The memory problem: Making learning content that actually sticks

by Thomas Espeseth, University of Oslo, Norway.

Every educator and content creator knows this frustration: you craft what you believe is clear,
compelling material, only to discover later that learners remember almost nothing. Despite our best efforts – engaging visuals, interactive elements, careful pacing – the information still evaporates from memory.

It’s not a matter of poor teaching or poor learning; it’s a fundamental challenge that stems from the way our brains work.

When experts miss the obvious

Consider this: trained radiologists, experts at detecting life-threatening lung nodules in CT scans, completely miss a gorilla image inserted directly into the scan – even when it’s much larger and more conspicuous than the average nodule. This striking finding builds on the famous “Invisible Gorilla” experiment, revealing a profound insight: even experts are limited by their attention span.

If radiologists can miss gorillas while looking for cancer, what are learners missing in our carefully designed courses and content?

As a cognitive scientist at the University of Oslo, I have become fascinated by this paradox. Why do we lose important information so easily, and how can we prevent it?

It’s not about the learner – it’s about the content

Our research revealed something unexpected: memorability isn’t primarily about individual learners or their study habits. Instead, certain content consistently sticks better than others, regardless of who’s learning. By analysing large datasets from thousands of participants, we discovered that memorability is largely determined by specific features of the content itself.

It opened an intriguing door for us: if memorability can be predicted, could we create content that is inherently memorable?

Teaching machines to make content stick

Working with neural networks trained on extensive memory data, we developed a system that can predict and enhance how memorable specific information will be – whether text or images. Our AI tool, when applied to text, consistently improves memory retention by an average of 28.4%, significantly flattening the typical forgetting curve.

Figure 1: AI-optimised content significantly reduces forgetting over time compared to standard content.

But the benefits extend beyond just remembering more. Highly memorable content requires less mental effort to retrieve and even mitigates common attention lapses like “attentional blink” – those brief moments when we miss important information because our attention is temporarily unavailable. The more memorable the content is, the stronger the neural activity in the brain regions associated with attention and learning.

Figure 2: Brain imaging shows heightened activity in regions critical for visual attention and
memory when learners engage with AI-optimised content.

From classrooms to boardrooms: where it works

The applications span everywhere learning matters. Textbook authors can craft explanations that naturally embed in long-term memory. Online course creators can ensure key concepts survive beyond the final quiz. Safety training developers can create materials that stick when lives depend on remembering. Corporate trainers can deliver impactful programmes that employees genuinely retain, while healthcare providers can ensure vital patient information and safety guidelines are consistently remembered.

Take a complex topic like photosynthesis. To avoid dense, technical explanations that students find hard to remember, our system identifies linguistic patterns that resonate with how the brain stores information. The result: students can remember the explanations for months after reading them.

Figure 3: In our AI optimisation tool, you can simply insert your text, click enhance, and get content optimised for maximum impact and memorability, as in this example.

The technology also transforms marketing and media content. With over $800 billion spent
annually on advertising that few remember, the potential impact is substantial. Content producers are using our system to optimise video captions, social media posts, and campaigns for maximum retention.

Putting it to the test

In our validation studies, I’ve been particularly encouraged by promising user adoption: 80% of users choose AI-optimised versions over their original content, rating the system as “excellent” in usability. It’s noteworthy that we’re witnessing a significant boost in both behavioural memory tests and brain-based memory measures. Rather than fighting against human cognitive limitations, the technology works with them, creating content that effortlessly embeds itself in memory.

The future of unforgettable content

This isn’t about replacing proven educational methods – it’s about enhancing them. If we can make content more memorable, we can support learners more effectively and sustainably. By combining cognitive science with AI, we develop innovative ways to enhance communication and learning. Instead of placing the entire burden on learners to memorise through sheer effort, we can ensure that information itself naturally stimulates memory. For educators and content creators tired of watching their best efforts fade from memory, this represents a fundamental shift: from hoping content will be remembered to ensuring it will be.

Although memorability optimisation holds substantial potential, it requires careful planning to ensure accuracy, avoid oversimplification, and achieve alignment with human psychology. Looking ahead, I’m especially interested in exploring the deeper interactions between curiosity, attention and learning, and expanding our technology to multimedia and interactive environments.

If you’re interested in exploring how memorable your own content can become, you can test our text memorability optimisation tool directly here. Please note that the tool is currently in active development – while it demonstrates our core approach, we’re continuously refining the algorithms and expanding our data. We welcome feedback from educators, researchers, and content creators as we work toward the full potential of this technology.

Editor’s note: We were delighted to have Thomas as a keynote speaker at the Media & Learning 2025 conference this year in Leuven. You will find a recording of his presentation under Keynotes on this page.

Author

Thomas Espeseth, affiliated with the Department of Psychology at the University of Oslo, TRUST, The Norwegian Centre for Trustworthy AI, and Mentalese AS.