Why AI Summaries Help You Skim but Self-Explanation Helps You Understand

When a piece of writing gets hard, most people do the same thing.

They stop reading and ask for a summary.

That makes sense. A summary is fast. It gives you the shape of the article, the headline ideas, and the feeling that you are caught up again. If your goal is orientation, that is often enough.

But if your goal is understanding, summaries are often solving the wrong problem.

The hardest moments in reading are usually not document-level problems. They are local problems.

one sentence that compresses too much
one term you almost recognize but cannot explain
one jump in the author's reasoning
one paragraph that depends on three ideas you do not fully have yet

A summary can tell you what the article is about. It often cannot get you through the exact point where your understanding broke.

That distinction matters more than it seems.

The problem is not always "too much text"

A lot of AI reading tools assume that difficulty comes from volume.

If the article is long, shorten it. If the paper is dense, simplify it. If the chapter is technical, extract the key points.

Sometimes that works. If you only need the gist, compression is useful.

But a lot of hard reading is not hard because it is long. It is hard because meaning is unevenly distributed.

You may understand 85 percent of an article and still be blocked by the remaining 15 percent. And that 15 percent is often not optional. It is the definition, the mechanism, the assumption, or the transition that makes the rest of the piece hang together.

This is why people can read a clean AI summary, nod along, and then return to the original text only to discover that they still cannot really read it.

The summary removed friction. It did not necessarily build understanding.

Why summaries feel more helpful than they are

One reason summaries are so appealing is that they create fluency.

Everything sounds smoother when someone else has already compressed the material, cleaned up the structure, and removed the parts that slowed you down. That smoothness can feel like learning.

But the science of learning has warned about this for a long time. In Make It Stick, the authors argue that many familiar study habits, including rereading and other low-friction review behaviors, often create the illusion of mastery rather than durable learning. Things that feel easy can be misleading. Things that feel effortful often produce stronger learning.

That does not mean summaries are bad. It means they are easy to over-credit.

A summary can make material feel familiar without making it usable. It can make an argument feel clear without making it retrievable. It can make a topic feel finished when you have only seen a compressed version of it.

If you have ever thought "I understood that summary perfectly" and then struggled to explain the original passage in your own words, you have felt this gap directly.

What learning science says instead

If summaries are good for orientation, what is better for actual learning?

Three ideas from cognitive science are especially useful here.

1. Self-explanation improves understanding

Research by Michelene Chi and colleagues found that when learners explain material to themselves, they often build a deeper understanding than when they simply reread it. In one influential study, students who were prompted to self-explain while reading an expository science text learned more than students who reread the same material. The gains were especially visible on harder questions that required inference and deeper understanding, not just surface recall.

That matters because real understanding is not just recognizing words on a page. It is integrating new information into what you already know. When you try to explain a sentence, you are forced to notice what you do and do not actually understand. You have to connect the new idea to prior knowledge, fill in missing links, and resolve ambiguity.

In other words, explanation is not just a way to display understanding. It is one of the ways understanding gets built.

This is especially relevant for difficult reading. If one paragraph in a paper loses you, the most useful next step is often not "summarize the whole thing." It is "explain this exact part in simpler language, and make the hidden assumptions visible."

2. More constructive engagement usually leads to more learning

The ICAP framework offers another useful lens. It proposes that learning tends to improve as engagement moves from passive to active to constructive to interactive.

A lot of summary consumption is passive or, at best, lightly active. You receive a cleaner version of the material, but you may not generate anything new from it.

Self-explanation is different. It is constructive. You are not just receiving information. You are producing an explanation, a bridge, an inference, or a reorganization of the idea in your own mental model.

That distinction helps explain why two reading experiences can feel similar in the moment but produce different results later. One leaves you with a pleasant sense of familiarity. The other leaves you able to use, connect, and recall the idea.

3. Retrieval matters if you want the learning to last

Even if you understand something while reading, that does not mean you will remember it tomorrow.

This is where retrieval practice becomes important.

Work by Jeffrey Karpicke and others has shown that actively retrieving knowledge leads to stronger long-term learning than simply restudying it. In one well-known study, retrieval practice outperformed concept mapping for learning science texts, including questions that required inference rather than verbatim recall. Later reviews have shown that retrieval practice benefits hold up across many real classroom settings too.

