Captions improve learning outcomes for hearing learners: the research case every L&D team should know

Why this is not an accessibility-only question: the research-framing problem in L&D

The dominant framing of captions in L&D literature and practice is the accommodation model: captions exist to make content accessible to learners who cannot access audio. This framing is accurate as far as it goes. ADA Title II, Section 504 of the Rehabilitation Act, and the 2026 enforcement deadline for public universities and state entities all operate within an accommodation model — the obligation is to provide access to individuals who would otherwise be excluded.

The accommodation model has a practical consequence for budget and procurement: it scopes captioning as a cost centre serving a small beneficiary population. If your training programme has a 2% hearing-impaired learner population, the accommodation model calculates captioning investment against 2% of learners. That framing consistently produces budget pressure: captioning is expensive per-learner when the served population is small, and it gets cut or delayed when resources are constrained.

The learning-effectiveness framing changes the calculation entirely. If accurate captions improve comprehension, retention, and completion for 100% of learners — at rates that vary by learner profile but are positive across every studied group — then captioning investment should be calculated against the entire learner population. The return-on-investment conversation with a finance audience is structurally different when the numerator is “improved training outcomes for all employees” rather than “accessibility compliance for 2% of employees.”

The research base that supports the learning-effectiveness framing has been building since the 1980s. The core finding — that redundant presentation of the same verbal information in audio and text simultaneously improves comprehension for hearing listeners — was established in controlled laboratory studies before most of today’s L&D practitioners were in the workforce. What has changed in the past fifteen years is the accumulation of applied research: studies in real educational and corporate training settings, with actual learner populations, measuring outcomes that L&D professionals care about (comprehension quiz scores, retention at follow-up intervals, completion rates, time-on-task). That applied research is now large enough to be treated as settled.

The problem is that most L&D practitioners have not read it. Caption research is published primarily in educational psychology, second-language acquisition, and human factors literatures — not in L&D trade publications. The practitioners who have read it are more likely to be instructional designers with graduate research backgrounds than training operations managers. This creates a gap: the people making caption budget decisions often have access only to the compliance framing, while the research framing exists but has not been translated into the business-case vocabulary that budget conversations require.

This post is that translation. It does not require reading the primary research literature. It requires understanding three core mechanisms — dual-coding, cognitive load, and ESL scaffolding — well enough to frame a business case, plus two applied findings — mobile audio dependency and caption-as-search-index behaviour — that are available from LMS data most organisations already have.

For the compliance framing that you still need in parallel, see how to build a caption compliance programme for L&D and the accessibility maturity model for captioning. The two arguments reinforce each other and should be made together in any budget or procurement conversation.

Dual-coding theory and captions: the verbal + visual channel mechanism

Dual-coding theory, developed by Allan Paivio beginning in the 1970s, proposes that humans process information through two distinct cognitive subsystems: one specialised for verbal information (words, whether spoken or written) and one specialised for non-verbal information (images, spatial relationships, visual patterns). The two systems are independent but can operate in parallel and can form referential connections to the same concept. When both systems encode the same information simultaneously, retention and retrieval improve because there are two independent memory traces rather than one.

Richard Mayer extended dual-coding theory into multimedia learning in a research programme spanning thirty years. His work produced eleven principles of multimedia design, each supported by experimental evidence, that specify how to structure audio, visual, and text information in educational media to maximise comprehension and retention. Two of those principles are directly relevant to captions.

The redundancy principle in Mayer’s original formulation suggested that adding text captions to an animation with spoken narration was counterproductive — the learner would have to split attention between reading and listening, and the redundancy of the same verbal information in two modalities would tax working memory rather than supporting it. That finding held for animated instructional content where the visual was the primary information carrier and the narration explained the visual.

But subsequent research — including replications by Mayer himself and substantial applied research in video-lecture and training-video contexts — found that the redundancy effect was context-dependent. In video lectures and training videos, where the visual channel carries much less instructional information than in animations (a speaker talking is visually less demanding than an animated diagram), the verbal channel capacity freed from visual processing is available to support caption reading. In those contexts, captions function as a redundant verbal channel that activates the same verbal system twice — once through auditory processing of the spoken narration, and once through the visual processing of written text. The result is stronger verbal encoding, not cognitive competition.

