, as it gives us a rough idea of how many questions we can ask in a given thread. Next, we have the temperature, which, in a nutshell, is the level of creativity.

Other settings are typically used when creating an API; for example, you can specify when the model should stop automatically, which can protect us from higher costs caused by user actions.

Video Generator in VEO 2

.

A fairly useful video generator; remember to save the video you create, as videos are not saved automatically.

."""

Starter apps – create your own apps in Google AI Studio

Google AI Studio enables the creation of custom applications. We can create them from scratch, while Google provides us with ready-made versions of functional applications with access to artificial intelligence, which we can modify.

.

We can choose from, for example, an app for creating flashcards, a Windows 95 simulator, or a travel-planning app.

.

Gemini: the chat that caught up with the competition (and sometimes overtakes it)

I waited a long time to be able to say this: Gemini is finally good.

And although the beginnings were weak (really weak), today we have a tool that works quickly, has access to fresh data, and is deeply integrated with Google.

Deep Research Function

Compared to the older version, the new Deep Search in Google Gemini is a major step forward. Previously, the feature was fast but rather superficial—it was more like classic search with a few added links, without any “wow” factor. Now, Deep Search creates a well-thought-out action plan (which we can modify before launching the feature), gathers information from multiple sources, and builds coherent, longer answers (even several thousand words long). Importantly, the new version filters content better, confuses facts less often, and more frequently provides specific sources. The older version was fast but shallow; the new one is still fast, but significantly deeper and more useful. As a result, Gemini deep search is now better than Grook in many cases and can sometimes compete with solutions such as ChatGPT deep search or Perplexity.

.

The OpenAI version still performs better, but Google has caught up and it is now a truly powerful tool. And for a user who doesn’t need 30 minutes of scientific research, Gemini is sufficient.

.

Another advantage is the ability to generate a short, several-minute podcast on a given topic, as well as quickly export the text to Google Docs. The short podcast is a good complement to NotebookLM, which by default creates recordings lasting several dozen minutes.

.

https://g.co/gemini/share/04428f5831f1

The Audio overview feature works not only with Deep Search; we can use it to analyze any documents, including Word files.

YouTube is the second-largest content search engine (right after Google). Many people use YouTube when looking for answers to a question or searching for a tutorial. However, YouTube’s own search engine has recently deteriorated. Google adds content to the search results that is completely unrelated to the search query and instead based on viewing history.

This problem can be solved by using Gemini to search YouTube content. This allows you to discover channels and search for specific statements by particular people (a feature similar to Grok when you want to search Twitter).

. .

Integration with YouTube, Maps, and Drive

This is where Google crushes the competition. Do you have a Google account? Do you have documents on Drive? Do you have videos with subtitles? Gemini sees all of it. You can analyze YouTube content, search through your own files, and even… use Google Maps.

At the moment, this works only with the 2.0 Flash model; the Deep Search model cannot create maps.

Gemini and Google Maps

You can ask about places based on specific criteria. Example:

  • Looking for a restaurant with a playground nearby?
  • Want to plan a weekend trip with the kids?
  • Need a café near a park?

Gemini Flash will map out a route, add pins, describe places, and even suggest opening hours. ChatGPT can’t do that—even with plugins.

Google Gemini

Custom Models

In ChatGPT, we need a paid version to create our own custom models, whereas in Gemini we have free access to this feature.

.

Google Gemini Canvas – A Canvas That Is Useful

I believe that using AI not only to create text, but also to edit text that has already been created (manually or with the help of artificial intelligence), has great potential. Therefore, as soon as this feature was added to ChatGPT, I tried it right away and … then I essentially never used it again. The Canvas in ChatGPT has a limited length (a maximum of 1,000 words), and combined with its slow performance and other limitations of this thread, it is practically pointless.

It’s different with Canvas in Google Gemini. In this case, we get a convenient text editor in which we can add headings, bold text, and create lists (or ask AI to do it for us). Such a text can be up to 2,000 words long (if we use a few prompts), although it is quite difficult to obtain longer texts. We can, however, paste more text into this working document, which is convenient when using multiple canvases within a single thread.

Most importantly, we can export such text to Google Docs while preserving all the formatting.

.

Image Generation in Google Gemini

Creating graphics in Google Gemini is at a fairly good level: images are generated relatively quickly and are of fairly good quality (realism, lack of errors). However, ChatGPT is much better in this regard. In Google Gemini, we have to describe more precisely what we want, while ChatGPT understands us better. Additionally, OpenAI’s tool allows us to change the proportions and edit graphics, which is not possible in Google Gemini.

Google AI Studio can be used to edit graphics, enabling modifications to uploaded photos.

Google Labs and experiments

It is also worth mentioning Google Labs — the place where Google tests new AI features. This is where the first Notebook LM appeared, before it became a public product.

Today, you’ll find, among other things:

  • language courses with AI,
  • content generators,
  • experimental ideas for collaboration between AI and users.

You can join the waiting list or start testing right away. A paradise for geeks and the curious.

tools currently available in Ai labs

Google vs. OpenAI: Who Is Winning?

There is no single answer. Both companies are taking different paths:

Feature | Google | OpenAI |

Speed of operation | ✅ very fast | ❌ slower (especially in browsing mode) |

Data access | ✅ YouTube, Drive, Maps | ⚠️ limited, but with unrestricted internet access |

Writing long texts | ⚠️ good, but not ideal | ✅ better |

Podcast creation | ✅ (Notebook LM) | ❌ none |

Price | ✅ AI Studio free | ❌ Playground paid |

Deep Search | ⚠️ good, but not perfect | ✅ Powerful tool |

Graphic generation | ⚠️ acceptable | ✅ High-level creation and editing |

Text quality | ✅ Good | ✅ High level |

Memory | ❌ None | ✅ Expanded memory across threads |

A comparison of the number of features cannot be made in a table because Google and OpenAI have adopted different strategies. OpenAI has practically everything in one tool (ChatGPT), which includes many features, such as projects, image generation, video generation, custom models, etc. Google, on the other hand, has many different tools (Gemini, Google AI Studio, NotebookLM, and a dozen or so smaller programs, as well as integrations into traditional programs, such as Google Workspace). Therefore, Gemini has far fewer features than ChatGPT, while Google’s AI-based tools have many more features than ChatGPT. One drawback is the lack of convenient integration between, for example, NotebookLM and Google Gemini.

Is It Worth Switching from ChatGPT to Google?

If:

  • you do quick research,
  • you work on YouTube or with videos,
  • you have a lot of files on Google Drive,
  • you are looking for integrations with Maps

Google tools may be a good alternative.

If:

  • if you care about writing style, long essays, and linguistic precision, ChatGPT is still better.

But in practice, it’s best to have both tools and use them depending on your needs.

Summary: Google Is Back in the Game

Google needed more than two years to respond to the challenge posed by ChatGPT. But today, it doesn’t just respond—it creates alternatives that make sense.

NotebookLM is a phenomenal tool for research and learning. Gemini finally works smoothly and has access to things that other models don’t. And Google AI Studio gives us the power to create our own solutions—for free.

This is not sponsored content. It was written by someone who, just like you, is looking for the best tool for their work. And after a long time, I can say: Google is really making good AI now.