How does Duolingo Make it Stick?

5 Proven Instructional Design Strategies
In this post, I want to look at two things that have made me sit up and think about how I approach instructional design.

One is the book Make it Stick by Peter Brown, Henry Roediger and Mark McDaniel, the other is an app called Duolingo.

It turns out the two are very closely linked. Make it stick – subtitled “the science of successful learning” – is a book about effective learning and instructional design strategies backed up by research. Duolingo uses most of the strategies the book explores and combines them with clever, gameful design. The combination of learning strategies and gamification make for an engaging, and compelling learning experience. I’ll look at the gamification more closely in a future post (this one is long enough). For now, I want to focus on the instructional design strategies.

I’m going to call out each of the strategies, summarise it, highlight how Duolingo does this and then add some ideas about how you and I can use it to design better e-learning experiences.


Instructional Design Strategy 1:

Retrieval practice

What is it?

Let’s start with an instructional design technique you’re probably already using – retrieval practice. Basically, it’s just asking questions.

“Retrieval practice – recalling facts or concepts or events from memory”

It can be as simple as using flash cards or multiple choice questions. It’s better to practise answering questions that actually make you recall the information instead of just reading and re-reading material. By forcing yourself to recall the answer you’re mentally rehearsing what you would do in the real world when you need that information. Every time you recall it, you strengthen the memory.

This approach can seem harder to alternatives like re-reading the content. The fact that it’s harder and slows us down means we’re less likely to want to do it. One of the terms used in the book is ‘desirable difficulties’ from research by psychologists Elizabeth and Robert Bjork. It describes how

“Some activities that slow down learning and require more effort will more than compensate for their inconvenience by making the learning stronger, more precise and more enduring.”

There is a balance to strike here – not too difficult and not too easy.

How does Duolingo do this?

It’s 100% question-based. There’s no long text to read or even videos to watch, you answer questions from the very start. I’m learning Portuguese. Despite having no knowledge of the language I can jump straight in because it gets the difficulty right. It starts with visual multiple choice questions and sticks to using variations of multiple choice most of the time. When it gets harder or introduces new words I can tap on the word for a definition or I can choose to guess it. As my knowledge of those words improves, it takes away the crutch and makes me think for myself. Scaffolding like this is a proven instructional design approach.

Duolingo screenshot1
Duolingo screenshot2
Duolingo screenshot3

How could you use this in e-learning?

Asking questions is one of the most fundamental interactions we have in our tool box. I have to confess that years of designing compliance training made me turn my nose up at basic recall and comprehension questions. Partly, because I had little control over what they were. Subject matter experts tend to get what they want in compliance. But also because it seemed a bit patronising asking basic questions straight after reading the answers. It’s just a test of short-term memory (or the ability to go back and look it up) isn’t it? Well yes, but it’s also the first step to consolidating it into a long term memory.

What makes e-learning different?

Questions (with feedback) are perhaps what distinguishes e-learning from most other content on the web. If you’re not asking questions then it’s just content, even if it’s got some other kind of interactivity. Asking questions that provide the right level of scaffolding and support as people progress is where the skill of instructional design lies. It’s about getting the level of challenge right as the user progresses. Low stakes questions throughout the content help people to understand the content better.

If you feel like this is pretty obvious then bear with me (or skip ahead!) – but it’s important to understand this point because it’s the foundation for the next two strategies.

At the moment we’re only talking about questions to aid comprehension and recall. You can use questions for much more of course – reflection, connection and elaboration which we’ll look at later.

We forget stuff pretty quickly. This kind of questioning alone isn’t going to change that, but the next two strategies just might.

Instructional Design Strategy 2:


What is it?

Repetition is an important part of learning anything. If you want to develop a skill, or you want to move new knowledge into long-term memory, it’s important to space out that repetition and find novel ways of repeating it.

Spaced learning is nothing new, you’ve probably heard of the Ebbinghaus forgetting curve Hermann Ebbinghaus came up with that gem in 1885. There is some dispute over how useful and relevant his experiments were, but there’s more recent and interesting evidence that shows spacing is important for learning. In one experiment two groups of surgeons were given four short lessons in microsurgery – reattaching blood vessels. One group received all four lessons in one day, the other had a week between each class. The surgeons who took all the lessons in one day scored lower on all the measures and 16% of them damaged the specimen’s they were practising on beyond repair.

