Continu LMS captions: Google Workspace LMS, YouTube auto-caption accuracy failures, and ADA Title I compliance for SaaS company training

How Continu handles video: the Google Workspace hosting model

Why Continu does not host video on its own servers

Most LMS platforms in the enterprise market — Docebo, TalentLMS, Cornerstone OnDemand — provide their own video hosting infrastructure: you upload an MP4 file to the LMS, the platform transcodes it, streams it to learners, and optionally generates auto-captions or accepts a VTT/SRT sidecar file through an admin interface. Continu's architecture is different by design. Because Continu is purpose-built around Google Workspace — the same ecosystem that tech company employees already use for email, documents, presentations, and meetings — Continu assumes that video lives where Google-native video lives: YouTube or Google Drive.

This is not a limitation but a deliberate architectural choice. An L&D manager building a course in Continu uses Google Slides for the presentation layer, embeds YouTube videos (the company's own channel or unlisted uploads), pulls screen-recording content uploaded to Google Drive, and uses Google Meet for live virtual instructor-led training sessions. The entire content creation workflow stays inside Google Workspace. Continu aggregates and structures this Google-native content into a cohesive course experience without duplicating the underlying media.

The captioning consequence follows directly from this architecture. When a learner clicks play on a training video embedded in a Continu course, they are watching a YouTube embed or a Google Drive video. The caption track they see — if any — is the caption track on that YouTube video or Google Drive file. For YouTube, that means the auto-generated captions that YouTube's speech recognition creates automatically when a video is uploaded. Unless the course creator has gone back to the YouTube video's settings and uploaded a corrected VTT or SRT file to replace the auto-generated track, the auto-captions are what hearing-impaired learners receive.

YouTube auto-captions: what they are and what they are not

YouTube auto-captions are generated by Google's speech-to-text engine and are made available automatically on most uploaded videos within a few hours of upload. They serve as better-than-nothing captions for general-audience content in common spoken language. For general conversational speech in standard American or British English, YouTube auto-captions typically achieve 80-90% word accuracy — meaning roughly 1 in 10 to 1 in 5 words is incorrect.

This accuracy range is adequate for informal supplementary content where minor errors do not impair comprehension. It is not adequate for:

• Mandatory training where comprehension accuracy is legally required — ADA Title I accommodation requires that hearing-impaired employees receive training that enables them to understand and apply the content, not a degraded version of the training. A security awareness training module where "Crowdstrike EDR" becomes "crowdstrike ether" and "Zscaler Zero Trust" becomes "z scaler zero trust" fails the accommodation standard.
• WCAG 2.1 AA compliance — WCAG 2.1 Success Criterion 1.2.2 requires captions for all pre-recorded synchronized audio content. The WCAG definition of "captions" includes the accuracy requirement: captions must convey all spoken information, including technical terms, product names, and specialized vocabulary. The 80-90% accuracy of YouTube auto-captions on technical content falls well below the 99%+ accuracy threshold that WCAG 2.1 AA compliance requires for technical vocabulary.
• Technical vocabulary in SaaS company training — YouTube auto-captions are trained on large corpora of general speech. They do not have exposure to the dense specialized vocabulary of a cybersecurity company's internal training, a fintech company's product documentation walkthrough, or a developer-tools company's SDK onboarding video. The error rate for technical vocabulary is substantially higher than the 80-90% general accuracy figure.

The caption upload path for YouTube-hosted Continu video

The correct workflow for adding compliant captions to a Continu course that uses YouTube video is to upload a corrected caption file directly to the YouTube video and then optionally disable the auto-generated track so it does not compete with the corrected version. YouTube accepts WebVTT (.vtt) and SubRip (.srt) caption files through the YouTube Studio interface. When a corrected VTT or SRT file is uploaded to a YouTube video, it replaces what the learner sees when they click the CC button in the YouTube player — regardless of whether that player is being viewed on YouTube.com or embedded in a Continu course.

For Google Drive-hosted video in Continu, the workflow differs: Google Drive video playback supports caption sidecar files, but the upload path is different from YouTube Studio. In Google Drive, a VTT file named identically to the video file (e.g., onboarding-module-3.mp4 and onboarding-module-3.vtt) stored in the same Drive folder is associated with the video for caption playback. Alternatively, a Drive video can have a caption track added directly through the Drive video player interface. Verify the current caption upload path in Google Drive's documentation, as the interface has evolved.

Continu may also provide a direct caption upload path in the course builder for certain video content types — check Continu's current product documentation for the latest supported workflow, as platform features evolve.

