Writing the internal captioning policy: approval workflows, exception handling, new-content gates, and the four clauses that stop the drift

Why "we have a process" is not the same as "we have a policy"

A captioning process is a sequence of actions: record, submit, review, publish. A captioning policy is a set of authoritative decisions about that process: who is responsible for what, what the minimum standard is, what happens when the process fails, who has authority to grant exceptions, and how the programme is reviewed and updated. The process can be communicated verbally, documented in a Notion page, or passed along by osmosis between team members. The policy must be written, versioned, approved by an appropriate authority (typically the L&D director or VP), and formally referenced in onboarding materials so that every new hire and every new video producer knows it exists.

The practical difference between having a process and having a policy becomes visible at three moments: when a team member leaves, when an exception arises, and when an auditor asks a question.

When a team member leaves, a process lives in their head and in their communication history. A policy lives in a document that the next person can read on day one. The enterprise LMS caption audit methodology documents the pattern consistently: organisations that lack a written policy have caption quality that degrades measurably over a 12-month period following a role change in the captioning function. Organisations with a written policy maintain quality within a few percentage points of baseline because the standard is documented and the new person can read what "compliant" means before they make their first decision.

When an exception arises — a producer needs to publish a compliance training video three days early because a regulatory deadline shifted, and the vendor cannot turn around the captions in time — the process provides no guidance. Someone decides informally, the exception is granted or denied based on whoever has the authority to say yes at that moment, and the decision is not recorded. The next exception is handled differently by a different person. Within twelve months there are seventeen precedents, none of them consistent, and no one can articulate what the actual rule is. A policy with an exception procedure handles the first exception the same way it will handle the seventeenth.

When an OCR investigator or an ADA plaintiff's counsel asks what your organisation's captioning standard is, "we have a process that the L&D team follows" is not an answer. A written policy is. The compliance matrix post covers the regulatory frameworks — Section 508, ADA Title II, and WCAG 2.1 AA — that your policy must reference. The policy is the document that ties your internal standard to those external requirements. It is also the document that demonstrates intent and systematic effort, which matters significantly in determining whether a compliance failure is treated as a violation or a good-faith operational gap.

What the captioning policy document is not

Before describing what to put in the policy, it helps to be explicit about what does not belong there, because the most common drafting failure is trying to put the wrong things in the policy and creating a document that is too long, too technical, and too fragile to maintain.

Not a how-to guide. The policy document should not describe how to submit a video for captioning, how to review a transcript, how to upload an SRT file to the LMS, or how to use the captioning platform. Those instructions belong in Standard Operating Procedures (SOPs) that sit alongside the policy but are maintained separately. SOPs change frequently — the vendor updates their portal, the LMS changes its file-upload interface, the review checklist gains a new step. If operational instructions are embedded in the policy, every SOP update requires a policy amendment, which requires re-approval, which creates friction that causes teams to skip the update and let the policy drift out of sync with actual practice. Separate the authority document (policy) from the operational instructions (SOPs).

Not the vendor contract. Your vendor contract establishes the SLA for turnaround time, the accuracy guarantee, the data-ownership provisions, and the termination mechanics. None of that language belongs in the internal policy. The policy should reference the vendor relationship in one sentence — "captions are produced via the organisation's contracted captioning vendor; vendor selection and contract terms are managed by the accessibility coordinator" — and stop there. If the vendor changes, the policy does not need to change. If the contract terms change, the policy does not need to change. Embedding contract language in the policy creates two copies of truth that will diverge.

Not the compliance matrix. Your organisation's legal obligation under Section 508, ADA Title II, AODA, EAA, or any other framework is documented in the compliance matrix. The policy should cite the applicable frameworks as a brief reference ("this policy implements the organisation's obligations under ADA Title II, Section 508, and WCAG 2.1 AA; see the compliance matrix for the full regulatory mapping") and move on. The detailed regulatory analysis belongs in the compliance matrix, not in the captioning policy.

Not a statement of aspirations. Policy language like "we are committed to producing high-quality, accessible captions for all learners" is not policy. It is a mission statement. Mission statements belong in the introduction paragraph or in the organisation's broader accessibility statement. The policy clauses themselves should be specific, measurable, and actionable: who does what, when, under what conditions, with what authority, with what consequences for non-compliance.

The four clauses that stop programme drift

Every captioning policy that actually maintains programme quality over time — surviving vendor changes, staff turnover, LMS migrations, and exception accumulation — contains four core clauses. These clauses are the load-bearing structure of the document. The rest of the policy (approval workflows, exception procedures, review cycles) hangs off these four clauses. A policy missing any one of them will show predictable failure patterns within eighteen months of deployment.

Clause 1: The content gate

The content gate clause defines what triggers the captioning requirement. It answers the question "does this video need captions?" without requiring a decision from a human every time. Without this clause, teams spend meaningful time debating whether a given video is in scope — and those debates tend to resolve in favour of the path of least resistance, which is almost always "not in scope."

