Captioning customer education and partner training academies: ADA Title III, Skilljar, Thought Industries, LearnUpon, and the public-academy compliance difference

Why customer education is a different compliance obligation than internal training

The Americans with Disabilities Act has three main titles that create accessibility obligations for video and training content. Understanding which title applies to which content is the starting point for building a caption compliance programme that actually covers your organisation's full exposure.

ADA Title I covers employment. Title I prohibits discrimination against qualified individuals with disabilities in job application procedures, hiring, firing, compensation, promotion, training, and all other terms and conditions of employment. The captioning obligation under Title I arises when an employer produces training video that employees with hearing disabilities must access as part of their job. The obligation is employer-to-employee. The employer must provide reasonable accommodations — which, for training video, means accurate synchronized captions at the WCAG 2.1 AA standard, or an equivalent accommodation. Coverage: employees of private employers with 15 or more employees.

ADA Title II covers state and local government entities, including public universities. If your organisation is a state agency, a county government, a public school district, or a public university, Title II applies to your training video library, your online course catalogue, and your instructional materials. The Title II deadline for web content became enforceable in April 2026 for entities with budgets over $5 million. Coverage: state and local governments and all their programmes, services, and activities.

ADA Title III covers places of public accommodation — private businesses that offer goods or services to the public. Title III prohibits discrimination on the basis of disability in the full and equal enjoyment of goods, services, facilities, privileges, advantages, and accommodations of any place of public accommodation. Congress enumerated twelve categories of public accommodation in the statute (hotels, restaurants, retail stores, service establishments, places of public display or collection, places of recreation, private schools, and others). The courts and the DOJ have, in a series of rulings since 2018, applied Title III to digital spaces including websites, mobile applications, and online services.

Customer education academies are places of public accommodation under Title III. The critical distinction is the audience: a customer academy is not exclusively for your employees. Any person with a disability who accesses your academy — whether they are a paying customer, a prospective customer, a community member who found the academy through a Google search, or a partner organisation employee — is protected by Title III's public accommodation obligation. The compliance driver is not employment; it is the act of opening a service to the public.

This creates a structural gap in many organisations' caption compliance programmes. The L&D team may have a mature, well-documented captioning workflow for internal training video — an accessibility coordinator, a QA gate, a captioning vendor with a glossary-biased model, and a compliance archive. But the customer education team, sitting in the Customer Success or Product organisation, may have no captioning workflow at all. Their content is published to Skilljar or Thought Industries, captioned with whatever the video hosting platform provides by default (usually AI-generated captions at 75–88% accuracy, well below the WCAG 2.1 AA 99% threshold), and treated as compliant because it has captions. That gap — between the internal L&D compliance programme and the external academy — is the compliance exposure most legal and compliance reviews of SaaS companies miss when they review accessibility.

The practical risk profile also differs between Title I, II, and III. Title I compliance failures typically generate EEOC complaints from individual employees, which are processed through administrative channels before reaching litigation. Title II failures generate OCR complaints to the Department of Education or DOJ, also administrative. Title III failures generate direct civil litigation under 42 U.S.C. § 12182 — private individuals can file in federal court without going through an administrative process first. This is why the wave of website accessibility litigation in the United States is overwhelmingly Title III litigation: plaintiffs' firms file complaints directly in federal court against businesses whose websites or digital services fail to meet WCAG 2.1 AA standards. If your customer academy has video content that fails the WCAG 2.1 AA caption accuracy standard, you are exposed to Title III civil litigation on the same basis as any other digital accessibility failure. The fact that the non-compliant content is inside a training academy platform rather than on your main marketing website does not change the exposure — both are places of public accommodation under Title III.

For the detailed compliance matrix across Title I, Title II, Section 508, state accessibility laws, and WCAG 2.1 AA, see the US caption compliance matrix.

ADA Title III and the public accommodation standard for digital academies

The ADA does not contain a specific section on websites. The statute lists twelve physical categories of public accommodation — written in 1990 before the commercial internet existed. The courts have developed two competing lines of doctrine on whether and how Title III applies to digital spaces:

The physical nexus doctrine, adopted by some circuits, holds that a website must have a sufficient nexus to a physical place of accommodation to be covered by Title III. Under this view, a website operated by a retail chain with physical stores is covered because the website is tied to the physical store experience; a website with no physical counterpart might not be.

The standalone digital service doctrine, adopted by the Ninth Circuit in Robles v. Domino's Pizza (2019), holds that Title III covers any place of public accommodation regardless of whether it has a physical location. Digital services are covered as standalone places of public accommodation.

Regardless of which circuit your business is subject to, the DOJ published guidance in March 2022 confirming that the department interprets Title III to require accessible websites and that WCAG 2.1 AA is the applicable standard. The DOJ supplemented this with a final rule on web accessibility for Title II entities in April 2024 and signalled that equivalent guidance for Title III entities is forthcoming. The litigation trend strongly favours the standalone digital service doctrine: by 2024, federal courts were processing thousands of Title III digital accessibility complaints per year, the majority targeting website content (including video) that fails to meet WCAG 2.1 AA.

For customer education academies, the most directly relevant standard is WCAG 2.1 Success Criterion 1.2.2, which requires synchronised captions for all pre-recorded audio in multimedia content. The failure mode under SC 1.2.2 is any video that has no captions, captions with accuracy below 99% as measured by the DCMP Captioning Key protocol, or captions that do not synchronise accurately with the audio. The specific error types that fail the standard: substituted words (ASR outputs a different word than was spoken), omitted words, incorrectly inserted words, and speaker identification errors for multi-speaker content.

