Digital learning and spelling: what the research says
Which parts of a learning app are evidence-based, which are plausible bets, and which are marketing.
Can software actually teach spelling?
For a long time the honest answer was "probably, but nobody has checked properly": reviewers noted that evidence for digital tools improving spelling specifically was scarce compared to reading. That has started to change. A randomised controlled trial of an adaptive spelling game combining sound-processing work, letter–sound correspondences and spelling rules found gains not only on trained words but on untrained words sharing the trained patterns, against an active control group (digital spelling RCT, 2023). The caveats are real: 65 children, twelve structured weeks, and German, whose spelling is far more rule-regular than English. But it establishes the category: well-designed software can produce measurable, generalising spelling gains.
Notably, the features that drove generalisation in that trial were instructional, not technological: the game was organised around named spelling patterns and taught the underlying skills explicitly. Software makes good practice scalable; it does not make weak practice effective.
Adaptive difficulty: matching the task to the learner
The most substantive thing software adds is adaptivity: measuring the learner and serving material at their level, continuously. A standard engine for this is the Elo rating system borrowed from chess, which estimates the learner's ability and every item's difficulty at once, updating both on each answer; the research literature describes it as simple, robust and effective for educational systems (Pelánek, 2016). The best-documented deployment is Math Garden, an adaptive arithmetic system in Dutch schools: thousands of children, millions of answers, items deliberately served at about 75% success, and difficulty calibrated on the fly (Klinkenberg, Straatemeier & van der Maas, 2011). Adaptivity matters because the alternative, one fixed word list for everyone, guarantees the list is too easy for some learners and demoralising for others.
Spaced repetition, industrialised
The spacing effect (covered in what makes spelling practice work) is the learning principle software implements best, because software remembers when you last saw every item. Modern schedulers go beyond fixed review ladders: they fit a memory model per item and schedule each review to hit a chosen probability of recall. Duolingo's published half-life regression model, trained on millions of learning traces, cut recall-prediction error substantially against baselines (Settles & Meeder, 2016); the open-source FSRS scheduler used by Anki reports needing 20–30% fewer reviews for equal retention (FSRS documentation). Read those two claims carefully, because they are different kinds: predicting recall better is peer-reviewed but is not itself a learning gain, and the fewer-reviews figure comes from the project's own benchmark. Both systems also agree on something refreshingly honest: there is no universal optimal schedule, and the right intervals differ by learner and material.
Gamification: what survives scrutiny
Points, streaks, celebrations and ratings are the most visible part of learning apps and the least certain part of the evidence. The headline meta-analysis found significant positive effects of gamification on learning outcomes, but when restricted to the most rigorous studies, the cognitive effect held while the motivational and behavioural effects did not (Sailer & Homner, 2020). Meanwhile the overjustification literature warns that expected, tangible, performance-contingent rewards can crowd out intrinsic interest in an activity a person already enjoys (Deci, Koestner & Ryan, 1999). That is a conditional risk rather than a law, and a 2023 meta-analysis found gamification can also support intrinsic motivation when rewards align with meaningful goals.
Streaks deserve their own honesty note. The retention statistics quoted for them come from company statements and trade press, not controlled studies, and users who keep streaks were plausibly going to keep practising anyway. Even the "losses loom twice as large as gains" figure usually invoked to explain streaks was estimated from monetary gambles (Tversky & Kahneman, 1992), not from any learning app. Streaks are an industry-standard bet, not a demonstrated learning mechanism, and that is a distinction any app's marketing is unlikely to make for you.
The wording of feedback matters
A consistent finding from developmental research: praise aimed at the person ("you're so smart") orients learners toward proving themselves and produces more helpless responses after later failure, while praise aimed at the process (effort, strategy) is associated with seeking challenge and persisting (IES Regional Educational Laboratory, 2025). The related growth-mindset interventions show only small average effects in meta-analysis, so the durable takeaway is the narrower one about wording. Most of this evidence comes from children; its extension to adults is plausible but untested. For an app, it is a cheap, risk-free design principle: describe what the learner did, not what they are.
Typing, handwriting, and what a keyboard changes
Does practising spelling on a keyboard teach as well as writing by hand? The evidence is genuinely mixed. One study found strong spelling learning with no overall difference between printing and typing, though children's keyboarding skill affected how well the typing group learned (printing vs typing study). Other work finds handwriting advantages in school-age children (typing and handwriting comparison, 2025). The recurring confound is that children in these samples type far less fluently than they write, which could account for the handwriting edge on its own; whether it persists for adults who type fluently has not been tested. The practical takeaway is not "avoid keyboards" but "know that input skill adds noise": a typed spelling test partly measures typing.