A functional content gate clause defines scope across four content categories:

Original content. Any video or audio-visual content produced by the organisation for internal or external training, compliance, onboarding, or communication purposes, with a runtime of 60 seconds or longer, requires captions before publication. The 60-second threshold excludes very short clips (animated GIFs, 15-second product teasers) while capturing all content that the WCAG 2.1 AA standard treats as requiring synchronized captions. Below 60 seconds, the content gate clause should specify that the relevant team lead uses their judgement about whether the content conveys meaningful verbal information that a deaf or hard-of-hearing viewer would need — and that if there is any doubt, captions are required.

Revised content. Existing content that has been re-recorded, or whose audio track has been changed by more than 10% of its total runtime, is treated as original content and requires a new captioning submission before republication. Minor edits that do not change the audio track (re-editing the B-roll, adding a graphic overlay, trimming a silent section) do not trigger a new captioning submission, but require the coordinator to confirm in the revision record that the audio track was unchanged. This distinction — audio changed vs. only visual changed — is the line that prevents teams from inadvertently republishing a video with outdated or incorrect captions that no longer match the new audio.

Republished content. Content that is moved from one location to another without any modification — for example, a course being migrated from one LMS to another — does not require a new captioning submission, but does require a caption format compatibility check as part of the migration process. The LMS migration caption checklist covers the specific format compatibility issues (SRT vs. VTT, time-code offset failures, embedded vs. sidecar caption files) that arise during LMS migrations. The content gate clause should reference the LMS migration checklist as the standard for republication events.

Third-party and embedded content. Content produced by an external vendor, partner, or third party and embedded in or linked from the organisation's LMS or intranet is not required to be re-captioned by the organisation. However, the organisation is required to confirm that the embedded content includes accessible captions before embedding it, and to provide an accessible alternative (text transcript or equivalent) if the third-party content does not include captions. The content gate clause should specify a 90-day remediation window for third-party content identified as non-compliant, after which the content must either have an accessible alternative in place or be removed from the platform.

The practical value of the content gate clause is that it removes discretion from the routine decision. When a producer asks "does this video need captions?" the answer is not a judgement call — it is a lookup against the four content categories. This matters especially for newer team members and for high-volume production environments where applying consistent standards across dozens of submissions per month requires a decision rule, not a conversation.

Clause 2: The accuracy standard

The accuracy standard clause defines what "compliant" means in measurable terms. It is the clause that makes your captioning standard auditable — and without auditability, every claim about caption quality is anecdotal.

The clause should specify three things: the accuracy threshold, the measurement protocol, and the remediation trigger.

Accuracy threshold. The organisation's captioning accuracy standard is 99% word-level accuracy, measured against the Described and Captioned Media Program (DCMP) captioning standards. This is the threshold required for WCAG 2.1 AA compliance on pre-recorded audio content. It corresponds to no more than one error per 100 words. For training content with high domain-specific vocabulary density (medical drug names, software SDK terms, regulatory acronyms), the 99% threshold applies to the domain vocabulary as a distinct measurement category — a file that achieves 99.2% overall but only 94% on domain vocabulary is not compliant, because the errors concentrate precisely in the terms that learners need to hear accurately. The caption QA methodology post covers the DCMP spot-check protocol in detail.

Measurement protocol. Accuracy is measured by spot-checking 10 consecutive minutes of content per video using the DCMP methodology. For content shorter than 10 minutes, the entire content is checked. Spot-check samples are drawn from the first third of the video, the middle third, and the final third, with the 10-minute sample spanning at least two of the three sections. The accuracy measurement is performed on a per-word basis, with each word scored as correct or incorrect regardless of whether the error changes meaning. This is the methodology used in OCR investigations and ADA enforcement proceedings; using a looser methodology internally and a stricter one when responding to a complaint creates a gap that is difficult to explain.

Remediation trigger. Any content that scores below 97% on the DCMP spot-check triggers a remediation request to the captioning vendor within five business days. Content that scores between 97% and 99% is flagged for review — the reviewer must determine whether the errors are concentrated in domain vocabulary or distributed across common words. Domain vocabulary errors at any level below 99% trigger a remediation request. General vocabulary errors in the 97%–99% range are reviewed by the accessibility coordinator and may be remediated if the content is in a high-stakes compliance or safety context. Content that scores 95% or below triggers an immediate remediation request and a vendor performance note in the SLA tracking log.

The accuracy standard clause is often the most contested clause when the policy is first drafted, because it commits the organisation to a measurable standard and creates accountability for enforcing it. Some teams resist setting a specific threshold because they do not want to be held to it. This is the wrong instinct: an unspecified accuracy standard means you have no standard, which means you have no basis for requesting remediation from a vendor, no basis for evaluating whether a vendor change is warranted, and no basis for responding to an auditor who asks what your quality target is. Committing to a measurable standard and having a documented record of measuring against it is a stronger compliance posture than claiming accuracy without demonstrating it.

Clause 3: The exception procedure

The exception procedure clause is the most frequently skipped clause in captioning policies, and its absence is the leading cause of programme drift. The reason teams skip it is that they expect exceptions to be rare — and initially, they are. But every captioning programme encounters operational pressure: a compliance deadline that shifts, a vendor turnaround delay, a production emergency, a resource shortage during a peak quarter. Without an exception procedure, each of these events is handled informally. The first informal exception is handled carefully. The tenth is handled carelessly. The fiftieth has become the de facto process.