The practical implication: for the vast majority of training videos in a typical L&D programme — talking-head expert interviews, screen-recorded software walkthroughs, live-event recordings, instructor-led training recordings — the dual-coding benefit of captions is positive. The animated-diagram case where captions create visual competition is a minority of corporate training content. L&D teams producing primarily video-lecture-style and screen-capture content can expect positive learning-outcomes effects from accurate captions across their portfolio.

The channel distinction also explains why caption accuracy matters for the dual-coding benefit. The redundant-verbal-channel mechanism works when the caption text matches the audio narration closely enough that both channels are encoding the same information. When caption text diverges from audio — a speaker says “QuantumStack SDK” and the caption renders “quantum stack SDG” — the two verbal channels are encoding different information. The learner’s working memory must now reconcile two conflicting verbal representations rather than encoding one reinforced representation. The dual-coding benefit disappears and is replaced by cognitive cost. This is the mechanism-level explanation for why 99% caption accuracy is a learning-effectiveness threshold, not just a compliance number.

The glossary-biased captioning approach addresses this directly. When a caption system knows that “QuantumStack SDK” is a term in the organisation’s product vocabulary, it can bias its decoding toward that term and produce the correct caption text. The dual-coding benefit is preserved. When a general-purpose auto-captioning system encounters the same audio, it renders based on phonetic proximity to common English words, producing errors on exactly the terms that matter most for the training content. See how glossary-biased captioning improves accuracy on engineering terms and the comparison of glossary methods with prompting and fine-tuning for the technical detail behind this distinction.

Cognitive load and caption accuracy: why inaccurate captions harm learning

Cognitive load theory, developed by John Sweller beginning in the 1980s, provides a second mechanism for understanding caption effects on learning. The theory proposes that human working memory has limited capacity, and that instructional design should minimise unnecessary demands on that capacity (extraneous cognitive load) while supporting the demands that are necessary for learning (germane cognitive load).

Accurate captions, in the right context, reduce extraneous cognitive load. For learners in a noisy environment, for learners with mild undiagnosed hearing difficulties, for learners whose native language is not English, and for all learners in content with dense technical vocabulary, understanding the audio stream requires active effort that draws on working memory capacity. Accurate captions reduce that effort by providing a redundant written record of the verbal content. The learner who missed a word in audio can resolve it from caption text without the cognitive cost of rewinding and replaying. The working memory capacity freed by that reduction in audio-processing effort is available for germane processing — integrating the new information with existing knowledge, connecting terms across the content, constructing the mental model that the training is designed to produce.

Inaccurate captions do the opposite. When caption text conflicts with audio, the learner must resolve the conflict. That resolution process is extraneous cognitive load: it consumes working memory capacity without contributing to the learning goal. Research by Donna Gilmore and others has documented this explicitly in second-language-learning contexts: learners exposed to caption errors had lower comprehension scores than learners with no captions at all, because the error-resolution cost exceeded the benefit of the redundant verbal channel.

The accuracy threshold at which this reversal occurs is not precisely established across all learner populations and content types, but the research cluster suggests it lies somewhere between 96% and 98% word-level accuracy. At 98% or above, the comprehension benefit of captions is positive across essentially all studied learner groups. At 90–96%, effects are mixed and population-dependent. At 80–90% — the accuracy range of general-purpose auto-captions on typical training content — the effect on hearing learners can be negative.

This finding has direct implications for the “auto-captions are better than nothing” argument that is sometimes made in content backlogs and remediation triage decisions. That argument is correct for deaf and hard-of-hearing learners, for whom 80% accurate captions provide substantially more access than no captions. It is not necessarily correct for hearing learners who are using captions primarily for comprehension support — for those learners, 80% accurate captions may be pedagogically counterproductive, producing lower outcomes than video without captions at all.

The practical implication for caption accuracy standards in vendor contracts: the 99% WCAG 2.1 AA threshold and the DCMP accuracy standard are not arbitrary regulatory numbers. They reflect a learning-science floor — the accuracy level at which the comprehension benefit of captions is reliably positive across learner populations. L&D teams that accept vendor contracts with lower accuracy standards because the vendor’s price is more attractive are accepting a service that may actively harm the training effectiveness of their programmes for the hearing-learner majority, while remaining legally non-compliant for the hearing-impaired minority.

