Reading time: 6 min.
Recently, I have discovered and tested NotebookLM, a new free platform from Google that can generate automatic summaries and highlight keywords or relevant information from entire documents. Using AI, the tool analyses the sources we input to return analyses, summaries, and even audio content according to our specific requests. The innovation lies in its local approach: unlike many AI tools, NotebookLM only processes answers from the documents we have uploaded, thus being very precise and a valuable study aid.
As a translator, however, I wondered what use I could make of this tool in my daily work. I imagined I had to document a new topic for a social campaign: cycle tourism in Italy.
About the Dashboard
The tool allows us to upload up to 50 sources in various formats, including PDF, .txt, markdown, audio files (mp3), external links (such as websites), Google Drive documents (Word or PowerPoint), YouTube links, and even text copied directly into the interface. My search is based on 17 sources of various types.
Once the sources are uploaded, NotebookLM creates a customized archive based on an analysis of the information entered, generating a dedicated window (in my case, ‘Cycle tourism in Italy’).
The dashboard includes sections for sources, the chat where we can enter our queries, and the notes section, which offers several automatic functions. After entering sources, you can easily query the tool via chat and easily retrieve information, such as technical terms, numerical or chronological data, and much more. The answer always shows the reference source.
In addition to the possibility of adding new sources, the tool also allows us to discover related sources via keywords, directly from Google.
Let us now explore the individual functions.
Overview Audio
Overview Audio is certainly the most innovative function of the platform, as it allows audio summaries to be created from the sources entered. In addition to educational use, many online resources cite this function for creating podcasts with AI: the output looks like a conversation between two hosts! If we do not provide precise indications, the tool analyses all the available sources for the creation of the audio, otherwise we can narrow the analysis by asking only to dwell on some of them; with 17 sources it takes just under 10 minutes to obtain an audio of about 7 minutes (in Italian, as the sources provided). To understand the potential of this function, I decided to make a first generic prompt and a second more targeted one.
Prompt 1: Create a Podcast on cycle tourism in Italy
I got a summary of about 7 minutes and 26 seconds. The playback is customizable (in terms of speed) and, as anticipated, the voices available are two: a male one, warm but slightly more artificial, and a female one, bright and surprisingly natural. Neither has obvious regional accents, although I did notice some inaccuracies in the pronunciation of proper names, for example, those of some cities. The language used is fluid and colloquial, with a brisk pace, well-calibrated intonations (questions, exclamations), and frequent use of interludes typical of the spoken language, such as ‘eh’, ‘here’, ‘exactly’, ‘great’. Foreign expressions such as ‘outdoor friendly’ are pronounced as a native Italian speaker would, contributing to the naturalness effect. It may sound repetitive at times, but this is related to the quantity and variety of sources available. Overall, the effect is similar to that of an episode of a podcast: pleasant to listen to, informative, and well-constructed. However, it is important to remember that these are still faithful summaries of the sources provided: the tool specifies this, stating that it does not make subjective judgments.
Prompt 2: Create a podcast on cycle tourism; address experienced and passionate people; use an engaging and promotional tone; highlight routes in the Alto Garda Area.
The result was an audio of 6 minutes and 20 seconds. Compared to the first experiment, the female voice changed: warmer and more pleasant to listen to, she also pronounced the place names correctly, as opposed to the male voice, which made a mistake. The content responded well to my request: the focus on Alto Garda was respected, with a good level of detail, technical references, and language more suited to those already familiar with the subject. The intonation, especially in the female voice, conveys enthusiasm and contributes to a more engaging listening experience. In terms of structure, however, the audio ends somewhat abruptly, with a final intonation that gives the impression of a sudden interruption rather than a natural closure. Again, the conversational component (interjections, colloquial tone) is very natural. Overall, the result is quite close to promotional audio content, although it is evident that it is a summary of sources.
The overall final effect is well constructed but lacks that communicative intention that makes a podcast designed to promote or engage really effective. Nevertheless, it remains a promising tool, especially considering how quickly it is possible to generate content that is listenable, informative, and already tailored to a specific audience.
Studio & Notes
Another interesting function of NotebookLM is the Studio section, designed for those who need to organize and explore information in a structured manner. The notes generated can become part of the sources themselves, and the tool automatically reprocesses them to create useful study or teaching materials.
The Study Guide, for example, offers short or open-ended quizzes, as well as lists of key terms with clear and concise definitions - the perfect tool for teaching or studying. It is also possible to generate a timeline, which lines up events and personalities mentioned in sources, or a FAQ section automatically derived from uploaded content.
The Summary function takes the form of a real document, complete with a title, biography of the sources, and paragraphs, similar to a term paper. In one of the tests, however, a date appeared that did not correspond to any element of the notes, pointing to a certain limitation in text processing.
Despite some uncertainties, the Studio function shows interesting potential for the organization of complex materials and the preparation of well-structured educational or professional content.
Final thoughts
NotebookLM proves that artificial intelligence is not a shortcut, but can become a concrete partner in everyday work: a valuable support to organize, search, and manage sources more efficiently. This tool, in particular, is not (at all) a tool to be used for writing or translating texts, but can be extremely useful in the preliminary work phase. For a translator, it means being able to collect disparate materials - briefings, articles, technical documentation, client notes - and turn them into concept maps, summaries, glossaries, or timelines that can be quickly consulted during the project. While on the one hand, we need a critical eye and human control (as always), on the other hand, this type of tool offers us the possibility to lighten some repetitive phases and gain time to do better what we already do very well: translate with care, creativity, and competence.