The 99% accuracy threshold is the standard that product walkthroughs, feature demonstrations, onboarding tutorials, and certification content must meet. It is not a high standard relative to human-level accuracy — a skilled captioner achieves 99%+ routinely. The challenge is that auto-captioning of customer education content hits this threshold particularly rarely because customer education content has the highest domain-specific vocabulary density of any training category. Every screen recording of your product contains proper nouns — product names, button labels, modal titles, API method names, error message text — that auto-captioning systems have never seen in their training data. The result is that auto-captions on a product walkthrough video will produce approximately 80–88% accuracy on the domain vocabulary, well below the WCAG 2.1 AA threshold, while sounding plausibly correct to a hearing reviewer who knows the product.

The accuracy failure is invisible in normal operation. A hearing viewer watching a product walkthrough with 85% caption accuracy will not notice — the errors are in product names and UI labels that the hearing viewer already knows. A Deaf or hard-of-hearing customer watching the same video is entirely dependent on the caption track for content. When the caption reads "click the Manage Users icon" but the audio says "click the Manage Licenses icon," the customer receives wrong information from a training video that your organisation published as a compliant learning resource. That is the compliance failure — not an abstract WCAG gap, but a real information failure for a real learner.

For a detailed treatment of WCAG accuracy measurement methodology, see the post on what 99% caption accuracy means and the caption QA methodology guide.

Authenticated academies and the "open to any registered user" rule

Many customer education teams, when they first encounter the Title III compliance argument, raise an objection: "Our academy requires login. It's not public. It's only accessible to our customers." This is a reasonable instinct, but it misunderstands the Title III standard in a way that creates genuine compliance risk.

The Title III public accommodation obligation does not require that a place be accessible without any registration or authentication. The relevant question is whether a member of the public can access the service by going through a normal registration or onboarding process without needing to prove a pre-existing employment relationship. If any person — regardless of whether they are a paying customer on day one — can create an account and access your training academy, your academy is a place of public accommodation under Title III.

The DOJ March 2022 guidance explicitly addresses this: "Covered entities must ensure that people with disabilities have equal access to goods and services even when those goods or services are provided through digital systems that require users to create accounts, register, or log in." The guidance does not create a "registration wall" exception to Title III. A website that requires login to access video content is not exempt from the video caption requirement. The authentication requirement addresses who can access the content, not whether the content must be accessible to those who do access it.

The practical test is straightforward: can a person with a hearing disability create an account in your customer academy by following the standard account creation process, and if so, will they encounter video content with synchronised accurate captions? If the answer to the second question is no, you have a Title III compliance gap regardless of the authentication requirement.

There are three common authentication structures in customer education, and each has a different risk profile:

Structure 1: Open self-registration. Any person can create an account by entering an email address and password, with no approval workflow. This is the most common structure for product-led growth companies — the academy is a marketing and onboarding tool as much as a training resource. Under this structure, Title III clearly applies. Any person can register, any person can access the video content, and any person with a hearing disability is entitled to accessible captions.

Structure 2: Customer-gated registration. Account creation requires a license key, a company email domain match, or an invitation from an existing customer. The academy is accessible only to verified customers. Under this structure, Title III still applies, but the risk profile is narrowed: the potential plaintiffs are your existing customers with hearing disabilities, not the general public. This is a smaller but still real population, and the DOJ guidance covers it explicitly. The Title III obligation runs to every customer who accesses your service.

Structure 3: Fully private partner portals. Access requires a manual approval process, a signed NDA, or a formal partner agreement. The academy is not open to any member of the public — only organisations with a formal commercial relationship can access it. Under this structure, the Title III argument is weaker (the physical nexus to a "place of public accommodation" is more attenuated), but the Title I argument may apply if the partner employees who access the training are effectively receiving mandatory training that affects their business relationship with you. The safest compliance posture is to caption partner training video regardless of the legal classification argument, because the vocabulary accuracy challenge and the business case for accuracy (accurate captions = better-trained partners who correctly represent your product) are identical regardless of the legal basis.

The compliance posture recommended for most SaaS customer education teams is to treat the entire customer academy as a Title III covered service and apply the WCAG 2.1 AA caption standard across all video content. The cost of treating a Structure 2 academy as Title III covered when it may technically be narrower is low — it means captioning video accurately. The cost of treating a Structure 1 academy as exempt from Title III when it is in fact fully public is material — it means litigation exposure for the entire uncaptioned or inaccurately captioned video library.

Platform-by-platform caption workflows: Skilljar, Thought Industries, LearnUpon, Gainsight CE, Docebo

The five most commonly used customer education platforms have fundamentally different caption architectures. Understanding where captions live in each platform's stack is essential to building a compliant workflow.

Skilljar

Skilljar is the dominant purpose-built customer education LMS in the SaaS market. It is not primarily a video hosting platform — it is a course-and-curriculum delivery platform. Video in Skilljar is embedded from external video hosting providers: Wistia, YouTube, Brightcove, Vimeo, and Kaltura are the most common. Skilljar does not store video files natively and does not have its own video caption workflow.