A functioning exception procedure clause has four components: the eligible exception categories, the approval authority, the exception duration limit, and the exception record requirement.

Eligible exception categories. Not every reason for missing the captioning standard is a legitimate exception. The policy should enumerate the categories that qualify: technical failure (captioning platform or vendor system unavailability that delays delivery beyond the normal turnaround SLA); production emergency (content must be published within 48 hours of creation due to a regulatory, safety, or compliance event and the standard captioning timeline cannot be compressed to fit); resource gap (the captioning coordinator role is vacant or the designated reviewer is on leave, creating a temporary review capacity shortage); and scope dispute (there is a genuine ambiguity about whether the content falls within the content gate that requires a formal determination by the coordinator before the submission can proceed). Any exception request that does not fit these four categories is denied by default.

Approval authority. All exception requests are approved or denied by the accessibility coordinator. In organisations where the coordinator role is unfilled or the coordinator is the person requesting the exception, the approval authority escalates to the L&D director. No other role has exception-granting authority. This matters because in the absence of a designated authority, exceptions get approved by whoever is available and willing to say yes — which tends to be the person with the least context about the compliance implications. The accessibility coordinator role design post covers how to structure the role so it has clear exception-granting authority without becoming a bottleneck in the production workflow.

Exception duration limit. Exceptions are temporary. A technical failure exception expires when the vendor system is restored, with a maximum duration of seven calendar days. A production emergency exception expires when the emergency content is replaced with a properly captioned version, with a maximum duration of 14 calendar days. A resource gap exception expires when the gap is filled or a remediation plan is approved, with a maximum duration of 30 calendar days. A scope dispute exception expires when the coordinator issues a scope determination, with a maximum duration of five business days. Exceptions that have not been resolved by their expiry date are escalated to the L&D director and recorded as overdue in the exception log. There is no mechanism for renewing an exception — a second extension requires a new exception request with a documented justification for why the original remediation plan failed.

Exception record requirement. Every exception must be recorded in the exception log at the time of approval. The log entry must contain: the video ID, the content title, the exception category, the date the exception was requested, the date it was approved or denied, the name of the approving authority, the maximum exception duration, the committed remediation date, and the actual completion date once remediation is finished. The exception log is reviewed quarterly by the L&D director. Any exception that appears in more than two consecutive quarterly reviews without resolution is escalated for executive review and may trigger a programme audit. The exception log is also the document that demonstrates due diligence in an OCR investigation or ADA proceeding — it shows that you knew about the gap, tracked it, and committed to remediation.

Clause 4: The audit trigger

The audit trigger clause defines the conditions that require a formal compliance review. Without it, audits are purely reactive — they happen when a complaint is filed or when a leadership team member asks a question. Reactive audits reveal problems that have already accumulated. Proactive audit triggers, built into the policy, catch drift before it becomes a liability.

The clause should define two types of triggers: scheduled audits and event-triggered audits.

Scheduled audits. A full caption library audit is conducted annually, using the methodology documented in the enterprise LMS caption audit post. The full audit covers every content item in the LMS with a runtime of 60 seconds or longer, verifying that a caption file exists, that the caption file is correctly linked, and that a random sample of 10 items per content domain (compliance training, product training, onboarding, soft skills) is spot-checked against the accuracy standard. In addition to the annual full audit, a quarterly spot-check is conducted on a random sample of 10 content items published in the prior quarter. The quarterly spot-check is not a full audit — it is a calibration check that confirms the quality of recently produced content matches the accuracy standard and provides early warning if the programme is drifting. The accessibility coordinator produces both the annual audit report and the quarterly spot-check report; the L&D director reviews and approves both.

Event-triggered audits. In addition to scheduled audits, the following events automatically trigger a captioning programme review: an OCR complaint or ADA claim related to video content accessibility; a change in the captioning coordinator role (new hire, departure, interim coverage longer than 30 days); an LMS migration or platform change that affects caption delivery; a captioning vendor change; and a material change in the regulatory framework applicable to the organisation (new enforcement guidance, new statute, change in compliance deadline). The event-triggered audit does not need to be as comprehensive as the annual full audit — it is a scoped review focused on the area affected by the triggering event. But it must happen within 30 days of the trigger event, and the results must be documented in a brief audit memo that is retained with the programme records.

The audit trigger clause is the mechanism that connects the policy to reality. It forces the organisation to look at actual compliance data on a schedule, not just when a crisis forces the issue. For teams that are subject to OCR investigation or ADA Title II enforcement, having documented evidence of proactive audit cycles is one of the strongest indicators of good-faith compliance effort.

Approval workflow architecture

The four policy clauses define the standards and authorities. The approval workflow defines how content moves through the system in practice. A well-designed workflow is invisible when everything is working correctly — standard submissions route automatically and the accessibility coordinator only sees the edge cases. A poorly designed workflow makes every submission a manual decision, creating a bottleneck that produces exactly the exceptions and workarounds the exception procedure clause is meant to manage.

Most captioning programme workflows have four routing paths: standard submission, exception request, emergency content, and third-party review.