The cognitive load argument also strengthens the case for glossary-accurate captioning specifically. Technical vocabulary items — product names, acronyms, proper nouns — are disproportionately high in cognitive cost when misrendered. A learner who encounters “QuantumStack SDK” in audio and “quantum stack SDG” in caption text faces not just an error-resolution cost for one word-pair but a vocabulary-anchoring failure: the incorrect text form competes with the correct audio form for the label slot in the learner’s mental model. That competition can interfere with retention of the term itself — a learner who is trying to learn what QuantumStack SDK is and how it works is now carrying a partially incorrect label for the concept. The instructional designer who meticulously crafted the definitional sequence for “QuantumStack SDK” in the training content has had their work partially undone by a caption rendering error.

ESL and EFL learner outcomes: the most consistent finding in the caption literature

The single most consistent finding in three decades of caption research is that captions improve comprehension and retention for learners whose native language is not the language of the audio content. This finding is consistent across proficiency levels from beginner to advanced, across source-language backgrounds, across content types, and across educational and professional training contexts. It is also the finding with the clearest practical implication for enterprise L&D programmes: the global workforce is not monolingual.

The mechanism is straightforward in structural terms. Audio comprehension for second-language listeners requires phoneme identification, word segmentation, lexical access, syntactic parsing, and semantic integration in real time — a processing chain that is substantially more demanding for L2 listeners than for native speakers, because each step requires more deliberate processing and has higher error rates. The result is a higher cognitive load for the same audio content, leaving less working memory capacity available for the germane learning tasks that the training is designed to produce.

Captions reduce this processing burden by providing a written version of the audio that the learner can process through the visual-verbal channel. For L2 listeners, written text is often more accessible than spoken audio — readers can control their processing speed through fixation time in a way that listeners cannot control the speech rate of audio. The combination of audio and caption text allows the L2 listener to use audio for prosodic and contextual cues while using the caption text for lexical access and syntactic parsing, a division of processing labour that reduces total cognitive load.

The proficiency-level finding is particularly important for enterprise L&D. Studies of caption effects on advanced-proficiency L2 learners — learners who score at IELTS 7+ or TOEFL 100+ equivalents, the proficiency level of many employees working in English as a second or third language in multinational organisations — consistently find positive comprehension effects from captions even at high proficiency levels. The magnitude of the effect is smaller than at lower proficiency levels, but it is positive and measurable. This means that the threshold below which captions provide no learning-effectiveness benefit for L2 listeners is above the proficiency level of most professional ESL/EFL users in enterprise settings.

The vocabulary interaction is also significant. Research by Markham, Peter, and others has found that caption access improves incidental vocabulary acquisition from video content for L2 learners: exposure to a new vocabulary item simultaneously in audio and written form produces stronger lexical retention than audio alone. In a training context where the L2 learner is encountering not just general English vocabulary but domain-specific technical vocabulary — product names, industry acronyms, platform-specific terminology — the caption benefit for vocabulary retention compounds with the comprehension benefit. A learner who has heard and read “SCORM xAPI” in a caption is better positioned to retain and deploy that term than a learner who has only heard it spoken.

This compounds the glossary-accuracy argument from a different direction. For native-English hearing learners, an error on “xAPI” rendered as “X API” is a minor disruption that the learner can resolve from context. For an ESL learner who is simultaneously trying to acquire the term itself and understand its instructional context, the same error creates both a vocabulary-acquisition failure and a comprehension disruption. The learner cannot distinguish whether “X API” in the caption is a different term from “xAPI” in the audio, a capitalization convention, or a captioning error. The cognitive cost of that ambiguity is higher for the ESL learner than for the native-English learner.

For enterprise organisations with substantial ESL learner populations — which includes most multinational organisations, many regional healthcare systems, and virtually all technology companies — the caption-effectiveness argument is strongest on ESL-population size alone, independently of any compliance consideration. A 10% improvement in comprehension outcomes for 30% of your learner population is a substantial return on the captioning investment, and it has nothing to do with ADA Title II. See captioning as a DEI investment in L&D programmes for the organisational equity framing that complements the learning-effectiveness argument.