The caption architecture in a Skilljar academy is: video file → external video host (Wistia/YouTube/Brightcove) → caption track added at the video host level → Skilljar embeds the video player from the host → captions delivered through the host's player controls. This means:

• If you use Wistia as your video host, captions must be managed in the Wistia media library. Wistia's AI captions are not WCAG 2.1 AA compliant — they typically reach 78–86% accuracy on product walkthrough content. You need to upload a corrected SRT or VTT file to replace the AI-generated track.
• If you use YouTube as your video host, YouTube's auto-captions are not WCAG 2.1 AA compliant for the same reason. Even if you embed a private or unlisted YouTube video in Skilljar, the YouTube auto-caption track does not meet the accuracy standard. You must upload a corrected caption track to the YouTube video and verify that the track is set to the correct language and labelled as "captions" not "subtitles."
• If you use Brightcove, Brightcove has its own caption management interface. Brightcove supports text track uploads (WebVTT format) and has a text track API. You must add the corrected caption track at the Brightcove media asset level.
• If you use Kaltura, Kaltura has a machine caption workflow with a correction interface. Kaltura machine captions reach approximately 80–88% accuracy on product walkthrough content and must be reviewed and corrected before the track meets the WCAG 2.1 AA standard. Kaltura's notification system can alert the caption workflow owner when new content is published, which is useful for building a triggered workflow.

The Skilljar platform itself does not have a "caption compliance" dashboard or an inventory of uncaptioned content across the academy. Compliance tracking requires working at the video host level: pulling a report of all video assets from Wistia, Brightcove, or YouTube and checking caption status asset by asset. If you have a large Skilljar academy with content from multiple video hosts, the compliance audit spans multiple platforms simultaneously.

For the Skilljar-specific caption setup walkthrough, see the Skilljar captions guide.

Thought Industries (Intellum/Eurekos)

Thought Industries is a white-label enterprise customer education platform with more flexible content architecture than Skilljar. Thought Industries supports native video hosting (direct upload to Thought Industries storage), external video embedding (same as Skilljar — Wistia, YouTube, Brightcove, Vimeo), and SCORM package delivery.

The caption workflow in Thought Industries depends on which video delivery method you use:

Native video hosting: Thought Industries stores video files on its own CDN. Caption tracks can be added to natively hosted videos through the Thought Industries content editor. The supported format is WebVTT. Thought Industries does not generate captions automatically — the caption file must be created externally and uploaded to the content object. The typical workflow for a Thought Industries customer with native video hosting is: record and edit video → submit to captioning vendor → receive corrected VTT file → upload to Thought Industries video content object → verify track in preview mode. This workflow is the most straightforward of any customer education platform because the caption track lives directly on the content object and is managed entirely within Thought Industries.

Embedded video from external host: Same architecture as Skilljar — captions are managed at the video host level, not the Thought Industries level. The Thought Industries embed carries whatever player and track configuration the external host provides. If your Thought Industries academy embeds from Wistia with AI-generated captions, the Wistia AI captions are what your learners see.

SCORM packages: Captions embedded within SCORM packages travel with the package and are not affected by the Thought Industries caption workflow. If your instructional design team produces SCORM-packaged course content with embedded HTML5 video and inline VTT caption tracks, those captions are inside the SCORM ZIP and are not visible to Thought Industries' content management layer. This is the best-practice architecture for Thought Industries customers with complex course content: SCORM packages with embedded captions, reviewed and accuracy-checked before the package is published to the Thought Industries academy. The SCORM-embedded caption approach also survives platform migrations — if you move from Thought Industries to a different customer education platform, the captions travel with the SCORM package. For the SCORM caption architecture in detail, see the SRT/VTT/TTML/STL format guide and the LMS migration caption checklist.

Thought Industries' white-label architecture means that the platform can be configured significantly by the customer. Some Thought Industries deployments have custom player integrations that bypass the platform's default caption controls. If your Thought Industries configuration uses a custom video player (through the platform's extensibility features), the caption track must be tested in that specific player configuration — the standard VTT upload workflow may not produce visible captions in a custom player without additional configuration.

LearnUpon (external portals)

LearnUpon occupies an unusual position in the customer education platform landscape: it was originally designed for internal corporate training (Title I territory) but has extensive external portal functionality that is widely used for customer and partner training (Title III territory). Many organisations run both internal and external training on the same LearnUpon instance, which creates the compliance classification problem described above — the same platform is serving both Title I and Title III obligations simultaneously, and the compliance documentation needs to distinguish between them.

LearnUpon's caption architecture for external portals is functionally identical to its internal LMS caption workflow. Video content in LearnUpon is typically delivered through one of three paths: SCORM packages with embedded captions, YouTube embeds, or Vimeo embeds. LearnUpon does not have native video hosting with caption management — it relies on the embedded player to deliver caption tracks.

The external portal feature in LearnUpon creates a separate learner-facing URL (typically a subdomain or custom domain) with independent branding. External portal learners create their own accounts and do not interact with the internal user management system. From a compliance perspective, the external portal is the place of public accommodation — internal learners (employees) are covered by Title I regardless of which portal they access, but external learners (customers, partners, community members) are covered by Title III when they access the external portal.

The compliance implication is that the caption audit for an organisation using LearnUpon for both internal and external training needs to cover both portals separately. The internal portal's caption library supports the Title I compliance argument (reasonable accommodation for employees with disabilities). The external portal's caption library supports the Title III compliance argument (equal access for all customers). An audit that covers only the internal portal, or that treats all content as equivalent regardless of which portal delivers it, will miss the external-portal Title III exposure.

For the full LearnUpon caption workflow including the SCORM and embed paths, see the LearnUpon captions guide.