Standard submission path

A standard submission is content that (a) falls within the content gate, (b) has no special timeline constraints, and (c) does not require any exception to the accuracy standard. The standard path routes automatically: the producer submits the content file via the captioning platform, the platform delivers the completed caption file to the designated output location (LMS, shared drive, or export queue), and the accessibility coordinator receives a delivery notification. The coordinator performs a spot-check review on a sampling basis — not on every submission, but on a rotating sample that ensures every content domain is reviewed at the frequency specified in the QA methodology. No human approval is required for standard submissions unless the QA spot-check identifies an accuracy issue that triggers the remediation process.

The standard path should be the path that at least 85% of submissions follow. If more than 15% of submissions require manual decision-making by the coordinator, the content gate definition or the vendor SLA is misaligned with actual production volume and needs to be recalibrated. A coordinator who is spending most of their time approving routine submissions is not performing the oversight function the policy intends — they are performing manual queue management, which is a workflow design failure.

Exception request path

An exception request is any submission where the producer believes that one or more of the policy requirements cannot be met on the standard timeline, or where the content's scope under the content gate is ambiguous. The exception request path requires the producer to submit a brief exception request form (five fields: content ID, reason for exception, exception category from the eligible list, proposed remediation plan, proposed remediation date) before publishing the content. The accessibility coordinator reviews the request within two business days and approves or denies it. Approved exceptions are logged. Denied exceptions are returned to the standard path with a note explaining what additional steps are required before the content can be published.

The exception request form is important because it forces the producer to articulate why the exception is necessary and commit to a remediation plan before the exception is granted. Many exception requests, when the producer has to fill out the form, turn out to be unnecessary — the timeline is workable if submitted now rather than after the content has been edited three more times. The form is not bureaucratic friction for its own sake; it is the friction that converts a casual verbal request into a documented record.

Emergency content path

Emergency content is a small subset of submissions where the content must be published within 48 hours and the standard captioning turnaround time (typically 24–48 hours for standard vendors) cannot be compressed further. Examples include a same-day safety notice following an incident, a regulatory compliance notice following new guidance, or a leadership communication that must go out before an all-hands. The emergency path allows the content to be published without captions, with a visible "captions forthcoming" notice displayed to learners, provided that: the production emergency exception has been approved by the accessibility coordinator; the coordinator has confirmed that the content does not involve safety instructions or ADA-critical compliance content (for which publishing without captions is not acceptable regardless of timeline); and the caption file is delivered and linked within 48 hours of publication. Any emergency content that exceeds the 48-hour caption delivery commitment is immediately escalated to the L&D director.

The 48-hour emergency path is not a back door for routine late submissions. The policy should specify explicitly that emergency path usage is logged, that the exception log is reviewed quarterly, and that any team or content domain where emergency-path submissions represent more than 5% of quarterly volume triggers a production workflow review. Teams that are routinely relying on the emergency path have a production scheduling problem that the captioning policy cannot solve — but the quarterly review data makes the problem visible so it can be addressed at the workflow level.

Third-party and embedded content path

Third-party content — licensed courseware, vendor-produced product training, partner webinar recordings — follows a different routing path because the organisation cannot submit it to its captioning vendor for re-processing without a license right to modify it. The third-party path requires that any new embed or link to external content be reviewed for caption availability before the embed is approved. The review is typically performed by the procurement or legal team at the time of content license negotiation, using a checklist that confirms whether the content includes captions and whether those captions meet the accuracy standard. If the content does not include captions, the procurement checklist requires the vendor to confirm whether captioned versions are available and at what additional cost. If captioned versions are not available and cannot be obtained, the organisation must either provide a text transcript as an accessible alternative or decline to embed the content.

For third-party content that is already embedded in the LMS and was purchased before the captioning policy was implemented, the annual audit identifies the non-compliant items and the coordinator works through them on the 90-day remediation timeline specified in the content gate clause.

Writing the new-content gate

The new-content gate is the decision logic that lives inside the content gate clause. It is the mechanism that answers "does this submission count as new content?" in the cases that are not obvious. Getting this right matters because the most common source of informal exceptions is content that a producer argues is "not really new" — a re-edited version, a re-used module, an updated slide-deck recording — and therefore does not need to go back through the captioning process. Most of the time, those arguments are wrong. The new-content gate definition makes it unambiguous.

When a revised submission is treated as new content

A revised submission (an edited version of existing content) is treated as new content — and requires a new captioning submission — when any of the following conditions are met:

• The audio track has been re-recorded, in full or in part, and the re-recorded portion is longer than 10% of the total runtime. A 20-minute module where 2 minutes 30 seconds of audio have been replaced exceeds the 10% threshold and requires a new captioning submission.
• New narration or spoken content has been added that did not appear in the original version, regardless of duration. Adding a 30-second intro voiceover to an existing module requires a new captioning submission for the added segment, even if the rest of the audio track is unchanged.
• The language of the audio track has changed. A module where the English narration has been replaced with a different-language narration requires a new captioning submission in the new language.
• The title or identifier of the content has changed and the content has been republished as a new course or module rather than as an update to the existing course. If the content appears in the LMS under a new title or new content ID, it is treated as original content for the purposes of the captioning requirement.