Noisy environments and mobile learning: captions as the primary audio channel

The dual-coding and cognitive-load research frames captions as a supplement to audio — a redundant verbal channel that helps learners process audio content more effectively. That framing assumes the learner can hear the audio. For a growing fraction of learning events in modern enterprise training programmes, that assumption is wrong.

Mobile device penetration in enterprise LMS access has been growing consistently. Most organisations with modern LMS platforms (TalentLMS, Docebo, Absorb, Cornerstone, Workday Learning) now report that 30–50% or more of course completions occur on mobile devices. Mobile video consumption behaviour is qualitatively different from desktop video consumption in one important respect: mobile learners frequently watch video without audio.

The reasons are environmental. Commuting by public transit is the most documented case: audio playback is disruptive to other passengers, earphone use varies by personal preference and availability, and transit environments are often too noisy to make audio comprehension reliable even with earphones. Shared workspaces — open-plan offices, hospital break rooms, retail back offices — present similar constraints. Watching a training video on a phone in a colleague’s eye-line with audio playing is a social intrusion that most learners avoid. Brief learning sessions during work breaks in environments where audio playback would be conspicuous produce learner behaviour patterns that systematically exclude audio.

In these contexts, captions are not a supplement to audio. They are the primary channel through which the learner is accessing the verbal content of the training. The learner has decided — for entirely rational environmental reasons — to process the training through the visual channel only. If accurate captions are present, the learner can access the full verbal content of the training through caption text. If captions are absent, the learner is watching a silent film with visual cues only, and the information density they can extract from training content is dramatically lower.

The completion-rate evidence is consistent with this pattern. LMS data from organisations that added captions to previously uncaptioned mobile-heavy content consistently show completion rate improvements in the 15–30% range for mobile completions, with smaller improvements for desktop completions. The effect magnitude is larger for longer content: a 5-minute microlearning module is completable on mobile without audio because the visual content and brief duration allow coherent consumption. A 45-minute compliance training module cannot be coherently consumed on mobile without audio unless captions are present. Organisations with high rates of longer mandatory training — annual compliance refreshers, new-hire onboarding modules, certification courses — see the largest caption impact on mobile completion rates.

This finding changes the ROI calculation for captions in ways that are difficult to ignore. If 40% of your mobile completions are occurring without audio, and 30% of all completions are on mobile, then approximately 12% of all training completions are occurring through caption text as the primary verbal channel. Any analysis of caption effectiveness that treats captions as a secondary channel for that 12% of completions is structurally wrong about the mechanism. Captions are the training for those learners in those sessions. Their quality is not a supplement to training quality — it is training quality.

The measurement approach for this finding requires LMS analytics segmentation that many organisations do not routinely perform: completion data segmented by device type, with cross-tabulation against caption availability. If your LMS platform supports this segmentation (most modern platforms do), running this analysis for a representative sample of your content portfolio will typically surface a completion-rate gap that makes the business case for captioning in terms that a finance audience will understand without any reference to learning theory at all.

Video review and caption-as-search-index behaviour: the retrieval-practice mechanism

The final mechanism is behavioural rather than cognitive-process research: the observation that learners who have access to accurate captions use video content differently, in ways that are consistent with deeper learning strategies.

The key behaviour pattern is segment re-watch with caption-guided navigation. In LMS platforms with video caption display, learners who access captions show higher rates of pausing and rewinding than learners without captions, but the rewinds are more targeted: they cluster around content boundaries (transitions between concepts), assessment-relevant passages (the minute before a quiz question that tests specific terminology), and dense technical sequences (the explanation of a process with multiple named steps). This is qualitatively different from the frustration-rewind pattern of learners who cannot follow the audio without captions — it is a deliberate retrieval behaviour in which the learner uses the caption text as a content map to navigate to a specific moment they want to review.