Gainsight CE and in-product customer education

Gainsight Customer Education (formerly Northpass, formerly Gainsight PX's learning module) represents a distinct category of customer education delivery: in-product training that is embedded directly in the application UI rather than delivered through a separate academy domain. This architecture has a different WCAG compliance profile than a standalone academy.

In-product training delivered through Gainsight CE typically consists of short video segments (1–5 minutes), contextual tooltips, and guided product walkthroughs. The video segments are served through an embedded player within the product UI. The product UI itself must meet WCAG 2.1 AA (it is a place of public accommodation), and the video content within the product UI is subject to the same SC 1.2.2 caption requirement as video in any external academy.

The caption architecture in Gainsight CE depends on the video hosting configuration. Most Gainsight CE deployments use Wistia or a similar video CDN for the video files embedded in the in-product player. Captions must be managed at the Wistia level (or equivalent) and must meet the same 99% accuracy standard. The in-product context makes caption accuracy failures particularly costly: a learner following an in-product tutorial is using the product at the same time as watching the tutorial, and a wrong caption (for example, a menu item name that is captioned incorrectly) can cause the learner to click the wrong UI element and fail to complete the onboarding flow. The accuracy failure has a direct impact on the customer success metric the training is designed to support.

Gainsight CE does not have a caption compliance dashboard. The platform surfaces product engagement metrics (completion rates, interaction rates, time-in-module) but does not track caption accuracy or completion rates specifically from Deaf or hard-of-hearing users. Building a compliance audit for Gainsight CE content requires pulling the video asset list from the underlying video host (Wistia, Brightcove) and auditing caption accuracy by asset.

Docebo (external portals and content marketplace)

Docebo is primarily an internal LMS, but it has well-developed external portal functionality that many enterprise SaaS companies use for customer and partner training. Docebo's external portal configuration ("Docebo Shape" in some versions) allows separate learner registration, custom branding, and independent content libraries for external audiences.

Docebo's caption architecture for external portals uses the same Central Repository model as the internal LMS. Video assets stored in the Docebo Central Repository (Docebo's native video management layer) can have caption tracks attached at the repository level, and those tracks are served across all portals — both internal and external — that reference the same asset. This means a well-configured Docebo deployment can manage caption compliance for internal and external content from a single asset library, with a single captioning workflow feeding both audiences.

The compliance tracking gap in Docebo's external portal is the same as in the internal LMS: caption accuracy is not tracked at the platform level. Docebo records whether a caption track exists, but it does not measure caption accuracy. The compliance audit must be conducted by spot-checking caption files against the DCMP protocol, not by relying on Docebo's content management reports. For the full Docebo caption management workflow, see the LMS caption audit methodology and the accessibility coordinator playbook.

Partner training legal status: employee, customer, or public accommodation?

Partner training academies — used to train resellers, implementation partners, technology integration partners, and referral partners — create a compliance classification problem that is more complex than either internal employee training (Title I) or public customer education (Title III).

The classification question is: who, legally, are the people accessing your partner training academy?

They are not your employees. Partners are independent businesses. Their employees who access your training academy are not your employees — they have an employment relationship with the partner organisation, not with you. ADA Title I covers the employment relationship between an employer and its employees. Your partner organisation's employees are not your employees under Title I, and you do not owe them a reasonable accommodation as an employer.

They may or may not be your customers. If the partner has signed a reseller agreement that includes access to your partner training academy as a commercial term, the partner employees who access the academy are accessing it as representatives of a customer organisation (the partner business). Title III covers individuals accessing a place of public accommodation — the question is whether the partner training academy meets the "open to any registered user" test described in the previous section.

They are almost certainly covered by Title III if the academy is accessible to any partner who applies. Most partner programmes have a standard application and onboarding process: a partner fills out an application form, completes a brief qualification step, and gains access to the partner training academy. This is functionally equivalent to open self-registration from a Title III coverage perspective. Any SaaS company that wants to maintain a partner channel is, by that business decision, opening its partner training academy to any organisation that meets the partnership criteria — which is typically not a restrictive set of criteria. The DOJ March 2022 guidance's "open to any registered user" framing covers this structure.

The compliance risk in partner training is often higher than in customer training for two reasons:

First, partner training content typically includes certification programmes with assessment components. Partners who fail to complete the certification correctly may lose partner status or not qualify for certain tiers of your partner programme. If a partner employee with a hearing disability cannot access the training video accurately (because captions are inaccurate), and as a result fails the certification assessment, the failure has direct financial consequences for the partner organisation. This creates a concrete harm that is much easier to link to the non-compliant caption than a generalised "equal access to training" argument.

Second, partner training content tends to have the highest technical vocabulary density of any category in your external training library. Partners need to understand your product at a level of technical depth that exceeds what most customers need. API documentation, SDK integration guides, configuration tutorials, and architecture walkthroughs all use technical vocabulary (method names, parameter names, error codes, protocol names) that auto-captioning systems handle poorly. A partner watching a Salesforce integration tutorial with inaccurate captions for API endpoint names has a genuinely degraded learning experience that affects their ability to implement the integration correctly.

The practical recommendation is to treat partner training academy content as subject to the Title III standard regardless of the technical legal argument about whether the specific partner programme structure creates full Title III coverage. The vocabulary accuracy requirements are identical, the QA process is identical, and the business case for accurate captions in partner training (better-trained partners who can implement your product correctly) is at least as strong as the legal argument. A partner training captioning programme that applies the WCAG 2.1 AA accuracy standard to all content is both legally defensible and commercially well-motivated.