“It appears that embedding new learning in long-term memory requires a process of consolidation, in which memory traces are strengthened, given meaning, and connected to prior knowledge – a process that unfolds over hours and may take several days.”

So how big should these intervals be?

Some people like to refer back to Ebbinghaus, but that’s mostly because he provides a simple answer. The truth is probably more complex. The right spacing will vary according to context and the individual. The book suggests that…

“Intervals should be far enough apart that practice doesn’t become a mindless repetition. At a minimum, enough time so that a little forgetting has set in.”

Sleep also plays a big part in memory consolidation, so it should be at least a day.

How does Duolingo do this?

Firstly the lessons are bitesize. Each one takes less than five minutes and the aim is to complete two lessons per day. You can do more, but the gamification engine steers you towards little and often.


One way it does this is through streaks. By getting enough XP each day you build a streak. It seems like a fairly meaningless mechanic, but it taps into our desire to get positive feedback and as the streak grows it becomes more valuable, psychologically.

Once you’ve completed two lessons you get enough XP to maintain your streak for the day. It’s a natural cutoff point to stop and switch the app off. Because you can do it whenever you have a few spare minutes, I can complete a lesson while I’m making a coffee or cooking dinner. Is that micro-learning? Who cares – it’s accessible, engaging and rewarding.

So Duolingo isn’t actually managing the spacing. Instead, it uses visual cues and gameful design to influence your behaviour. Content is broken down into chapters and chunks. Once you complete a chapter you’ll notice the “volume” starts to fade. It’s a simple visual representation of your memory of that content.

It’s a bit like spinning plates.

You don’t want any of them to hit the ground. So you go back in and practice them to get the level back up. The more chapters you complete the more time you have to spend revisiting and reinforcing your knowledge.

Duolingo screenshot4
Duolingo screenshot5

How could you use this in e-learning?

You have two choices here. First, you could buy a dedicated tool like Elephants don’t forget or Q-Stream. These are dedicated quizzing tools built around spacing. I haven’t used Q-Stream, but in the case of Elephants, it works like this:

  • You create a bank of questions covering different topics
  • The questions go out by email (2-3 a day)
  • The results are recorded and an algorithm adjusts the frequency and spacing of questions according to what you get right and wrong.
  • It’s a deceptively simple tool that generates really interesting data about teams and managers – not just what they know!

But maybe you don’t have the budget for another tool, so you need to hack something similar together?

What can you do?

If your learning management system is any good you could build multiple mini-quizzes or scenarios and space them out according to your best guess. Don’t let a mediocre LMS limit your instructional design options. You could just use email and Google forms. Build a bank of questions and shuffle them around into different quizzes then schedule the emails for the next 3 months.

If you want to get really funky you could use Google forms, Zapier and an email marketing system like Active Campaign. You could use this combination to build a spacing automation that adapts to how well people are doing on the questions. That way you only have to build the email/quiz sequence once and it’s easy to repeat in the future. Benefits include…

  • a tonne of actionable data
  • massive geek fun to get it working
  • it will cost you less than £10 a month!

Right onto number 3…

Instructional Design Strategy 3:


What is it?

Remember back to your exam days when you would cram one topic over and over and over?

Here’s how my biology revision plan went (at least here’s how I imagine it went, I can’t remember back that far and I doubt I was organised enough to have a plan).

  • Day 1 – Photosynthesis
  • Day 2 – Respiration
  • Day 3 – Cell Division

Turns out that was a dumb way to learn. Learning related topics in silos like this is not effective.  It feels like we’re learning, but all we’re really doing is cycling it through our short term memory. This is called “momentary strength” and it’s an illusion. Memory and learning don’t work like that.

The more effective and counterintuitive approach is to mix up the topics. But that’s harder, so we shy away from it instinctively. Part of the reason it works is that it requires more effort.

“It’s more effective to distribute practice across different skills than polish each one in turn”

How does Duolingo do this?

As mentioned above, language content is broken down into chapters and chunks. But as the lessons progress the older vocabulary is worked back into the exercises.

With language learning that’s quite simple and logical to do. You aren’t going to stop using common words just because you’ve covered them once. But there’s no reason this approach can’t be applied to other disciplines.

How could you use this in e-learning?

One obvious answer is to mix up the questions you use when spacing.