A revised submission is not treated as new content — and does not require a new captioning submission — when the following conditions all apply: the audio track has not changed; the visual changes are limited to B-roll substitution, graphic overlay, or silent section trimming; and the content ID and LMS location remain the same. In this case, the existing caption file remains valid, and the producer must note in the content management system that the audio track was unchanged in this revision.

The format compatibility check for republished content

Even when republished content is not treated as new content, it requires a caption format compatibility check before the migration is complete. The most common failure in LMS migrations is that the caption file format supported by the source LMS is not the same as the format supported by the destination LMS. An SRT file that works in TalentLMS may require conversion to VTT for Docebo; a caption file embedded in an MP4 container may need to be extracted as a sidecar file for Cornerstone. The LMS migration caption checklist provides the format compatibility matrix for the major LMS platforms; the new-content gate policy should require that the format compatibility check be completed and documented before any migrated content is published on the new platform. A failed migration that results in content without visible captions is a compliance gap even if the caption file technically exists — the standard is that captions must be displayed to the learner, not that a caption file is present somewhere in the system.

The 60-second threshold in practice

The 60-second threshold in the content gate clause requires clarification for a few common edge cases. A series of short clips that together constitute a module — each clip is 45 seconds, but the module has 12 clips totalling nine minutes — should be treated as a single piece of content for purposes of the captioning requirement, not as twelve separate clips each below the threshold. The captioning requirement applies to the module as a whole. A video that runs 58 seconds but contains a formal spoken presentation of compliance requirements (rather than incidental spoken words) should be flagged by the producer and treated as in-scope by the coordinator, even though it falls below the 60-second threshold. The threshold is a default rule, not an absolute carve-out — the content gate clause should specify that the coordinator can extend the captioning requirement to content shorter than 60 seconds when the content conveys material information that a deaf or hard-of-hearing viewer would need in order to complete the learning objective.

Exception handling in practice

The exception procedure clause specifies the rules. What those rules look like in practice is a set of recurring decisions the accessibility coordinator makes weekly. Understanding the failure patterns helps in writing the clause with the right guardrails.

The most common exception requests

In a functioning captioning programme with 20–50 video submissions per month, the most common exception requests are, in approximate order of frequency:

Vendor turnaround delay. The captioning vendor has missed the standard turnaround SLA, and the content has a hard publish date. This is a legitimate exception trigger (technical failure category) if the delay is on the vendor side. The policy response is to approve the exception for the delay period and simultaneously record the vendor SLA miss in the performance log. If vendor SLA misses are triggering more than two exceptions per quarter, the coordinator should review whether the contracted SLA is realistic for the organisation's submission volume and whether a vendor performance discussion or vendor change review is warranted. The vendor transition playbook covers the decision framework for vendor changes; frequent exception requests due to vendor delay are one of the leading indicators in the stay-vs-switch analysis.

Scope dispute on short-form content. A producer submits a 90-second "microlearning" clip and argues it does not qualify as training content and therefore is not in scope. These disputes are resolved by applying the content gate clause: is the audio track carrying substantive verbal information that a learner needs to achieve the learning objective? If yes, the content is in scope. If the producer can document that the clip contains only background music, ambient sound, or visual-only information (no spoken narration), the coordinator can confirm it is out of scope and note the determination in the exception log. This is not an exception in the traditional sense — it is a scope determination — but it uses the same record-keeping infrastructure.

Production deadline pressure for new-hire onboarding content. New-hire start dates and onboarding programme launch dates create genuine hard deadlines. The correct approach is to plan captioning turnaround time into the onboarding content production calendar, not to treat late submission as a basis for an exception. However, in the first six months after a captioning programme is formally launched, teams frequently underestimate the turnaround time and produce content on a timeline that does not allow for captioning before the new-hire cohort starts. The policy should address this directly: onboarding content with a hard start-date deadline is a high-priority submission; the coordinator should work with the vendor to confirm turnaround time at the time of submission, not at the time of the deadline. If the turnaround cannot be compressed, the emergency path applies — with the "captions forthcoming" notice displayed and the captions delivered within 48 hours of publication.

The exception log as a management tool

The exception log is not just a compliance record — it is a management tool that reveals operational problems in the captioning programme before they become compliance problems. Quarterly review of the exception log should surface: content domains that generate disproportionate exceptions (indicating a production workflow or vendor SLA problem); exception categories that are recurring (indicating a systemic gap in programme design); exceptions that repeatedly miss their remediation commitments (indicating resource or vendor problems that need structural attention); and scope disputes that are recurring on the same content type (indicating a content gate definition that needs to be more specific).

The quarterly review should produce a one-page summary for the L&D director: number of exceptions by category, average exception duration, number of overdue exceptions, and any systemic patterns the coordinator has identified. This summary is the operational data that informs the annual policy review and provides early warning if the programme is drifting.

Policy governance and the annual review cycle

A captioning policy that is written once and never reviewed is a policy that will drift out of alignment with actual practice, regulatory changes, and technology changes. The annual review cycle is the mechanism that keeps the policy current without requiring constant amendments.

What the annual review covers

The annual review should address five questions: Has the regulatory framework changed in ways that affect the accuracy standard or the scope of the captioning requirement? Has the organisation's content production volume or content mix changed in ways that require updating the content gate or the exception procedure? Have exception log patterns revealed systemic gaps in any of the four core clauses? Have vendor or technology changes created new compliance risks or operational opportunities that the policy does not address? And has the policy been applied consistently, or have informal variations accumulated that need to be either incorporated into the policy or corrected?