For the full compliance framework comparison between Title I, Title II, Title III, and Section 508, see the caption compliance matrix.

Vocabulary and accuracy challenges for customer education content

Customer education content — product walkthroughs, feature tutorials, onboarding sequences, certification modules, partner enablement content — has the highest domain-specific vocabulary density of any training category. This creates an accuracy challenge that is structurally different from the challenge in internal L&D content, healthcare training, or compliance training.

Categories of vocabulary in customer education content

Product and feature names. Every product has a proper-noun vocabulary that is unique to that product: feature names, module names, workflow names, navigation labels, and product tier names. For a typical SaaS product with 3–5 major product areas and 15–30 sub-features, this vocabulary runs to 50–150 terms that auto-captioning systems have never seen. A navigation item called "Outcome Tracking" becomes "outcome tracking" or "out-come tracking" in auto-captions. A feature called "Milestone Builder" becomes "milestone builder" or "miles tone builder." None of these errors are individually catastrophic, but in aggregate they produce a learner experience where the caption track and the on-screen UI labels do not match — which is confusing for any learner and a compliance failure for learners who rely on captions.

UI labels and modal text. Screen recordings of product walkthroughs contain on-screen text that the narrator reads aloud. The narrator's voice says "click the Manage Billing dropdown" while the screen shows a dropdown labelled "Manage Billing." The caption track, to be useful, must produce "Manage Billing" — not "manage billing" (incorrect capitalisation of a proper UI element) or "manage billing dropdown" (omitting the product-specific label). Auto-captioning systems produce consistent casing failures on UI labels because they apply standard sentence-case normalisation to transcribed text, stripping the title-case that UI labels use.

API and SDK terminology. Partner training content and developer-focused customer education content is extremely dense with technical identifiers: API endpoint paths (/api/v2/users/{userId}/roles), SDK method names (client.createSession(options)), configuration parameter names (enableAutoRenew, maxRetryCount), HTTP status codes (404, 422, 503), and response body keys. These identifiers are completely outside the vocabulary of any general-purpose ASR model. Auto-captioning of developer documentation screen recordings typically produces 65–80% accuracy on the identifier vocabulary — well below the general content baseline.

Version identifiers. Customer education content references version numbers that change with every product release: Salesforce Winter '26, Workday 2026R1, HubSpot CRM 2.0, API v3.2. Release-specific vocabulary is a moving target: the version numbers in your Q1 customer education content are different from the version numbers in your Q3 content. A glossary model built from Q1 content will produce errors on Q3 version identifiers unless it is updated with each major release.

Pricing tier and plan names. Product pricing tiers appear in onboarding and customer education content more frequently than in internal L&D content. "The Enterprise plan includes…", "if you're on the Growth tier…", "Pro users can access…" — all of these require correct rendering of proprietary plan names. A pricing tier called "Business Plus" becomes "business plus" or "business plus" in auto-captions, stripping the capitalisation that distinguishes the plan name from the descriptive phrase.

Competitor product names referenced in context. Some customer education content — particularly migration tutorials, integration guides, and competitive positioning content — references competitor product names. "If you're migrating from HubSpot…", "unlike Salesforce's approach…", "this replaces what you'd build with Zapier." Competitor product names are in ASR training data (unlike your own product's obscure feature names), so they are generally captioned correctly. But the combination of your product vocabulary with competitor product vocabulary in the same content creates a vocabulary-density profile that degrades overall accuracy more than either category alone.

Building a customer education glossary

The solution to customer education vocabulary accuracy is a glossary model seeded specifically from your product documentation. This is not the same glossary you build for internal L&D training. An internal L&D glossary covers your company's internal jargon, HR vocabulary, and department-specific terms. A customer education glossary covers your product's public-facing vocabulary: the terms that appear in your help documentation, your API reference, your feature changelog, and your support knowledge base.

The most efficient seed source for a customer education glossary is your existing documentation:

• Help centre / knowledge base — your help articles are a dense source of product vocabulary. Every article title, every UI label mentioned in a step ("click the Settings gear icon"), every feature name in a procedure is glossary-appropriate. A systematically processed help centre of 200–500 articles will yield 300–800 unique product terms.
• API reference documentation — if your product has a public API reference (whether in Swagger/OpenAPI format, Postman, or Stoplight), extract all endpoint paths, method names, and parameter names. These are high-priority glossary terms for developer-facing customer education content.
• Release notes / changelog — release notes name new features, deprecated features, and changed UI labels. A processed release notes archive of 12–18 months will identify terms that are in current use and should be in the active glossary, as well as terms that have been renamed or deprecated and should be removed or aliased.
• Pricing and plan documentation — your pricing page is a compact source of plan names, feature entitlement labels, and tier-specific vocabulary. Plan names and tier names should be in the glossary with their canonical capitalisation.

The customer education glossary needs a cadence update trigger tied to product releases. When a major product release changes UI labels, adds new features, or renames existing capabilities, the glossary model must be updated before the customer education content for that release is published. A 48-hour SLA between a major release going live and the corresponding glossary update propagating to the captioning model is a reasonable target for most teams. For the full glossary architecture and update workflow, see the customer glossary architecture post.