Let’s say you’re running an onboarding program which could spread out over months. You may need to cover dozens of topics that people need to master to become competent. You can test them at the end of each topic, but you could keep adding in a proportion of questions from previous topics. It might go something like this.

Week 1: 100% topic 1
Week 2: 80% topic 2 | 20% topic 1
Week 3: 70% topic 3 | 20% topic 2 | 10% topic 1
Week 4: 70% topic 4 | 20% topic 3 | 10% topic 2
Week 5: Review and consolidate

There’s no magic formula, but interleaving topics can help people make connections between the material. It leads to deeper understanding and faster recall.

The ability to recall information is rather unfashionable. After all, we can always just Google it, but there are still many workplace situations where faster recall equates to cost savings and greater profit.

Imagine you work on a customer helpline. If you can recall the answer to the question without putting the caller on hold, you can deliver a better experience for the customer and move onto the next call more quickly. That’s a measurable performance benefit to your organisation.

A salesperson who can accurately recall the specs of the product they are selling, without looking them up, is going to appear more confident in front of clients and stand a better chance of closing that deal.

Does your organisation expect everyone to contribute new ideas and innovations? Then they’ll need to be able to do more than just look up the answers. They’ll need a deeper understanding of how the business works and how their role contributes to it.

Knowledge still matters.


Instructional Design Strategy 4:

Generation and Desirable Difficulty

What is it?

Is it fair to ask questions about new topics without presenting some relevant information first?

It’s an instructional design method I’ve always liked, so it was good to see that the research backs up the instinct. But it’s also one that has got me into heated discussions with subject matter experts. Before we look at the pros and cons, what is generation?

“The act of trying to answer a question or attempting to solve a problem rather than being presented with the information or solution”

When we have to struggle with a question about something new, we tend to think of terms of reference that could be relevant. That helps us make connections with prior knowledge and deepens learning.

But that’s just one form of generation. It also refers to everything from simple fill-in-the-blank and open input questions to essay writing. In other words, any kind of question that requires you to come up with the answer yourself rather than choose from a list of possible right answers.

Why does this style of question lead to stronger memories and better retrieval? It’s about desirable difficulty which I mentioned further up.

“The more effort you have to expend to retrieve knowledge or a skill, the more the practice of retrieval will entrench it.”

How does Duolingo do this?

It does it in stages. When you start a new topic you get no background information, no learning objectives, just a topic name like “Food”.

Stage 1

The first question is always a graphical multiple choice question – so far they’ve all been nouns. So although you don’t know the vocabulary, even a small child can recognise the right answer.

Stage 2

Not all vocabulary is easy to represent visually – pronouns, some verbs etc. Instead, new structures are introduced with multiple choice answers, so you can make an educated guess. You also have the option to tap for a definition and explanation.

Stage 3

You have to type the answer in, either from what you heard or what you read. These questions have no support, so the stakes are higher. If you get any question wrong in Duolingo you lose a life. It’s a very gamified learning experience, but we’ll look at that in another post.

Stage 4

Beyond the core language learning system, there are other features which encourage higher levels of learning and application. These include chatbots and social learning communities.


There are several characters represented as chatbots in typical situations like restaurants, shops etc. This is your chance to practise using actual phrases in response to “real” native speakers. It’s naturally more challenging, but it works really well. They’ve put a lot of thought into how the conversations flow and I never once felt frustrated that my answer wasn’t recognised (unless it was complete rubbish). You also get additional points for coming up with more elaborate answers. I haven’t tried out the community groups yet, but it’s great to see they offer ongoing options to raise your mastery levels in each language.

Duolingo screenshot6
Duolingo screenshot7

How could you use this in e-learning?

Earlier I mentioned the pros and cons of elaboration in e-learning and how they can get instructional designers into trouble.

The first problem comes when people feel it’s unfair. It’s about high stakes vs low stakes testing and it’s one of the reasons people hate e-learning. If you’re expecting people to answer questions they’ve never encountered before, it needs to be clear that it’s ok to get it wrong. If it’s a compliance course, people are likely to be on edge about getting questions wrong. When we use this approach we need to make it clear which questions contribute to completing a course and which don’t.

User frustration

The other challenge comes when we deviate from multiple choice questions. Any time we ask people to type in answers to questions online we create the potential for user frustration.

