What an AI video instructional designer does

An AI video instructional designer creates learning experiences that help a defined audience perform a real task. The role begins with needs analysis and continues through learning objectives, content architecture, modality choice, storyboards, scripts, practice, assessment, production, learning-platform delivery, measurement, maintenance, and improvement. AI video may be the product being taught, a medium used to teach, or a tool inside the designer's production workflow. In every case, the instructional designer remains accountable for learning quality and responsible media use. Current first-party job descriptions demonstrate the work. Synthesia's senior instructional designer role calls for customer discovery, product and support data, audience and prerequisite definitions, measurable objectives, learning paths, assessments, and defensible modality decisions. Its instructional designer and knowledge-and-enablement roles add LMS ownership, videos, live sessions, taxonomy, quality assurance, and cross-functional launch support. Harvey's customer-education role includes outcome-based courses, instructional videos, interactive exercises, storyboards, scripts, AI-tool fluency, and an inspectable portfolio. Job pages can change, so use them as evidence of the profession and verify status before applying.

Why this is not simply making training videos

A polished video can still be instructionally useless. It may explain information the audience already knows, omit the decisions required on the job, provide no practice, or measure only completion. Instructional design asks whether a performance gap exists, whether learning can address it, what learners must do afterward, and what evidence will show improvement. Video is selected when its ability to demonstrate change over time, show a person or process, model judgment, or provide a reusable narrative serves that goal. The role is distinct from video editor, motion designer, learning presenter, and product marketer, although it collaborates with all of them. A designer may write and storyboard but not perform or edit. Another may produce the whole asset. Read the listing for needs analysis, curriculum, LMS, facilitation, video production, product expertise, analytics, and stakeholder ownership. A good portfolio makes your exact contribution visible.

Diagnose a performance problem before prescribing content

Begin with the desired performance, current performance, affected audience, environment, consequence, and evidence. Interview sponsors, subject-matter experts, managers, support teams, and representative learners. Observe the task where permission allows. Review product usage, errors, quality reviews, help searches, support patterns, and existing learning materials. Ask what changed and why the issue matters now. Then test whether instruction is the right response. The cause may be missing access, unclear policy, poor interface, insufficient time, absent feedback, conflicting incentives, or a product defect. Training cannot repair those conditions by itself. Document the performance gap, likely causes, evidence, non-learning actions, learning need, and unknowns. Synthesia's current senior role explicitly asks designers to separate real learning needs from content requests and bring defensible evidence. That is a useful portfolio standard: show one request you would decline or reshape, not only courses you would build.

Define the audience and context

A learner profile should include the job, prior knowledge, motivation, language, access needs, tools, permissions, environment, workflow frequency, time available, and consequence of error. Avoid inventing personality labels that do not change design. An enterprise administrator learning governance needs different examples and practice from a creator learning shot iteration or a developer learning an asynchronous video API. Identify prerequisites and provide a way to check them. If learners need an approved workspace, source asset, browser, role, or API credential, state that before the course begins. Consider whether work happens at a desk, on set, in a studio, on mobile, or under a deadline. Use inclusive, rights-cleared examples that reflect the audience without stereotyping it. Segment only when the difference leads to a meaningful design decision. The audience definition should help you choose language, modality, practice, support, and assessment—not merely fill a template.

Write measurable learning objectives

Describe what the learner will do, under what conditions, and to what standard. 'Understand AI video' is not observable. A stronger objective is: given an approved script and brand template, the learner will create a draft, correct captions, document voice permission, submit it for review, and resolve the specified quality checks. For technical learners, an objective may require sending an authenticated request, handling a failed asynchronous job, and recording the request identifier without exposing a secret. Use objectives to decide what to exclude. Each lesson, example, practice activity, and assessment should support one or more objectives. Separate awareness, procedural performance, judgment, and transfer because they need different evidence. Share objectives with subject-matter experts and workflow owners before production. If stakeholders cannot agree on the desired behavior, a polished storyboard will not solve the underlying ambiguity.

Choose modality from the learning job

Use video when motion, sequence, performance, tone, visual change, or demonstration matters. Use concise text for scanning and exact reference. Use an interactive exercise for decisions and feedback. Use a sandbox for tool practice. Use a live session when facilitation, discussion, coaching, or rapidly changing context justifies the cost. Use a job aid when the learner must perform an infrequent procedure accurately at work. A course can combine these, but every format adds production and maintenance burden. Document why the modality fits the objective, audience, access conditions, and consequence of error. Do not turn a paragraph into a presenter video merely because an AI tool makes production fast. Learners cannot search speech as efficiently as structured text, and a long video is costly to update. Synthesia's senior role asks for deliberate format decisions; your portfolio should show at least one choice to use less video.