This behaviour is a form of retrieval practice, one of the most robustly supported learning strategies in educational psychology. Retrieval practice — the process of actively recalling information from memory, as opposed to passively re-reading it — improves long-term retention substantially compared to passive review. The mechanism by which captions enable retrieval-practice-style video review is the availability of the caption text as an index. The learner remembers that a concept was explained in the video, uses the caption stream to navigate to the relevant section, and re-processes the explanation. Without captions, the same learner must either rewatch the entire video or use chapter markers if they exist — a much higher friction re-engagement path that most learners do not take.

The LMS data signature of this behaviour is detectable with standard analytics: video engagement rate (the fraction of learners who replay at least one segment), segment replay clustering (whether replays concentrate on specific content passages or are distributed randomly), and assessment score correlation with replay behaviour (whether learners who replay specific segments perform better on questions about that content). Organisations with LMS platforms that provide granular video engagement data (Kaltura, Panopto, and Docebo with native video have the most robust reporting here) can test the retrieval-practice hypothesis directly from their own content analytics.

The assessment-score correlation is the business-case clincher for the learning-effectiveness argument. If learners who caption-navigate to specific content segments before answering assessment questions perform measurably better on those questions than learners without caption access who answer without segment review, then captions are demonstrably improving measured learning outcomes. That finding — translatable directly into “training effectiveness” language — is the basis for a captioning investment case that does not require any reference to accessibility law at all.

The caption-as-search-index behaviour also has implications for how to frame caption ROI for a finance audience. The standard compliance framing presents captioning as a cost: spend X to achieve compliance, avoid penalty Y. The learning-effectiveness framing presents captioning as an investment: spend X, and measurably improve the training outcomes that justify your L&D budget. In organisations where the L&D function is under productivity pressure — which is most organisations post-2024 — a framing that positions captioning as a performance investment rather than a compliance cost has a structurally better chance of surviving a budget review.

The glossary accuracy dimension surfaces again here. Caption-guided navigation to a specific content segment depends on the caption text accurately representing the audio at that segment. A learner who is searching for the explanation of “QuantumStack SDK” using the caption stream will not find it if the caption at that moment reads “quantum stack SDG.” The search-index function of captions breaks at the point where the index entry is wrong. For technical training content with dense product vocabulary, glossary-accurate captions are the prerequisite for the retrieval-practice mechanism to operate. General-purpose auto-captions at 80–90% accuracy will produce a search index with errors concentrated on the terms most likely to be the subject of learner searches — because those terms are the high-frequency technical vocabulary that auto-captions fail on.

What this means for L&D programme design

The research findings above converge on a set of programme-design implications that are distinct from the compliance-only framing. The compliance frame drives caption production as a remediation exercise: audit content, identify non-compliant material, produce captions, verify accuracy. The learning-effectiveness frame drives caption production as a programme-design standard: captions are a required component of video-based training design, produced as part of the production workflow, not added after the fact.

Caption before publish

The most effective organisational change that follows from the learning-effectiveness framing is treating caption production as a production gate, not a post-production remediation step. A training video that ships without captions has shipped a degraded learning artefact. The review and approval gate for training content should include caption verification the same way it includes content-accuracy review, accessibility review, and instructional-design quality review.

This requires a production workflow that generates captions in parallel with video production rather than sequentially after it. The organisations that have implemented this most successfully treat the SRT or VTT file as a required deliverable in the same brief as the video file — the video is not accepted into the LMS without both the MP4 and the caption file attached. See caption API automation and webhook workflows for LMS integration for the technical implementation patterns.

Glossary-accurate captions as a vocabulary reinforcement tool

The ESL/EFL research finding has a design implication that is separate from accuracy compliance. When a learner encounters a product term, technical acronym, or industry concept in training content, they are often encountering the term for the first or second time. The caption text is a simultaneous written presentation of that term that can function as a vocabulary-acquisition scaffold: the learner hears the term in context and reads it in text simultaneously, a dual-channel encoding event that improves retention of the term itself.

This only works if the caption accurately renders the term. A training programme that uses “DCMP” (Described and Captioned Media Program) throughout its accessibility training, but whose captions render that term as “D-C-M-P,” “DCMB,” or “deep-sea-am-pee” (all attested auto-caption failure modes for this acronym) is not providing a vocabulary-acquisition scaffold for that term — it is providing a vocabulary confusion event. The glossary architecture that prevents these errors is simultaneously an accuracy-compliance mechanism and a learning-design quality mechanism.