The regulatory question is particularly relevant for organisations subject to ADA Title II. The Department of Justice issued updated Title II technical guidance in 2024 and 2025 that clarified the application of WCAG 2.1 AA to state and local government entities. Public universities and government-affiliated training programmes should confirm annually that their accuracy standard and content gate definitions align with the current DOJ technical guidance, not just the standard as understood at the time the policy was first written.

The policy amendment process

Between annual reviews, the captioning policy should only be amended when a material change in circumstances makes an existing clause unworkable or non-compliant. The amendment process should require: a written proposal from the accessibility coordinator or L&D director specifying what clause is being changed, why the change is necessary, and what operational impact the change will have; review and approval by the L&D director (and by the legal or compliance function if the change involves the accuracy standard or audit trigger clause); and distribution of the amended policy to all staff in the L&D function and all video producers within 14 days of the amendment taking effect.

Minor clarifications — adding a specific LMS to the format compatibility reference, updating the vendor name, clarifying an ambiguous phrase — do not require a formal amendment. The coordinator can update the policy document with a version note (e.g., "v1.2 — clarified TalentLMS format compatibility guidance, 2026-09-15") without triggering the full amendment process. What does require the formal amendment process: any change to the accuracy threshold, any change to the exception categories, any change to the audit trigger conditions, and any change to the content gate scope definitions.

Onboarding new L&D staff and video producers

The captioning policy should be included in the onboarding documentation for all L&D staff and all video producers, regardless of whether their role directly involves caption production. L&D staff need to know the policy because they may be the first point of contact when a business stakeholder asks about captioning timelines or exceptions. Video producers need to know the policy because they make the submissions, identify the scope questions, and initiate the exception requests. The onboarding reference does not need to be the full policy document — a one-page summary of the four core clauses, the approval workflow paths, and the exception log process is sufficient for most producers. The full policy should be linked and accessible from the one-pager.

The accessibility coordinator role is typically responsible for delivering the captioning policy orientation to new L&D hires, and for re-confirming with existing staff annually that they have read the current version of the policy. This is not a compliance formality — it is the mechanism that closes the single largest cause of informal exception accumulation, which is team members who are unaware of the standard and therefore default to their own judgement about what is acceptable.

The captioning policy document structure

A complete internal captioning policy document has seven sections. The following is the heading-level architecture; each section's content is drawn from the clauses and procedures described above.

Section 1: Purpose and scope

Three to five sentences stating the purpose of the policy (to establish the organisation's internal standard for video caption production, review, and compliance), the legal basis (citing the applicable frameworks: Section 508, ADA Title II, WCAG 2.1 AA, and any other applicable law), and the scope of applicability (all internal and external video content produced by the L&D function, and all third-party video content embedded in or linked from the organisation's LMS or intranet). The purpose section is also where you state the policy effective date, the version number, the approving authority, and the review cycle (annual).

Section 2: Definitions

A glossary of terms used in the policy: original content, revised content, republished content, third-party content, WCAG 2.1 AA, DCMP methodology, accessibility coordinator, exception, audit, caption file, sidecar file, and any other term whose meaning must be specified to prevent ambiguous interpretation. The definitions section is where you establish the 60-second threshold, the 10% audio-change threshold for revised content, and the 90-day remediation window for third-party content.

Section 3: Content gate

The full content gate clause, covering the four content categories (original, revised, republished, third-party) and the specific conditions that trigger or exempt the captioning requirement for each.

Section 4: Accuracy standard

The full accuracy standard clause, specifying the 99% threshold, the DCMP measurement protocol, and the remediation trigger conditions.

Section 5: Approval workflow

The four routing paths (standard, exception, emergency, third-party) with the decision criteria and role responsibilities for each.

Section 6: Exception procedure

The full exception procedure clause: eligible categories, approval authority, duration limits, record requirements, and the quarterly review process.

Section 7: Audit and review

The audit trigger clause (scheduled and event-triggered), the annual policy review process, and the amendment procedure. This section should also specify where the policy document, exception log, and audit reports are stored and who has access to them.

The complete policy document for a mid-size L&D team (10–40 video producers, 20–100 submissions per month) is typically four to eight pages in length. Longer is not better — a policy that can be read in 15 minutes by a new video producer is more likely to be read and followed than a 25-page document that nobody finishes.

Common policy drafting mistakes

The four most common mistakes in captioning policy drafting, and what they look like in practice:

Writing a compliance summary instead of an authority document. The most common version of this mistake is a policy that correctly describes the regulatory requirements (ADA Title II requires WCAG 2.1 AA, which requires 99% accuracy, which means...) but never specifies what the organisation's internal roles, authorities, and procedures are. The reader finishes the document knowing what the law requires and having no idea what they are supposed to do when they have a video to submit. A policy that does not specify who does what is not a policy — it is a regulatory overview. The regulatory overview can be a linked appendix; the policy itself should be entirely focused on internal authority and procedure.