Design the learning path

Sequence from prerequisite concepts to authentic performance. Introduce a mental model before a complicated procedure, then provide a demonstration, guided practice, feedback, independent practice, and a transfer aid. Break the path into meaningful tasks rather than arbitrary durations. Give experienced learners a way to assess what they can skip without hiding required safety or governance content. Map dependencies and likely misconceptions. In AI video, learners may need to understand authorized inputs before generation, review before publishing, and deletion or provenance after export. A developer may need authentication and asynchronous states before building a user interface. An administrator may need roles and templates before inviting a large group. Use a curriculum map that links objectives, content, practice, assessment, owner, and maintenance trigger. This makes gaps and duplication visible before expensive production begins.

Storyboard for instruction, not decoration

A useful storyboard defines the learning objective, narration or on-screen copy, visual action, interaction, feedback, accessibility treatment, source asset, and production note for every segment. Show only the interface area or visual detail needed for the step. Use callouts and zoom deliberately; constant motion competes with the task. Keep realistic pauses where a learner must inspect or decide. Mark what is captured live, animated, generated, simulated, or edited. Record the product version and plan for interface changes. Include captions and transcript text during review rather than after final approval. Give subject-matter experts a focused review question: is the procedure accurate, safe, current, and complete for the objective? A storyboard should expose instructional risk early, when changing the sequence costs less than rebuilding the finished module.

Write scripts that teach decisions

Use plain language, concrete verbs, consistent terminology, and short explanations tied to action. Explain why a consequential step matters and what can go wrong. Avoid narrating every visible click or filling time with enthusiasm. When several paths are valid, state how the learner chooses. Give examples and non-examples for judgment tasks. The federal plain-language guidelines provide useful principles for audience-centered organization and readable wording. Write for listening and reading because the script becomes narration, captions, and often a transcript. Spell or pronounce specialized terms deliberately. Keep on-screen text concise and allow enough time to read it. Do not claim that AI removes the need for expertise, guarantees an outcome, or produces universally correct results. If a person, voice, or likeness appears, confirm permission for the intended use and distribution before production.

Produce AI-assisted video with a documented workflow

Define which production steps use AI: outline support, script revision, translation, voice generation, avatar presentation, screen capture cleanup, illustration, caption drafting, or editing assistance. Use only approved tools and inputs. Preserve source material, review decisions, tool or model version when relevant, permissions, and final approval. AI acceleration does not change who owns factual, instructional, brand, accessibility, or rights quality. Evaluate the complete asset at normal speed. Check pronunciation, pacing, gesture, gaze, lip synchronization, temporal stability, on-screen interface accuracy, captions, graphics, audio, and transitions. Confirm that generated elements do not introduce false information or unapproved identity. Maintain an editable source and conventional fallback for high-consequence material. A portfolio process note that shows human review and corrections is more credible than presenting one generated video as an effortless result.

Build practice that resembles the real task

Practice should require the learner to retrieve, decide, create, diagnose, or perform—not merely watch another example. Give a realistic scenario, authorized assets, constraints, and a clear success standard. Begin with support, then reduce it. For a creator, practice might involve correcting a generated draft against brand, caption, and consent criteria. For a support specialist, it might involve turning a vague report into a reproducible escalation. For a developer, it might involve recovering from a failed job safely. Feedback should identify the relevant principle and next action. Avoid feedback that only says correct or incorrect. Include plausible mistakes based on research. Allow safe retry when the work permits it. If the real task has serious consequences, use a sandbox or structured simulation before production access. Learning transfer is more likely when the practice preserves the decisions and constraints of the job instead of replacing them with trivia.

Design assessments that measure performance

Start from the objective and decide what evidence would convince a qualified reviewer. A knowledge check can test recognition, but it cannot prove that a learner can produce, diagnose, or approve a video workflow. Use an authentic task, scenario, critique, project, or observation where feasible. Create a rubric with observable criteria such as accuracy, workflow sequence, rights checks, accessibility, quality review, error handling, and escalation judgment. Establish scoring, feedback, retake, and accommodation rules. Pilot the assessment and inspect items that nearly everyone misses or passes; the problem may be unclear instruction or a weak item rather than learner ability. Protect assessment data and do not use automated scoring for consequential decisions without representative validation and human oversight. Explain what the assessment does not measure. Completion is an activity; demonstrated performance is stronger learning evidence.