The accuracy improvement from a well-maintained customer education glossary is consistent and measurable. On a typical SaaS product walkthrough with 50–80 domain-specific terms in a 10-minute video, a 300-term product glossary applied to Whisper-large produces a 8–14 percentage point accuracy improvement on the domain vocabulary compared to default-model output. A well-seeded 800-term glossary covering the full product vocabulary can push customer education content accuracy from the 80–88% auto-caption baseline to 96–99% on product-vocabulary content, with the remaining gap coming from audio quality issues (home-office recording, background noise, speaker accent variation) rather than vocabulary recognition failures. For the accuracy benchmarks by content type, see the Whisper accuracy benchmarks post.

Caption production workflow for customer education teams

The structural challenge with customer education captioning is that customer education content is not produced by L&D teams. It is produced by Customer Success, Customer Education, Product Marketing, Technical Writing, and Developer Relations teams — none of which have a training operations infrastructure that includes a captioning workflow. The contrast with internal L&D is significant.

An internal L&D team producing training content has typically established:

• A captioning vendor with a signed SOW and a defined accuracy SLA
• A submission workflow (who submits content to the vendor, in what format, by what deadline)
• A QA gate (who checks the returned caption track before publication, what accuracy check is run)
• A compliance archive (where caption files and accuracy records are stored for audit purposes)
• An accessibility coordinator who owns the workflow and reports on compliance metrics

A customer education team producing content for a Skilljar academy has typically established:

• A recording workflow (Loom, Zoom, Camtasia, or screen recording software)
• An editing workflow (video editor, usually not professional post-production)
• An upload workflow (upload to Wistia or YouTube, embed in Skilljar)
• No captioning workflow at all, or "enable auto-captions on Wistia/YouTube" as the entire caption programme

The gap between these two workflow states is large. Adding a captioning programme to a customer education team that has never had one requires building the entire infrastructure from scratch — not adapting an existing L&D caption workflow to a new platform. That infrastructure gap is why the eight failure modes in the next section are so common: they are not the result of poor execution of an existing workflow, but of the complete absence of a workflow where one is legally required.

The production workflow for customer education captioning needs to account for several characteristics of customer education content production that differ from internal L&D:

Volume and cadence. Customer education content is typically produced in higher volume and on a more frequent cadence than internal training. Every product release generates new tutorial content. Every new feature generates onboarding content. Every major customer segment generates a dedicated use-case walkthrough. A SaaS company with two major releases per year, three minor releases per quarter, and a 200-article help centre translating to video might produce 50–100 new video assets per quarter for its customer academy. This volume requires a scalable, low-friction submission workflow — a captioning programme that requires a manual email to a vendor for each submission will not scale to 50-100 submissions per quarter.

Screen recording quality. Customer education content is almost always screen-recorded, usually with a voice-over recorded through a computer microphone. The audio quality on screen recordings is typically worse than controlled-studio internal training video. The common issues: laptop microphone (omnidirectional, high ambient noise pickup), home-office or open-office recording environment (room reverb, HVAC noise, keyboard noise), no audio processing (no noise reduction, no normalisation, no compression). The accuracy impact is a 5–10 percentage point degradation on top of the vocabulary challenge described above. A screen recording with home-office audio and dense product vocabulary may start at 72–80% auto-caption accuracy before any glossary correction, versus 85–92% for studio-recorded internal training with a standard professional vocabulary.

Update frequency. Customer education content is updated more frequently than internal training. When a UI label changes in a product release, all video content that shows and narrates that UI label needs either recaptioning or a caption correction. An internal compliance training video from two years ago is still compliant; an onboarding tutorial from two years ago that shows an interface that no longer exists is both outdated content and potentially a source of caption accuracy failures for the new UI vocabulary. The recaption trigger for customer education content is not a compliance schedule — it is a product release event.

Multi-author production. Customer education content is produced by a distributed team — Customer Success Managers recording customer-facing tutorials, Solutions Engineers recording technical integration guides, Technical Writers producing documentation-to-video content, Developer Relations recording API walkthroughs. Each author has a different recording environment, a different audio quality profile, and a different vocabulary accuracy baseline. A captioning programme for a customer education library needs to handle content from 10–30 different authors, each with their own acoustic profile, rather than from a controlled set of professional narrators.

The workflow architecture that addresses these characteristics is an event-triggered submission model: when a new video is uploaded to the video hosting platform (Wistia, Brightcove, YouTube), an automated trigger sends the video to the captioning vendor API, the captioned VTT file is returned and attached to the video asset, and a QA notification is sent to the designated caption reviewer. This workflow requires a captioning vendor with an API submission capability and a webhook or notification return path, and it requires the video hosting platform to support upload event triggers (Wistia, Brightcove, and Kaltura all support this natively or through Zapier/Make integration).

The alternative — a manual submission workflow where a team member submits videos to the captioning vendor on a periodic schedule — works for teams with low video production volume but breaks at the cadence typical for customer education teams. The key failure mode of the manual workflow is the compliance gap during the publication window: video is published to the academy with no caption track (or with auto-captions) and the manual submission happens days or weeks later, during which time the uncaptioned or inaccurately captioned video is live for customer access.

For the event-triggered publication model and how it prevents compliance drift, see the accessibility coordinator playbook section on the new-content gate.

Building a captioning programme from the customer education seat

Building a captioning programme for a customer education academy requires the same structural components as building a captioning programme for an internal L&D library — a captioning vendor, a submission workflow, a QA process, a compliance archive, and an ownership assignment — but the organisational context is different in ways that affect how the programme is designed.