Omitting the exception procedure. Teams that write a policy without an exception procedure often argue that they want to hold a strict line and not create an official exception pathway. The result is not a stricter programme — it is a programme where exceptions happen informally and without records. An official exception procedure with high standards (specific eligible categories, coordinator approval required, short duration limits, mandatory logging) is more restrictive than an informal culture of occasional waiver, because informal exceptions tend to accumulate without limit. The exception procedure clause is not a concession to programme softness; it is the mechanism that makes programme strictness sustainable over time.

Setting the accuracy threshold too high to be operational. A policy that requires 100% accuracy on every submission before publication creates a situation where any caption file with a single error triggers a remediation cycle that blocks publication. In practice, this results in either chronic production delays or chronic informal waivers of the accuracy requirement. The 99% threshold (no more than one error per 100 words) is the WCAG 2.1 AA standard and is also operationally achievable with a quality vendor and a structured feedback loop. Setting the threshold at 99% is both legally defensible and operationally sustainable.

Copying vendor contract language into the policy. Some teams draft the accuracy standard clause by copying the accuracy guarantee language from their vendor contract (e.g., "vendor shall deliver captions at 98% accuracy as measured by the vendor's internal quality assessment process"). This creates two problems: first, the vendor's internal quality assessment process is not the same as the DCMP methodology, and the difference in measurement approach can result in a vendor claiming compliance while the DCMP spot-check shows non-compliance. Second, when the vendor relationship changes, the policy accuracy standard changes with it — which is the wrong dependency. The internal policy should specify the organisation's standard and the measurement protocol, independent of any vendor's claims. The vendor contract is evaluated against the internal standard, not the other way around.

Connecting the policy to the operational layer

A captioning policy is only as effective as the operational infrastructure that executes it. Three connections between the policy and the operational layer are particularly important.

The QA methodology as the policy's measurement arm. The accuracy standard clause specifies a 99% threshold measured by the DCMP protocol. The QA methodology operationalises that measurement: it specifies how to draw the spot-check sample, how to score errors using the DCMP taxonomy, how to record the results, and how to initiate a remediation request when the threshold is not met. The policy and the QA methodology must be explicitly linked — the policy should reference the QA methodology as the measurement standard, and the QA methodology document should reference the accuracy standard clause as its governing authority.

The vendor SLA as the policy's delivery arm. The exception procedure clause has a vendor turnaround delay category because vendor delays are the most common operational trigger for exception requests. The vendor's SLA commitments — turnaround time, accuracy guarantee, escalation process — are the external dependency that determines how often the exception procedure is invoked. A vendor whose standard turnaround is 24 hours with a 98% accuracy guarantee creates a different exception frequency than a vendor whose standard turnaround is 72 hours with a 95% accuracy guarantee. Before finalising the captioning policy, review the vendor SLA contract checklist to confirm that the vendor contract's commitments are consistent with the policy's delivery expectations. If the vendor SLA cannot support the policy's timelines, either the SLA needs to be renegotiated or the policy's timelines need to be recalibrated.

The audit programme as the policy's accountability arm. The audit trigger clause specifies when audits happen and what they cover. The LMS caption audit methodology provides the operational detail: how to enumerate the content library, how to prioritise the audit sample, how to document findings, and how to track remediation. The audit programme is what converts the policy's commitments into verified compliance — without it, the policy is a statement of intent with no mechanism for confirming whether the intent has been achieved.

For organisations that have not yet run a structured audit, the audit methodology post describes a three-week first-audit process that produces a baseline compliance picture. That baseline is valuable not just as a compliance record but as a management data point: it tells you how far the current state of your content library is from the policy standard, which informs the remediation timeline and the exception log review schedule. Teams that run their first audit expecting to find a compliant library and instead find a 60–70% non-compliance rate across the pre-policy content are often surprised — but having that data is better than not having it. The exception log and remediation records that flow from the audit are the evidence of systematic effort that matters in a regulatory proceeding.

Policy considerations for remote and distributed teams

The approval workflow architecture described above assumes a reasonably centralised production function: producers submit content, the coordinator reviews exceptions, the director approves policy amendments. For organisations with distributed or remote L&D teams, where video is produced across multiple business units, geographies, or time zones, the policy requires two additional specifications.

Distributed producer responsibilities. In a distributed production environment, individual producers or business-unit L&D leads may have limited awareness of the captioning policy and no direct relationship with the accessibility coordinator. The policy should specify a "captioning liaison" role within each major business unit — this is typically an existing L&D or instructional design role that takes on the captioning policy orientation for their unit, serves as the first contact for scope questions, and ensures that new producers in their unit complete the captioning policy orientation. The liaison role does not have exception-granting authority — that remains with the accessibility coordinator — but it provides a distributed point of contact that reduces the volume of routine questions reaching the coordinator directly. The remote and hybrid video captioning post covers the production workflow considerations for distributed teams; the policy layer needs to mirror the distributed structure of the production workflow.

Submission routing in multi-LMS environments. Organisations that run multiple LMS platforms — one for internal training, one for customer education, one for compliance training — need to specify in the policy which submission routing rules apply to which platform. The content gate and accuracy standard apply equally across all platforms, but the format compatibility check and the third-party content review process may differ per platform. The policy should include a brief platform-specific appendix, or reference a platform configuration guide maintained by the coordinator, rather than embedding platform-specific instructions in the main policy document. This keeps the policy document platform-agnostic and easier to maintain as platforms change.

