Learning with AI isn't asking for summaries: how to sort sources and actually understand

While I was preparing my final thesis, the one that got me my degree in information security, I had to study foundational topics from many different sources.

It wasn’t a single, neat, closed bibliography. There were books, PDFs, YouTube videos, materials on cybersecurity frameworks, and also references tied to software development. The problem wasn’t getting information. The problem was sorting that information, understanding which ideas kept repeating, which concepts were truly central, and how to connect them so I could write with judgment.

That’s where an important difference shows up: using AI to study shouldn’t mean asking it to “summarize this for me.” That can work as a first approach, but it can also create a dangerous illusion: feeling like you understood something because you read a shorter version of it.

Studying isn’t about consuming less text. It’s about being able to explain better.

A summary isn’t enough

A summary tells you what a source says. But when you’re seriously learning a topic, you need something more:

• Detecting which concepts hold up the entire topic
• Understanding which ideas show up repeated across multiple sources
• Separating what’s important from what’s secondary
• Finding contradictions or nuances
• Identifying what you still can’t explain in your own words

That difference shows up a lot in technical topics. In cybersecurity, for example, you can read about frameworks, controls, risk management, secure development, or maturity models. But if you just pile up definitions, you end up with a list of concepts, not real understanding.

And a list of concepts isn’t enough to write a thesis, defend an idea, or make decisions.

The problem wasn’t just understanding, it was also sorting

In my case, there was another fairly concrete problem: frameworks change.

I’d often find a useful guide, PDF, or reference, but then a newer version of the same framework would show up. Some sources cited older versions. Others mixed concepts from different editions. And if I wasn’t paying attention, I could end up using duplicate, outdated, or less official bibliography than the original source.

That’s not a minor detail in a final thesis.

When you work with cybersecurity or development frameworks, you need to know:

• What the current version is
• What the official source is
• Which documents are duplicates
• What changed between one version and another
• Which reference is worth citing
• What material is useful for understanding versus what serves as primary bibliography

AI shouldn’t replace that validation. But it can help you sort out the map.

You can load several sources and ask it to detect whether they’re talking about the same framework, whether they cite different versions, or whether some references look derived from an official source. After that, the final decision is still yours: verify the current version, go to the original document, and choose what bibliography stays.

The difference is that you’re no longer reviewing a pile of documents blindly. You’re working with a first layer of organization.

AI as a tutor, not a shortcut

The most useful way I found to think about tools like NotebookLM is this: not as a summary generator, but as a tutor that works on top of your sources.

That changes the kind of questions you ask.

Instead of asking:

Summarize these documents.

It’s better to ask:

Identify the five central concepts that appear across these sources and explain how they connect.

Or also:

What would I really need to understand to be able to explain this topic to someone from scratch?

The difference is huge. The first prompt compresses information. The second forces you to build structure.

When you work with books, PDFs, and videos, that structure is worth more than a summary. Because the value isn’t in having a shorter version of each source, but in detecting the common map across all of them.

The questions that expose whether you understood

One of the best ways to study with AI is to ask hard questions.

Not memory questions. Questions that connect concepts.

For example:

• In what case would this framework not be enough?
• What mistake would someone make applying this control without understanding the context?
• What secure development concepts connect with this cybersecurity practice?
• What part of my explanation is superficial?
• What source should I review to improve this point?

At that point, AI stops being an answer machine and becomes a tool for detecting gaps.

That’s the key point. Learning isn’t confirming you’re right. Learning is finding where your explanation still breaks down.

A more useful study session

If I had to lay out this method as a practical session, I’d do it like this:

1. Load several sources on the topic.
2. Ask for the central concepts, not a summary.
3. Ask what you’d need to understand to be able to teach it.
4. Generate questions that require reasoning.
5. Answer without looking at the sources.
6. Ask for hard but useful correction.
7. Close with a review plan.

This is especially useful when the goal isn’t “knowing what a topic is about,” but being able to use it: write, explain, decide, or defend a position.

For a final thesis, that difference matters. Because collecting information isn’t enough. You have to turn it into your own line of argument.

The risk of studying on autopilot

AI can make studying more active or more passive. It depends on how you use it.

If you only ask for summaries, you’ll probably study faster, but not necessarily better. You can end up outsourcing exactly the most important part: connecting ideas.

But if you use it to ask better questions, compare sources, challenge your explanations, and detect errors, the outcome changes.

The tool doesn’t replace the effort of understanding. It can make it more visible.

Closing

For me, the best way to study with AI isn’t asking it to simplify everything. It’s asking it to force you to think better.

NotebookLM, or any similar tool, can be useful if you use it as an active learning system: sources in, better questions, your own answers, correction, and review.

The question isn’t whether AI can summarize a topic.

The question is whether it can help you realize you still don’t understand it as well as you thought.

And there, used with intention, it starts to have a lot more value.