The practical lesson is simple: if you want learning to stick, you cannot rely only on seeing information again. At some point, you need to pull it back out.

That does not have to mean formal testing. It can mean trying to restate the main idea after reading, turning a key point into a question, or revisiting a saved passage and asking yourself what it meant before revealing the explanation.

The important shift is from recognition to reconstruction.

What this means for AI reading tools

This is why "summary versus reading" is the wrong comparison.

The more useful comparison is between two workflows.

Workflow A: compress and move on

hit a difficult passage
ask for a summary
feel caught up
continue reading with only partial understanding

Workflow B: locate the break in understanding

hit a difficult passage
isolate the exact sentence or paragraph that caused the problem
ask for an explanation of that part in simpler language
drill into the concept that is still unclear
continue reading once the local confusion is resolved
later, retrieve the important idea from memory

The second workflow is slower in the moment. It is also much closer to how understanding is actually built.

The point is not to avoid summaries completely. The point is to use them for what they are good at and stop expecting them to do the work of explanation and retention.

This is where Drill fits

Drill is not meant to be just another summarizer.

It is built around a different assumption: the most valuable help usually happens at the exact place where reading breaks.

Instead of pulling you away from the text, Drill keeps the learning work anchored to the original passage.

You can:

highlight the sentence that confused you
get an explanation inline
drill deeper on the part of the explanation that is still unclear
keep your place in the original article
save the passages that are worth revisiting
come back later and turn them into review material

That workflow matters because it combines several things that the learning literature treats as important:

active reading rather than passive skimming
self-explanation rather than surface familiarity
constructive engagement rather than simple consumption
retrieval and review rather than one-time exposure

Drill does not invent these principles. It puts them back where they belong: inside the act of reading.

Summaries still have a place

To be clear, summaries are useful.

They are great for:

deciding whether something is worth reading
getting oriented before a deeper read
reviewing a familiar topic quickly
scanning across many sources

But they are weaker when you need:

to understand a difficult sentence
to unpack a chain of reasoning
to connect a concept to what came before
to retain what you learned after the reading session ends

In those cases, the better question is not "Can you summarize this?"

It is usually one of these:

"What does this sentence mean in plain language?"
"Why does this follow from the previous point?"
"What assumption is the author making here?"
"Can you explain this term using the context of this paragraph?"
"What would I need to remember from this section tomorrow?"

That is a different posture toward reading. Less compression. More understanding.

Read for understanding, not just for relief

One reason people love summaries is that they relieve the discomfort of not understanding.

But relief is not the same as learning.

Learning usually happens when you stay close to the confusion long enough to work on it: clarify the sentence, connect the concept, rebuild the idea in your own words, and then retrieve it again later.

That process is less satisfying in the first 30 seconds. It is much more satisfying the next day, when the idea is still there.

So the next time a hard article starts to lose you, do not ask only for the gist.

Start smaller.

Find the exact line where your understanding breaks. Explain that line. Drill into what still does not make sense. Then see what you can recall once you are done.

If your goal is to really learn from what you read, that is usually where progress starts.

Research notes

This post is shaped by a few recurring findings from learning science:

Dunlosky et al. (2013) reviewed common learning techniques and argued that summarization has narrower benefits than stronger strategies such as practice testing, while self-explanation is especially relevant for comprehension.
Chi et al. (1994) showed that prompting students to self-explain while reading improved deeper understanding of a science text.
Chi and Wylie (2014) proposed that more constructive forms of engagement usually produce more learning than more passive ones.
Karpicke and Blunt (2011) found that retrieval practice produced stronger long-term learning than elaborative studying with concept mapping for science texts.
Agarwal, Nunes, and Blunt (2021) reviewed classroom retrieval-practice research and found robust benefits across a range of real educational settings.
How to Read a Book remains a useful classic on reading as an active process aimed at understanding rather than passive intake.

If you want to test the difference for yourself, take a dense article, use a summary only to orient, then switch to explaining the exact sentence that blocks you. The gap becomes obvious very quickly.