Why YouTube auto-captions fail for tech company training content

The vocabulary gap between general STT and tech-company training

YouTube's speech-to-text engine is optimized for the vocabulary distribution of YouTube content overall: general speech, casual conversation, broadly popular topics. The vocabulary of a SaaS company's internal training is radically different from this distribution in two compounding ways.

First, Google's own product names — the vocabulary that appears in any training for Google Workspace users, Google Cloud customers, or companies using Google tools in their stack — are a consistent failure surface despite being Google's own property. YouTube's auto-captioning engine does not give preferential treatment to Google product names. Second, the organization's own proprietary vocabulary — its product names, API names, internal system names, project codenames, and service identifiers — has zero exposure in any general STT training corpus. Every proprietary term is an out-of-vocabulary problem for the model.

Google Workspace and Google Cloud vocabulary failures

For Continu's tech-company customer base, the following Google product name failures are the most commonly encountered in training video captions:

• Looker / Looker Studio — "Looker" is transcribed as "locker" (phonetically close in casual speech) or "looker" (correctly, but lowercased as a common adjective). "Looker Studio" becomes "locker studio" consistently. For a company training its sales team on Looker dashboards or its data team on Looker Studio reports, the primary product name is systematically wrong in every mention.
• BigQuery — transcribed as "big quarry" (the dominant failure mode), "big query" (correctly split), or "big quarry" with capitalization lost. A data engineering onboarding video that references BigQuery dozens of times will have the term wrong in a significant fraction of occurrences.
• Vertex AI — transcribed as "vortex AI" (rhyme-neighbor substitution) or "vertex a.i." (correctly split but with the period notation). Machine learning training content from companies using Google Cloud AI/ML infrastructure sees this failure on a high-frequency term.
• Apigee — transcribed as "apagy," "apadgy," or "a pee gee" depending on speaker cadence. API management training content built around Apigee loses the product name in most instances.
• Duet AI / Gemini — "Gemini" (Google's AI branding) transcribed as "geminy" or "jimmy" at faster speech rates. "Duet AI" is transcribed as "duet a.i." or "do it AI" with inflection-driven substitution.
• Pub/Sub — transcribed as "pub sub," "pubsub" (merged), or "pub sub" (with the slash dropped). For companies running event-driven architectures on Google Cloud, Pub/Sub appears in technical training constantly.
• Anthos — the name itself is relatively stable, but "Anthos" as Google's hybrid cloud platform is context-confused with common uses of the Greek prefix "anthos" (flower) and the Anthos name in other contexts.
• AlloyDB — transcribed as "alloy d.b." (correctly split) or "alloy dB" (dB as a decibel unit) in speech contexts where the speaker pauses between the components.
• Spanner — correctly transcribed but semantically confused with the common English noun. A training video saying "configure Spanner" is correctly transcribed but a caption reader may not immediately parse the capitalized Google Cloud database product from the tool noun.
• Chronicle — Google's SIEM (Security Information and Event Management) product is correctly transcribed as "chronicle" but contextually confused with the common English noun. For cybersecurity companies using Chronicle for their SOC, training video referencing "Chronicle alerts" or "Chronicle detection rules" loses the product specificity in lower-case auto-caption output.

SaaS company proprietary vocabulary: the primary failure surface

Beyond Google's own product names, the most significant vocabulary failure surface for Continu customers is each organization's proprietary vocabulary. Continu's ICP — SaaS and technology companies with 50 to 1,000 employees — are by definition companies whose internal training is saturated with proprietary terminology that has zero presence in any general STT training corpus:

• Product names and version strings — the company's own SaaS product name, sub-product names, module names, and version identifiers. A company named "Axiom" trains employees on "Axiom Analytics" and "Axiom Insights" — both names are homophones or near-homophones for common English words and are systematically misrecognized without a glossary.
• API endpoint and SDK function names — developer-tools training and engineering onboarding videos reference specific API endpoints (/v2/events/ingest), SDK function names (initializeClient(), sendEventBatch()), and configuration parameter names. These are entirely out-of-vocabulary for general STT.
• Internal system and project names — internal codenames for projects, infrastructure component names, data pipeline names, and organizational program names are proprietary vocabulary with no general-corpus exposure.
• Cybersecurity vocabulary (for security-sector Continu customers) — security companies training employees on their own security operations stack see compound failure: industry vocabulary (SIEM, SOAR, EDR, XDR, MITRE ATT&CK, CVE, IOC, TTPs) alongside vendor product names (CrowdStrike, Splunk, Okta, Zscaler, SentinelOne, Palo Alto Networks) alongside proprietary internal tool names. Each layer multiplies the error surface. See cybersecurity training captions for the full analysis.
• Fintech and financial services vocabulary — fintech companies using Continu for internal training have a second vocabulary layer: financial terminology (ACH, SWIFT, FBO accounts, escrow, reconciliation, chargeback, interchange), regulatory abbreviations (PCI DSS, SOX, BSA/AML, KYC), and proprietary product names. Generic STT handles financial regulatory acronyms inconsistently.