Scope definition and ownership

The first decision is scope. A customer education captioning programme covers: all video content published in the customer academy (Skilljar, Thought Industries, LearnUpon external portal, Gainsight CE, Docebo external portal), all video content embedded in the company's public help centre, all video content in the partner training academy, and all video content on the company's YouTube channel that is linked from or embedded in any of the above. The scope does not cover internal L&D content (which is covered by the separate internal captioning programme, if one exists) or marketing videos on the company's homepage (which may have a separate accessibility review programme under the marketing team's web accessibility initiative).

Ownership of the customer education captioning programme should sit with the Customer Education Director or VP of Customer Success, not with the L&D team. The L&D team's ownership extends to the internal training library. The customer education team's ownership extends to the external academy. Both leaders need to know they own their respective captioning obligations — an ambiguous ownership model that assumes "the L&D team handles captioning" will result in the external academy being unreviewed when the internal library is audited.

In practice, the customer education team will not have caption expertise — that expertise is in the L&D team (if a captioning programme exists there) or in the captioning vendor. The recommended structure is for the customer education team to own the compliance obligation and the workflow management, while leveraging the same captioning vendor used by the internal L&D team. Using a shared vendor creates a shared glossary opportunity: the internal L&D glossary and the customer education glossary are both managed by the same vendor, are maintained on the same accuracy improvement cycle, and can share company-level vocabulary (company name, executive names, broad product vocabulary) while maintaining separate domain-specific layers (internal HR vocabulary vs. customer-facing product documentation vocabulary).

Inventory and audit baseline

Before building the ongoing workflow, the customer education team needs an inventory of the existing video library and its caption status. The inventory should cover:

• All video assets in each hosting platform (Wistia, Brightcove, YouTube, Kaltura) linked to or embedded in the customer academy
• For each asset: whether a caption track exists, the caption track format (SRT, VTT, auto-generated vs. manually uploaded), the track language, and the estimated or measured accuracy
• For assets with existing caption tracks: a sample accuracy spot-check using the DCMP Captioning Key protocol (minimum 5% of video duration, scored at the word level)
• The publication date of each asset (older content should be prioritised for remediation if the UI labels it references have changed in subsequent product releases)
• The platform delivery path (which academy or portal each asset appears in, which learner population accesses it)

The inventory output is a compliance baseline: the percentage of the customer education video library that meets the WCAG 2.1 AA caption accuracy standard. In most customer education teams that have not previously run a captioning programme, this baseline is 0–20% of the library — the majority of content is either uncaptioned or captioned with auto-generated tracks that fail the accuracy standard. This is not unusual; it is the starting state for most customer education libraries. The baseline is the starting point for the remediation plan, not a source of liability in itself (provided the plan is documented and executed against a defined schedule).

Vendor selection for customer education captioning

The vendor selection criteria for customer education captioning differ slightly from the internal L&D criteria. The specific requirements for customer education are:

API submission capability. The event-triggered submission model described in the previous section requires the captioning vendor to accept video submissions via API, not just through a web uploader or email. The API should support webhook callbacks when the captioned file is ready, to enable automated delivery of the corrected VTT file back to the video hosting platform. Not all captioning vendors have this capability.

Per-customer glossary with product-documentation seeding. The glossary model must be buildable from product documentation sources — help centre exports, API reference exports, changelog archives — rather than requiring manual term-by-term entry. A vendor that requires manual glossary entry cannot efficiently onboard a 500-term customer education vocabulary in the pre-launch window. For the glossary architecture that supports this use case, see the customer glossary architecture post.

Release-trigger glossary update workflow. The vendor must support a defined process for updating the glossary model when new product vocabulary is introduced in a release. This should be documented as a term in the vendor SLA or SOW, not left as an informal process. For the SLA requirements for captioning vendor contracts, see the captioning vendor RFP playbook.

Wistia, Brightcove, Kaltura, and YouTube integration. The vendor must have demonstrated integrations with the video hosting platforms used in your customer education stack. Manually uploading corrected VTT files to Wistia and Brightcove for every video is feasible for 20 videos per quarter but not for 80 per quarter. Automated delivery via platform API is required at scale.

For a vendor comparison across the major captioning providers, see the Rev vs GlossCap comparison and the 3Play vs GlossCap comparison.

QA gate for customer education content

The QA gate for customer education captioning should be designed as a publication gate: no new video content is published to the customer academy without a passed caption QA check. This is operationally different from the internal L&D compliance gate, which may allow content to be published before caption review (for time-critical compliance training) and then remediated. Customer education content does not typically have the same emergency-publish use case — a product tutorial can be held for 24–48 hours for caption QA without business impact.

The QA check for customer education content should focus specifically on product vocabulary: spot-check the first 3 minutes and one random 2-minute segment for product name, UI label, and API identifier accuracy. If the spot-check reveals substitution errors on product vocabulary (the glossary model is missing terms from the latest release), those terms should be added to the glossary model and the full video recaptioned before publication. The publication gate prevents the accuracy failure from going live; the glossary update prevents it from recurring for the next piece of content on the same release vocabulary.

For the spot-check protocol in detail, see the caption QA methodology post. For the compliance dashboard metrics that the customer education captioning programme should feed into, see the accessibility coordinator playbook.

Eight failure modes

These are the failure modes that leave customer education academies non-compliant despite the team's belief that the caption obligation is covered. They are documented here because they are specific to the customer education context and are not covered by the internal L&D captioning failure mode literature.