Frequently asked questions

How long should the policy document be?

Four to eight pages is the target range for a mid-size L&D team. The four core clauses (content gate, accuracy standard, exception procedure, audit trigger) and the approval workflow each require one to two pages to specify clearly. The definitions section adds a page. The governance and review section adds half a page. A complete, functional policy should fit within eight pages — if it is longer, look for content that belongs in an SOP or an appendix rather than in the policy document itself.

Who should approve the policy?

The captioning policy should be approved by the L&D director or VP, with review (not approval authority) by the legal or compliance function for the regulatory citations and the accuracy standard clause. In organisations with a dedicated accessibility officer or ADA compliance officer, that role should also review the policy before it is finalised. The accessibility coordinator typically drafts the policy and manages the review process, but the policy is approved by the L&D director because it carries organisational authority and commits resources (the coordinator's time, the vendor relationship, the audit cycle) that the director controls.

Does the policy apply to video content on the public website?

A captioning policy scoped to the L&D function applies to internal training content and to customer education content produced and hosted by the L&D or customer success function. Public website video content (product demos, testimonials, marketing videos) is typically governed by a separate web accessibility policy maintained by the digital or marketing team. If the organisation does not have a web accessibility policy, the L&D captioning policy should note this gap explicitly and recommend that a web accessibility policy be developed for externally-facing video content — particularly for organisations subject to ADA Title III or the European Accessibility Act. The compliance matrix post maps the applicable frameworks to the organisation type and content category.

How do we handle content that was produced before the policy was implemented?

Pre-policy content is addressed through the annual audit and the remediation tracking process. The annual audit identifies which pre-policy content is non-compliant; the remediation plan prioritises by risk (ADA Title II scope, training content with active learners, content in frequently audited domains) and provides a timeline for bringing the content library up to the policy standard. The remediation plan does not require that all pre-policy content be re-captioned immediately — that is operationally infeasible for most organisations — but it does require that the organisation have a documented plan and be making measurable progress against it. The exception log should include a pre-policy remediation category that tracks the progress of the historical correction work separately from current-production exceptions.

What happens if the accessibility coordinator role is vacant?

The policy should specify a coverage protocol for the coordinator role. During a vacancy or extended leave, the approval authority for exception requests escalates to the L&D director. The standard submission path (which does not require coordinator approval) continues unaffected. The coordinator should designate a deputy from among the existing L&D team who has been trained on the exception categories and the exception log process — the deputy does not have policy-amendment authority but can handle routine exception approvals while the full coordinator role is filled. The accessibility coordinator playbook covers the deputy-designation process and the knowledge transfer requirements that make coverage functional rather than nominal.

How does the policy interact with a vendor change?

The captioning policy is vendor-agnostic: it specifies the organisation's internal standard, not the vendor's delivery process. When the vendor changes, the policy does not change. What changes is the operational configuration (submission workflow, format compatibility, delivery integration) and potentially the exception frequency if the new vendor's turnaround time or accuracy level differs from the prior vendor's. The vendor transition playbook covers the operational steps; the policy covers the standard that the new vendor's output must meet. The audit trigger clause specifies that a vendor change automatically triggers a scoped compliance review — typically a spot-check of the first 20 submissions from the new vendor — to confirm that the new vendor is delivering to the policy standard before the full production volume transitions.

Can the policy be shared with external stakeholders or auditors?

Yes. The internal captioning policy is a document that can and should be shared with OCR investigators, ADA compliance auditors, and accessibility certification bodies on request. The fact that the organisation has a written policy, that the policy specifies measurable standards, and that the policy includes a systematic audit process is strong evidence of good-faith compliance effort. The exception log and audit reports are also shareable but should be reviewed by the legal function before external disclosure, because they document the instances where the organisation fell short of its own standard — which provides context about the nature and frequency of gaps but should be characterised carefully in an enforcement context.

Further reading

• The L&D accessibility coordinator playbook — role design, authority structure, and the weekly operating cadence for the role that owns the captioning programme.
• Building a caption compliance programme from scratch — the 90-day sequence for moving from ad hoc captioning to a structured programme; this post covers what you do after the programme is built.
• Enterprise LMS caption audit methodology — the operational process for the annual audit and event-triggered audits referenced in the audit trigger clause.
• Caption QA methodology for training video teams — the measurement protocol that operationalises the accuracy standard clause.
• Section 508 vs ADA Title II vs WCAG 2.1 AA compliance matrix — the regulatory mapping that the policy's purpose section should reference.
• Caption vendor SLA and contract review checklist — the vendor contract provisions that the captioning policy depends on for delivery performance.
• Captioning vendor transition playbook — operational steps for the vendor change event that triggers the policy's audit clause and exception procedure.
• LMS migration caption checklist — format compatibility process for the republished-content path in the new-content gate.
• Captioning async training video for remote and hybrid workforces — production workflow considerations that inform the distributed-team policy specifications.
• Making the caption ROI argument to a finance executive audience — the business case framing for securing the resources the policy commits (coordinator time, vendor budget, audit cycles).