Accuracy numbers in context

The 80-90% general accuracy figure for YouTube auto-captions translates to material error densities in technical training content. For a 20-minute training video with 2,800 spoken words, 80% accuracy means approximately 560 incorrect words — roughly one error every two seconds of caption display. In a security awareness training module, those errors cluster on the most important terms: product names, threat actors, attack techniques, and regulatory requirements are the highest-entropy vocabulary items and the first to fail. A hearing-impaired employee receiving 80% accurate captions of a mandatory security awareness training is receiving materially degraded training relative to hearing employees — not an accommodation but a second-class substitute.

The Google Meet recording-to-Continu publishing pipeline

How L&D teams use Google Meet with Continu

One of Continu's most valued Google Workspace integrations is its connection to Google Meet for live virtual instructor-led training (vILT). L&D teams can schedule and host live training sessions in Google Meet and surface those sessions in Continu's course calendar. After a live session ends, the Google Meet recording is typically saved to the host's Google Drive. The L&D team can then take that recording and publish it to Continu as on-demand content — making the live session available to employees who couldn't attend live or as a permanent course asset.

This pipeline — live Google Meet session → Google Drive recording → published to Continu course — is the highest-failure-rate context for YouTube/Drive-hosted caption quality. The reasons compound:

• Live session audio quality — Google Meet recordings capture audio from participants over internet connections of variable quality, with background noise, microphone variation, and the latency artifacts of real-time conferencing. This audio quality is substantially worse than the clean close-mic recordings of a professionally produced training video, and STT accuracy is strongly correlated with audio input quality.
• Live demo narration — when a trainer shares their screen in Google Meet to walk through a product demo, API documentation, or technical workflow, the narration is improvisational and fast-paced. Technical product names, API endpoints, function names, and UI element labels are spoken in rapid succession without the measured pacing of a scripted training video. This is the highest-error-rate speech context for technical vocabulary.
• Q&A and participant contributions — live Q&A in Google Meet captures multiple speakers, often with overlapping or interrupting speech, accents, and varying technical fluency. Participants asking questions may use technical terms in non-standard pronunciations or abbreviated forms that further challenge STT accuracy.
• Screen-sharing narration vocabulary — when a trainer narrates what is on screen (UI elements, navigation paths, menu names, configuration fields), the vocabulary is the product's own interface terminology. For a trainer demonstrating BigQuery, the narration includes table names, SQL function names, project identifiers, and dataset names that are entirely proprietary and unpredictable by a general STT model.

The Google Meet caption problem specifically

Google Meet provides live captions during a session (available to participants in the meeting interface). These live Meet captions are not the same as the recorded video captions. The recorded video uploaded to Google Drive does not inherit the in-meeting live caption track. When the Drive recording is published to Continu or uploaded to YouTube, the caption generation starts from scratch — the audio is re-processed by YouTube's or Drive's STT engine, with no memory of the live caption session.

For L&D teams that record live sessions and publish them to Continu, the captioning workflow must be treated as a post-production step, not a live-session feature. The recording must be transcribed (or auto-transcribed and corrected), and a VTT or SRT file must be uploaded to the YouTube video or Drive file before the on-demand course version is made available to hearing-impaired learners.

GlossCap's glossary-biased decoding is particularly valuable for this use case: when an organization provides its product glossary and Google Cloud service glossary, the transcription engine applies those glossary terms preferentially during decoding, recovering correct product names from the acoustically ambiguous input of a Google Meet recording. For more on the underlying technology, see glossary-biased captioning for engineering training videos. For Google Meet-specific captioning considerations in more detail, see Google Meet captions for training.

Engineering onboarding as the primary high-stakes pipeline

For tech companies using Continu, engineering onboarding training is the most common use of the Google Meet recording-to-Continu pipeline. A staff engineer hosts a live "architecture deep dive" session for new hires in Google Meet, the session is recorded to Google Drive, and that recording becomes the permanent architecture overview module in the Continu engineering onboarding course. This single recording — one live session captured once — may be watched by every new engineer for the next two or three years. The technical vocabulary density is extreme: microservice names, database identifiers, message queue topics, deployment pipeline stages, and infrastructure configuration terminology. A single uncorrected auto-caption file on this recording creates an accessibility gap for every hearing-impaired engineer who joins the company for the life of the recording.