Apply accessibility from the first draft

Plan accurate captions, transcripts, audio description or descriptive narration where needed, keyboard operation, visible focus, readable contrast, meaningful headings, alt text, and accessible interactions before production. WCAG 2.2 is the W3C Recommendation used as a core web-accessibility reference. W3C's media accessibility guidance and the U.S. Section 508 resources provide practical context for audiovisual content. The applicable requirements depend on audience and organization, so involve qualified accessibility reviewers. Automatic captions are a draft. Correct names, technical terms, punctuation, timing, speaker changes, and meaningful sound. Ensure essential visual information is available without sight and essential audio without hearing. Do not create an inaccessible transcript viewer or assessment around an accessible video. Test the learning experience with keyboard navigation and appropriate assistive technologies, then document known limits. Accessibility is part of instructional accuracy because excluded learners cannot achieve the objective.

Use inclusive design and multiple paths

CAST's Universal Design for Learning Guidelines offer a framework for supporting learner agency through multiple means of engagement, representation, and action or expression. Apply the framework with purpose rather than duplicating every asset in every format. Offer a searchable text reference beside a demonstration, meaningful practice choices, pacing control, and alternative ways to access essential information where they support the objective. Avoid cultural shorthand, token representation, or generated people whose appearance reinforces stereotypes. Test examples, names, gestures, idioms, color meaning, and scenarios with relevant reviewers. Localization is not word replacement; it includes timing, layout, pronunciation, images, policy, and learner context. Preserve the same learning objective while allowing an appropriate route to reach it. Document design decisions so future updates do not accidentally remove an access path.

Localize learning without losing meaning

Create a source script with controlled terminology, clear references, and a translation-ready structure. Identify text embedded in images or screen captures, interface language, captions, transcripts, downloadable files, narration, assessments, and feedback. Use a glossary and style guide. Engage reviewers who understand the language, subject, audience, and product; a fluent speaker may still miss a technical or instructional error. Check text expansion, line breaks, reading speed, pronunciation, lip synchronization where relevant, date and number formats, examples, cultural meaning, and the equivalence of assessment difficulty. Record which source version each translation follows. When product terminology changes, update the dependent assets together. AI-assisted translation can accelerate a draft, but the designer must verify meaning and preserve the instructional decision. More language versions are not valuable if learners receive inconsistent or unsafe guidance.

Manage privacy and learner data

Learning systems may record identity, enrollment, progress, attempts, scores, feedback, behavior, and accessibility information. Define why each field is collected, who can access it, how long it remains, and how it affects decisions. The NIST Privacy Framework offers a structured way to identify and manage privacy risk. Employer policy and applicable law determine the actual controls. Use the minimum data needed for a legitimate learning purpose. Avoid sending learner records, customer examples, or confidential support data into unapproved AI tools. If a course uses personalization or automated recommendations, explain the behavior and provide a human route for consequential issues. Separate improvement analytics from employee-performance decisions unless governance explicitly permits the connection. In a portfolio, use synthetic learner data and label it. Ethical measurement begins before the dashboard exists.

Govern AI risk in the learning workflow

Use NIST's AI Risk Management Framework to organize governance, context mapping, measurement, and management without claiming that a course or product is certified. Identify where AI enters the learning system, which people or groups could be affected, what errors matter, and who approves output. Test factual accuracy, representation, accessibility, privacy, rights, safety, consistency, and the ability to correct or withdraw material. Set different review levels for different consequences. A low-risk internal draft and a mandatory safety certification should not share the same approval path. Preserve source evidence and require subject-matter review for consequential guidance. Provide a reporting and correction process. AI can accelerate instructional production, but faster publication also accelerates error unless the review system scales with it. The designer's role is to make quality and accountability executable, not merely state them as principles.

Publish through an LMS with reliable standards

An LMS delivery plan includes enrollment, roles, prerequisites, completion, attempts, scoring, certificates, versioning, localization, accessibility, reporting, and retirement. Test the package in the target platform rather than assuming it behaves like the authoring preview. SCORM remains a common interoperability model, while the Experience API, or xAPI, supports statements about learning experiences. The Advanced Distributed Learning Initiative maintains official resources for both. Choose a standard from the actual reporting and portability need, not familiarity alone. Define completion and success precisely. Test suspend and resume, repeated attempts, time zones, browser behavior, keyboard access, reporting, and course replacement. Preserve a version manifest and rollback path. Do not collect granular activity merely because the standard permits it. The platform is part of the learner experience, and a course that cannot launch, resume, or report accurately is not complete.