For reporting caption programme effectiveness to leadership, the vocabulary-reinforcement framing gives you a metric that is visible in assessment data: track assessment question performance on technical terminology items across content with glossary-accurate captions versus content with general auto-captions. The performance gap, if measurable in your LMS data, is a direct learning-effectiveness metric that translates into business value without any reference to compliance law.

The learning-effectiveness argument in procurement decisions

Procurement conversations about caption vendors are typically structured around price and compliance: which vendor produces captions at 99%+ accuracy at an acceptable cost. The learning-effectiveness research provides additional selection criteria that are harder for a price-competitive vendor to satisfy:

• Glossary accuracy on domain vocabulary. The vendor’s accuracy metric should include performance on your organisation’s specific technical vocabulary, not just general English. See how to evaluate vendor accuracy and pilot programme design for the evaluation methodology.
• Caption timing precision. The learning-effectiveness benefits of captions — particularly the retrieval-practice navigation behaviour — depend on caption segments aligning closely with the audio they represent. Captions that lag the audio by more than 100–200 milliseconds, or that use very long segments (10+ seconds per caption card), reduce the usability of captions as a navigation index. Timing specifications should appear in vendor contracts alongside accuracy specifications.
• LMS compatibility for caption display. The mobile-completion and retrieval-navigation benefits of captions only materialise if learners can actually see caption text during video playback in your LMS. Caption delivery format compatibility — VTT for HTML5 players, SRT for players with format conversion, TTML for Kaltura and enterprise players — is a prerequisite for any of the learning-effectiveness benefits to reach the learner. See the SCORM and xAPI caption delivery and tracking guide for the format compatibility detail by platform.

Connecting the compliance and effectiveness arguments in business-case conversations

The most effective business case for captioning investment combines both arguments in a structure that addresses different audience concerns. For legal and risk stakeholders: the compliance framing (ADA Title II, Section 508, WCAG 2.1 AA, EAA) establishes the regulatory floor and the penalty exposure for non-compliance. For learning-effectiveness and L&D leadership stakeholders: the dual-coding, cognitive-load, and ESL-outcome research establishes the positive return on investment beyond compliance. For finance stakeholders: the mobile completion rate gap and the assessment-score correlation data translate the research into measurable business outcomes that do not require familiarity with educational psychology.

The ROI framing guide for finance executives walks through this argument structure in detail. The leadership reporting guide covers how to present the combined case in a board or executive team context. Together, they give you the full business-case architecture for captioning as both a compliance requirement and a performance investment.

Seven questions L&D teams ask about the research

Does caption timing matter for the learning-effectiveness benefit, or is accuracy the only variable that matters?

Both matter, but they affect different mechanisms. Accuracy affects the dual-coding and cognitive-load benefits: inaccurate captions create conflict between audio and text channels, increasing extraneous cognitive load. Timing affects the synchronisation prerequisite for dual-coding: captions that appear significantly before or after the corresponding audio break the simultaneous-channel-activation condition that Paivio and Mayer identified as the mechanism for enhanced verbal encoding. Research suggests that caption delays of more than 250–300 milliseconds reduce the dual-coding benefit, and that caption segments spanning 10 or more seconds make it difficult for learners to anchor caption text to the specific audio passage it represents. Vendor specifications should include both accuracy (99%+ word accuracy) and timing constraints (maximum segment duration and maximum audio-to-text delay). The DCMP timing guidelines provide a practical standard for the timing dimension that parallels the WCAG accuracy standard. See caption accuracy standards in vendor contracts for both dimensions in a contractual specification format.

Do accurate captions outperform printed transcripts for learning outcomes?

Yes, consistently in the research literature, though the effect depends on what is being measured. Transcripts have an advantage for content look-up after the fact: learners can read a transcript as a document and find specific passages faster than they can navigate video captions. But for during-playback comprehension, the synchronised presentation of caption text with audio produces stronger dual-coding effects than the separate-document presentation of a transcript. The learner with synchronised captions can maintain visual attention on the speaker (for non-verbal cues, expression, and demonstration) while processing both audio and caption channels. The learner with a separate transcript must switch visual attention between the video and the document, which breaks the dual-channel simultaneous-processing mechanism and adds the cognitive cost of tracking position in two parallel content streams. The practical implication: produce both captions (for during-playback learning) and transcripts (for post-session reference and accessibility), but do not substitute one for the other in your caption compliance programme.