Failure mode 1: Wistia or YouTube AI captions treated as WCAG-compliant

Both Wistia and YouTube offer AI-generated auto-captions as a standard feature. Both label these captions in their platform UI as "captions," not as "draft captions requiring review" or "machine-generated captions not meeting WCAG 2.1 AA standards." Customer education teams who enable auto-captions on their video hosting platform often believe they have resolved their caption compliance obligation. They have not. Wistia AI captions and YouTube auto-captions routinely reach 80–88% accuracy on product walkthrough content — well below the 99% WCAG 2.1 AA threshold. The presence of a caption track is not compliance. The quality of the caption track determines compliance. The audit check is not "does this video have captions?" but "do these captions meet the accuracy standard?"

Failure mode 2: Skilljar academy treated as captioned because Wistia has captions enabled

This failure mode is an extension of failure mode 1. A team using Skilljar with Wistia as the video host enables Wistia AI captions, sees captions appearing on videos when previewed in Wistia, and concludes that the Skilljar academy is compliant. The captions are present but do not meet the accuracy standard. An additional complication: some Wistia embed configurations suppress the caption toggle in the embedded player even when a caption track exists in Wistia — meaning learners in the Skilljar academy cannot access the caption track at all because the player controls do not expose it. The compliance check must be run from the learner's perspective, within the Skilljar academy, not from the Wistia library management view.

Failure mode 3: Partner training academy classified as "private" and exempted from caption compliance

A partner training team tells legal or compliance that the partner academy is "private" (requires a signed partner agreement to access), and the legal team accepts this as a basis for exempting the academy from Title III. As described in the section on authenticated academies, the DOJ March 2022 guidance does not create a "signed agreement required" exception to Title III. Any academy that is open to any organisation that meets the partnership criteria — and most partner programmes are open to any organisation that fills out an application — is covered by Title III. The classification error leaves the partner training academy's video library unaudited and uncaptioned while the organisation believes it has legal cover.

Failure mode 4: LearnUpon external portal not covered by internal L&D caption audit

An organisation using LearnUpon for both internal training (Title I) and external customer/partner training (Title III) conducts a caption audit of the internal portal and documents compliance for the internal library. The external portal is not included in the audit scope because the audit was scoped to "internal training compliance." The external portal's video library — which is the Title III covered place of public accommodation — has never been audited. The result is a compliance documentation package that appears thorough (it covers all internal content) but misses the entire external-academy exposure. The audit scope must explicitly include the LearnUpon external portal as a separate audit target with a separate compliance record.

Failure mode 5: Product release creates new UI vocabulary, existing captions become inaccurate

A product release renames a feature: "User Management" becomes "Workforce Administration," "Settings" becomes "Configuration," a navigation menu item previously called "Reports" becomes "Insights." All existing customer education videos that reference the old UI labels now have caption tracks where the narration says "click Workforce Administration" but the caption reads "click User Management." The caption was accurate when it was published; it is inaccurate after the rename. This is not a captioning vendor failure — the caption file was correctly created from the original audio. It is a content lifecycle management failure: no process exists to identify which videos reference vocabulary that changed in the product release, flag them for recaptioning, and track the remediation to completion. The compliance gap is invisible until an auditor or a Deaf user encounters the discrepancy.

Failure mode 6: Customer education compliance documentation covers only internally-hosted content

An accessibility audit covers all video in the company's Docebo LMS (the internal platform). The audit report states that 94% of the internal training library is captioned to WCAG 2.1 AA standards. The customer academy is hosted on Skilljar with Wistia as the video host. No Wistia content appears in the Docebo caption audit. The accessibility report is technically accurate for what it audited, but the customer education library — the Title III covered content — has no compliance documentation. If the organisation is presented with a Title III complaint about the customer academy, the Docebo compliance audit provides no cover because it documents a different content library. The compliance documentation for the customer academy must cover the customer academy's content, not the internal LMS content.

Failure mode 7: New content published to the academy without a caption QA gate

A Customer Success Manager records a product walkthrough after a feature launch on Friday afternoon and publishes it directly to the Skilljar academy over the weekend to support the Monday customer email announcing the launch. No caption QA gate exists. The video goes live with Wistia AI captions at 81% accuracy on the new feature vocabulary (the feature was released last week and the glossary model does not yet have its terminology). Monday morning, the caption-first learner email goes out with a link to the tutorial. Every Deaf or hard-of-hearing customer who clicks the link encounters a video with inaccurate captions for the vocabulary they most need to understand — the new feature names. The business urgency of the launch override conflicted with the compliance gate that did not exist. Building the publication gate before the first urgent publish event, not after, is the only way to prevent this failure mode.

Failure mode 8: In-product tutorial captions not updated when product UI changes

An organisation uses Gainsight CE or a similar in-product education system to deliver contextual tutorials within the product UI. The tutorials contain screen recordings of the product at a specific version. A product redesign changes the UI layout and navigation structure. The in-product tutorials now show an old UI while the learner is looking at the new UI. The caption tracks were correct for the old UI; they reference UI labels and navigation paths that no longer exist. The learner following the tutorial cannot reconcile the caption ("click the left navigation item labelled Campaigns") with the current UI (which now has a different navigation structure). The in-product tutorial content needs to be treated as a product artefact with a release-gated update cycle, not as evergreen content. A content audit trigger keyed to major product releases — which tests all in-product tutorials against the current UI and flags caption accuracy failures from changed vocabulary — is the structural fix.

FAQ