If someone types in a correct answer that we didn’t think of, or they made a small typo, they’ll get upset when the software tells them they’re wrong. In Duolingo every feedback box gives the option to report any problems back to Duolingo which is a good way of putting power back in the hands of the user (and picking up on any errors).

I tend to shy away from using open input questions, unless it’s for reflective purposes, because of the trouble I’ve caused myself in the past. You’d think that with advances in predictive text and natural language processing, today’s authoring tools could auto-suggest all the most likely answers, rather than rely on the poor old instructional designer to come up with them and enter them manually. If such a tool exists I haven’t heard about it yet – feel free to enlighten me in the comments!

So what about chatbots?

Surely simulating conversations in safe online environments must have a wide range of applications? Are there any authoring tools that have incorporated this as a feature? I know the latest version of Evolve is heading in this direction, but it’s not actually interactive or branching yet.

Building interactive chat sequences isn’t really that hard. I’ve been playing with a tool called Tars that makes it easy to create branching chat sequences. Here’s an example we built to capture leads on a client’s website. It wouldn’t be difficult to do something similar for a branching scenario. If there isn’t a tool that does this already, expect something on the market soon.


Instructional Design Strategy 5:


What is it?

This post is an example of elaboration. When I read Make it stick I started to notice the strategies from the book in Duolingo. I studied the app and developed a deeper understanding of the strategies because I could see and experience them in the real world.

By taking those observations and reflections and writing them into a rather long blog post, I’ve developed my understanding further. By sharing it with you I’m inviting comment, criticism and further learning opportunities. You might be inspired to write something yourself, you might want to share something you’ve read or watched elsewhere. Now we’re moving into social learning which isn’t really touched on in the book, but you can see how elaboration can be an effective instructional design strategy.


How does Duolingo do this?

As we’ve already seen the Chatbots and online communities (they’re called Clubs in Duolingo) give you options for elaboration. You can practice using different phrases in chatbot conversations and receive instant feedback on how you’re doing.

To be honest I’m not too sure what format the clubs follow because when I tried to join one it said they were all full!

How could you use this in e-learning?

In my opinion, this is an essential one because it’s about learning transfer. It’s that critical and sometimes messy stage, where you take what you’ve learned and start trying to use it in the real world. It’s the experiential side of learning where you take the concept or idea and try to make sense of it. It’s that moment you take the stabilisers off your bike.

This is where reflection becomes incredibly valuable and in e-learning, we don’t always make the best use of it. It doesn’t have to be complicated. Asking people to think and write about how the content you’ve covered relates to their world, or building an action plan to help them take the first steps are really effective. You don’t have to keep it all on the screen. Encourage people to write it down physically (this approach has other positive effects that support learning – more here)

The problem with this approach is that you can’t automate feedback. Especially if people are writing it down in a journal. Instead, you could consider building communities – whether they’re face-to-face or online. This is where social learning comes into its own. If we can find ways for people to share their reflections and relevant experiences, everyone benefits even more.




You may be thinking this focus on retrieval is all a bit old fashioned, when everyone else is talking about performance support, micro-learning and “resources not courses”. That was my first reaction when reading the book, but I like to think I value data and evidence over what feels good.

All the strategies in Make it stick are based on empirical evidence, not marketing soundbites or personal ideologies. The challenge for 21st-century instructional design is to adapt these proven techniques to the changing world of workplace learning and performance, not ignore them. Quizzes and scenario based simulations can be resources too. They don’t have to be prescriptive. It’s irresponsible to ignore what we know works just because it’s not fashionable.


All of these instructional design strategies have the potential to improve the learning outcomes for the people who engage in the experiences we create, but one thing they don’t really touch on is motivation.

I bought Make it stick about eight months ago and I’ve only just finished it. I read the first 40 pages and then didn’t pick it up for another seven months. What changed? I felt a need to invest in my own development. I felt like I was stagnating a little. When that itch had to be scratched I picked the book back up and I started to read a little every day.

Equally, I’ve heard people talking about Duolingo for well over a year, my kids were using it six months ago, but it was only when I booked a trip to Portugal that I was motivated to learn the language and the app became relevant.

Without motivation or at least discipline, instructional design strategies are useless. So here’s a question to lead into a future post – to what extent can gamification or gameful design, build and support motivation for learning where it might be lacking?