Is there a proficiency threshold for ESL learners below which captions provide no additional benefit?

No threshold has been identified in the research at the professional proficiency levels typical of enterprise employees. Studies of advanced-proficiency L2 learners — equivalent to CEFR C1 level or IELTS 7+ — consistently find positive comprehension and vocabulary-retention effects from captions, though the magnitude is smaller than at lower proficiency levels. The only population where caption benefits approach zero is native-speaker-equivalent proficiency (bilingual or heritage-speaker populations), which is not the relevant comparison in most enterprise L&D contexts. For the practical purpose of programme design, the correct assumption is that every non-native-English learner in your programme benefits from accurate captions, at every proficiency level you are likely to encounter.

How do I cite the research to a leadership audience that doesn’t have time to read the primary literature?

Three citations carry the most organisational credibility in this context. First, the Mayer multimedia learning research programme, summarised in “Multimedia Learning” (Cambridge University Press, third edition); this is the most-cited educational-technology research in the enterprise training literature. Second, the 3Play Media “Captions and Learning” white paper, which aggregates the applied research in a practitioner-friendly format and is widely cited in L&D trade publications. Third, your own LMS data: completion-rate gaps between captioned and uncaptioned content on mobile, and assessment-score differences between content with glossary-accurate captions and content with general auto-captions, are first-party evidence that requires no external citations. The combination of an established theoretical framework (Mayer), a practitioner-facing summary (3Play white paper), and your own analytics data is the business-case package that translates most effectively for a leadership audience without research backgrounds.

Does the research apply to microlearning as well as long-form content?

Yes, but the magnitude of the effect is smaller for short content and the mechanism differs. For microlearning modules of 3–5 minutes, the cognitive-load benefit of captions is present but small: the duration is short enough that the total extraneous cognitive load is manageable without caption assistance. The retrieval-navigation mechanism is essentially absent: there is not enough content to warrant segment-level navigation, and the learner is more likely to replay the whole module than to navigate to a specific passage. The ESL benefit persists at all content lengths, as does the mobile-audio-absent benefit. The business case for captioning microlearning is therefore stronger on ESL-population size and mobile-completion patterns than on cognitive-load reduction. The compliance case applies equally regardless of content length — WCAG 2.1 AA has no minimum-duration exemption.

How should we handle multilingual programmes where the primary content language is not English?

The research findings apply in the same structural form to any first-language/second-language pairing. If your training content is in French and a portion of your learner population speaks French as a second language, captions in French provide the same ESL comprehension benefits as English captions for English-L2 learners. The additional complexity in multilingual programmes is the glossary architecture: domain vocabulary (product names, technical acronyms, regulatory standards) must be captured in every language the training is delivered in, and the glossary-accurate captioning system must be configured for each language pair. See the multilingual caption workflow guide for global L&D teams for the implementation detail, including translation pipeline, LMS delivery by platform, and EAA/AODA compliance considerations for multilingual content.

What is the correct unit of analysis for measuring the learning-effectiveness return on captioning investment?

Three units of analysis are defensible and serve different audiences. The per-learner completion rate is the most immediately available from LMS data: for a given piece of content, compare completion rates between captioned and uncaptioned versions, or between versions with glossary-accurate and general auto-captions, segmented by device type. This metric translates directly to “how much of the training is actually being completed” — a language finance audiences understand without learning-theory background. The assessment score on technical vocabulary items is the most direct learning-outcome measure: track quiz performance on questions targeting domain-specific terms across content types. The time-to-competency measure is the most powerful for organisations that track competency development over time: if learners who receive glossary-accurate captioned training reach competency benchmarks faster than comparable learners without accurate captions, the ROI case closes without any compliance reference at all. See reporting caption programme effectiveness to leadership for how to structure the measurement programme and present results in a format appropriate for a C-suite audience.