ADA Title I, California FEHA, and EAA for Continu customers

ADA Title I: mandatory training for hearing-impaired tech company employees

ADA Title I (42 U.S.C. § 12112) requires employers with 15 or more employees to provide reasonable accommodations for employees with disabilities, including hearing impairments. Training video delivered through Continu that is mandatory for all employees — security awareness training, compliance training, product onboarding, policy acknowledgment training — must be accessible to hearing-impaired employees. The ADA Title I obligation attaches to the mandatory nature of the assignment, not to the platform. Using Continu as the LMS does not create or remove ADA obligations; the obligation exists because the employer has assigned the training as a condition of employment or role performance.

YouTube auto-captions do not satisfy the ADA Title I accommodation standard for mandatory training. An employer that provides a hearing-impaired employee with access to a Continu course containing uncaptioned or auto-captioned video has not provided equivalent access. The standard is functional equivalence: the hearing-impaired employee should be able to obtain the same information and understanding from the training as a hearing employee watching without captions. 80-90% accuracy auto-captions do not provide functional equivalence when technical vocabulary comprehension is material to the training's purpose.

California FEHA: five employees and near-universal applicability to Continu's ICP

California FEHA (Gov. Code § 12940(m)) requires employers with five or more employees to provide reasonable accommodations for employees with disabilities. The five-employee threshold makes FEHA obligations near-universal for California-registered SaaS and technology companies — Continu's primary customer base. Continu was founded in San Francisco and its customer base is concentrated in the California technology sector: companies registered in California, with employees in California, delivering training through Continu to those California employees.

California FEHA applies regardless of whether the employer also meets the ADA's 15-employee threshold. A six-person startup using Continu for engineering onboarding is a FEHA-covered employer for any hearing-impaired employee it hires. As the company grows past 15 employees, ADA Title I adds federal obligations on top of the state obligation. The practical captioning requirement is the same at both thresholds: accurate captions for mandatory training video.

California also imposes specific mandatory training requirements that carry their own captioning obligations when delivered via video. AB 1825 and SB 1343 require harassment prevention training for supervisors (2 hours) and non-supervisors (1 hour) in California. If this training is delivered as video through Continu, it must include accurate captions. The harassment prevention training content includes legal terminology (DFEH/EEOC, FEHA, hostile work environment, quid pro quo, bystander intervention) that generic STT handles inconsistently. Delivering inaccurate captions on legally mandated harassment prevention training is a compliance failure at multiple levels simultaneously.

European Accessibility Act for international SaaS companies using Continu

The European Accessibility Act (EAA), fully effective from June 28, 2025, requires that digital products and services placed on the EU market meet WCAG 2.1 AA accessibility standards. For international SaaS companies using Continu for internal employee training — companies with employees in EU member states — the EAA applies to training content accessed by those EU-based employees if the content falls within scope of the Directive's transposition in the relevant member state.

The practical consequence for a US-headquartered SaaS company using Continu with remote employees in Germany, France, the Netherlands, or other EU countries: training video delivered to those EU employees through Continu should meet WCAG 2.1 AA caption accuracy standards. YouTube auto-captions at 80-90% general accuracy do not meet the WCAG 2.1 AA standard. Corrected VTT or SRT files meeting the 99% threshold are the appropriate solution. For the broader digital accessibility framework, see state and international digital accessibility laws.

Compliance training as the mandatory-training floor

For Continu customers in regulated industries or with California employees, the following categories of mandatory training create the highest-priority captioning obligations:

• Security awareness training — virtually all tech companies require annual security awareness training. This training covers phishing recognition, password hygiene, multi-factor authentication, data handling policies, and incident reporting procedures. For cybersecurity companies, the training vocabulary is dense with product names, attack technique identifiers, and threat actor terminology. All employees are mandatory assignees, making this the broadest mandatory training captioning obligation.
• Data privacy and GDPR/CCPA training — SaaS companies with EU or California customers are required to train employees on data privacy obligations. Training video on GDPR data subject rights, CCPA opt-out workflows, data minimization principles, and breach notification procedures must be accurately captioned for hearing-impaired employees assigned this mandatory training.
• Harassment prevention (California SB 1343) — mandatory for all California employees as described above. The two-year re-training cycle means hearing-impaired employees encounter this training repeatedly, amplifying any captioning quality failure over time.
• Product compliance and certification training — fintech companies must train employees on BSA/AML, KYC, and financial regulatory requirements. Healthcare tech companies must train on HIPAA. Mandatory regulatory compliance training carries both ADA accommodation obligations and, for public companies, regulatory audit exposure if accessibility is challenged.