Why hearing the word matters
Working memory is the bottleneck of all instruction: it holds only a few elements at once, and presentation design either spends that capacity on learning or wastes it (Centre for Education Statistics and Evaluation, 2017). One well-supported way to spend it well is to move the prompt off the visual channel: people learn better when words are spoken rather than printed alongside visual material, because the two stop competing (Mayer). For spelling there is a more fundamental reason the prompt should be spoken: spelling is the translation of sound to letters. Structured-literacy guidance treats hearing-to-writing as the defining direction of spelling practice, distinct from reading's letters-to-sound (International Dyslexia Association). A dictated word exercises the skill; a copied word does not.
Accessibility is part of the design, not a checkbox
Dyslexia affects a meaningful share of the population: the International Dyslexia Association cites figures as high as 15–20% for some symptoms, with wide variation in severity (IDA, Dyslexia Basics). People who struggle with reading and spelling are disproportionately the people a spelling tool exists for. Concretely, accessibility guidance converges on: clear, non-decorative typefaces (when the best-known specialised "dyslexia font" was tested directly, it did not outperform an ordinary sans-serif, per Kuster et al., 2018); text that survives user-applied spacing adjustments, which is what WCAG's text-spacing criterion actually requires (W3C, Text Spacing); colour never carrying meaning alone, which is its own WCAG criterion (W3C, Use of Color); and spoken prompts with a replay control, which remove the reading demand from the question itself. An audio-first spelling task is, incidentally, one of the more accessible task designs possible for this audience: the impaired channel is not where the question arrives.
Questions worth asking of any learning app
The research above compresses into a short evaluation checklist for a teacher or parent looking at any spelling product:
- Does the learner produce spellings from memory, or mostly recognise, copy, or rearrange them? Production is the skill; recognition is easier and teaches less.
- Do missed words come back later, on another day, not just later in the session? Spacing across days is where retention comes from.
- Is the word spoken, and can the learner replay it? Spelling practice without the pronunciation is pattern-matching, not spelling.
- What happens on an error? A useful tool shows the correct form promptly and gets the learner to engage with the correction, not just tap past it.
- Does difficulty adapt, or does every learner get the same list?
- When the marketing quotes numbers, are they from published studies or from the company's own reporting? Both exist in this industry, and they are not the same kind of claim.
Keep reading
The two companion guides cover the underlying science: how people learn to spell and what makes spelling practice work. For how these principles shape SpellingIQ's own choices, see the philosophy page.
References
- Digital spelling RCT (2023). Digital game-based spelling intervention for children with spelling deficits. Learning and Instruction. Randomised controlled trial, 65 German-speaking children
- Pelánek (2016). Applications of the Elo rating system in adaptive educational systems. Computers & Education
- Klinkenberg, Straatemeier & van der Maas (2011). Computer adaptive practice of Maths ability using a new item response model for on the fly ability and difficulty estimation. Computers & Education
- Settles & Meeder (2016). A Trainable Spaced Repetition Model for Language Learning. Proceedings of ACL
- FSRS documentation. The optimal retention (Free Spaced Repetition Scheduler wiki). Community/vendor documentation; its efficiency figures are self-benchmarked, not peer-reviewed
- Sailer & Homner (2020). The Gamification of Learning: a Meta-analysis. Educational Psychology Review
- Deci, Koestner & Ryan (1999). A meta-analytic review of experiments examining the effects of extrinsic rewards on intrinsic motivation. Psychological Bulletin. Contested, and its scope conditions (tangible, expected, performance-contingent rewards) are often dropped in citation
- Gamification and intrinsic motivation meta-analysis (2023). Effects of gamification on intrinsic motivation (SDT-framed meta-analysis). Educational Technology Research and Development
- Tversky & Kahneman (1992). Advances in prospect theory: Cumulative representation of uncertainty. Journal of Risk and Uncertainty
- IES Regional Educational Laboratory (2025). A Closer Look at Growth Mindset Research
- Printing vs typing study. The write way to spell: printing vs. typing effects on orthographic learning
- Typing and handwriting comparison (2025). Comparing typing and handwriting on spelling performance in school. Scientific Reports
- Centre for Education Statistics and Evaluation (2017). Cognitive load theory: Research that teachers really need to understand. NSW Department of Education
- Mayer. Cognitive theory of multimedia learning: the modality principle (overview)
- International Dyslexia Association. Structured Literacy Defined: An Explanation and Rationale
- IDA, Dyslexia Basics. Dyslexia Basics. International Dyslexia Association. The 15–20% figure is IDA's estimate for some symptoms of dyslexia; prevalence estimates vary with the definition used
- Kuster et al. (2018). Dyslexie font does not benefit reading in children with or without dyslexia. Annals of Dyslexia
- W3C, Text Spacing. Web Content Accessibility Guidelines 2.1, Success Criterion 1.4.12 Text Spacing
- W3C, Use of Color. Web Content Accessibility Guidelines 2.1, Success Criterion 1.4.1 Use of Color