Caption workflow for Continu

Step 1: identify which videos in your Continu courses need captioning

Before addressing the captioning workflow, audit which videos in your Continu courses are YouTube-hosted versus Google Drive-hosted, and whether any have existing corrected caption tracks or only YouTube auto-captions. In Continu's course builder, video modules show the underlying video source (YouTube URL or Google Drive link). For each YouTube-hosted video, check the YouTube Studio Subtitles section to determine whether the video has only the auto-generated track (labeled "Automatic captions" or "Auto-translated") or a manually uploaded corrected track. Prioritize mandatory training courses and content assigned to hearing-impaired employees or flagged as ADA/FEHA accommodation requirements.

Step 2: generate corrected VTT or SRT files with a company-specific glossary

For each video requiring corrected captions:

1. Submit the video audio to GlossCap with your company glossary. The glossary should include: your company's product names, API names, internal system identifiers, Google Cloud services used in training content, and any other high-frequency technical terms that appear in the video. GlossCap's glossary-biased decoding applies these terms preferentially during speech-to-text processing, recovering correct terminology from acoustically ambiguous input.
2. Review the returned VTT or SRT draft for any remaining errors, particularly on speaker-specific pronunciation variants or new terms not yet in the glossary. Add newly encountered terms to your glossary for future submissions.
3. Finalize the corrected VTT or SRT file. The output should meet the WCAG 2.1 AA 99% accuracy standard for the specific technical vocabulary in your training content.

Step 3: upload the corrected caption file to YouTube or Google Drive

For YouTube-hosted Continu video:

1. Go to YouTube Studio (studio.youtube.com) and open the video.
2. In the left menu, click Subtitles.
3. Click Add language if English is not already listed, or click the existing English row.
4. Under Subtitles, click Add. Upload your corrected VTT or SRT file.
5. After uploading, consider turning off the auto-generated track if YouTube allows it for your video to prevent caption-track confusion. YouTube's interface may only allow hiding auto-captions when a manual track is published for the same language.
6. Publish the corrected caption track. The corrected track now appears as the default CC option when the video is played on YouTube.com or embedded in a Continu course.

For Google Drive-hosted Continu video:

1. Prepare a VTT file with the same base filename as your video (e.g., if your video is security-awareness-2026.mp4, name the caption file security-awareness-2026.vtt).
2. Upload the VTT file to the same Google Drive folder as the video file.
3. In Google Drive, open the video and verify the caption track is available through the Drive video player's CC control. Alternatively, use Google Drive's video player interface to add a subtitle/caption track directly through the player menu.
4. Confirm the caption track appears correctly when the video is played through Continu's course player.

Step 4: establish a captioning-before-publish policy

The most effective long-term solution is to make caption upload a required step before any video is published to a live Continu course. Establish a content checklist that includes caption file upload as a mandatory pre-publish verification step. For Google Meet recordings being published to Continu as on-demand content, the post-production workflow should include: download the recording, submit to GlossCap with the session vocabulary glossary, upload the corrected VTT to YouTube or Drive, then mark the Continu course module as ready for publication. This policy prevents the accumulation of uncaptioned video assets and eliminates the need for retrospective remediation campaigns.

See GlossCap pricing

FAQ — Continu captions

Further reading

• WCAG 2.1 AA captions: the accuracy standard for technical training video
• Google Meet captions: recording live training sessions and captioning for on-demand publishing
• ADA Title I captions: employer accommodation obligations for training video
• VTT captions for training videos: uploading to YouTube and Google Drive
• SRT captions for training videos: format, upload workflow, and accuracy requirements
• Engineering onboarding captions: architecture deep-dives and technical vocabulary density
• Cybersecurity training captions: SIEM, SOAR, EDR vocabulary and accuracy requirements
• Bridge LMS captions: Instructure corporate L&D platform for technology employers
• 360Learning captions: peer-authored learning and EU-focused LMS captioning
• Glossary-biased captioning for engineering training: how technical vocabulary is recovered from ambiguous audio
• ADA compliance for training video: what L&D teams need to fix in employer-assigned training
• State and international digital accessibility laws: California FEHA, EAA, and AODA