Measure learning and business transfer

Build a measurement chain: participation, learning evidence, behavior at work, and the operational outcome that motivated the program. Define the baseline and data source before launch. Completion and satisfaction can reveal access or experience problems but do not prove transfer. Assessment can show performance in the learning environment. Workflow quality, reduced error, faster approved work, or successful adoption may show transfer, but other changes can influence them. Combine quantitative signals with interviews, observation, artifact review, and manager feedback. Segment only where the analysis is legitimate and sufficiently supported. State uncertainty and avoid attributing a business result to training without evidence. Set a review date and decision: maintain, revise, add practice, change the non-learning environment, or retire the material. Synthesia's roles emphasize learner feedback and performance signals used for evidence-led improvement, not reporting for its own sake.

Maintain learning as the product changes

AI video products can change models, interfaces, terminology, policies, limits, and output behavior quickly. Assign every learning asset an owner, source references, product version, last review, next review, and change triggers. Connect release planning with customer education so the course does not teach a retired interface on launch day. Use modular design to update one procedure without rebuilding an entire curriculum. Run link, caption, transcript, metadata, and package checks. Review whether screenshots, pronunciations, rights, and accessibility remain valid. Announce meaningful changes to learners and customer-facing teams. Archive obsolete guidance or redirect it clearly; silent contradictions erode trust. A maintenance portfolio artifact can include a content inventory and change-impact map. Employers need designers who care for the system after the attractive launch asset ships.

Collaborate with subject-matter experts

Define what you need from the expert: accurate task steps, exceptions, consequences, examples, review, and approval. Prepare focused questions and observe the work rather than asking the expert to design the course. Separate what experts do automatically from what a novice needs explained. Validate disagreements against product evidence, policy, or the accountable owner instead of averaging opinions. Use a review matrix that assigns instructional, technical, brand, legal, accessibility, and final publication decisions. Time-box reviews and mark which feedback is blocking. Protect the learner objective when a stakeholder tries to turn the module into a feature catalog. After launch, share evidence with experts and update the source of truth. Strong collaboration makes expertise teachable without asking the expert or designer to surrender their distinct craft.

Build an instructional design portfolio

Create one small, complete AI video learning experience using fictional or fully authorized content. Include the performance diagnosis, audience profile, objectives, curriculum map, modality rationale, storyboard, script, accessible video, transcript, practice, rubric-based assessment, facilitator or job aid, LMS test plan, measurement plan, rights manifest, and maintenance map. The finished video is only one part of the case. Annotate decisions and show what changed after a learner test. Provide captions and accessible documents. Use synthetic learner data and label it. Link product facts to current official documentation and identify version assumptions. Do not publish employer curricula, customer recordings, proprietary product captures, assessment banks, or learner records. A hiring manager should be able to see that you can diagnose, design, produce, evaluate, and maintain—not merely operate an authoring tool.

Prepare for an instructional design interview

Expect to critique a course, write objectives, create a storyboard, choose a modality, design an assessment, or respond to a vague training request. Begin with the performance problem and audience. State assumptions and the evidence you would collect. Show how the design supports practice, feedback, accessibility, responsible media use, measurement, and maintenance. If the exercise asks for video, keep production proportional so instructional reasoning remains visible. Prepare stories about a request you reframed, learner evidence that changed a design, difficult expert feedback, an accessibility improvement, and an asset you maintained through product change. Explain your exact contribution. Ask the employer how learning needs are prioritized, how designers access users and data, who owns the LMS, how product releases trigger updates, and how AI-generated media is reviewed. The answers reveal whether education is treated as a measurable customer capability or a content queue.

A practical 30-day preparation plan

In week one, choose one AI video workflow and conduct a small, ethical needs analysis using public and authorized information. Define the audience, performance gap, non-learning causes, and objectives. In week two, map the curriculum, select modalities, and create a storyboard, script, practice activity, and rubric. Review the technical procedure with a knowledgeable peer. In week three, produce a short accessible module with corrected captions, transcript, rights-cleared assets, and a job aid. Ask representative testers to complete the task without live rescue. In week four, revise from evidence, design LMS and measurement plans, and assemble the portfolio case. Verify every claim, permission, link, and format. This preparation cannot guarantee a job, but it shows the complete discipline behind a learning experience rather than only the visible media.

Find AI video instructional designer jobs with purpose

Search instructional designer, senior instructional designer, learning experience designer, customer education, academy content, knowledge and enablement, curriculum designer, e-learning developer, technical trainer, and learning program manager. Read whether the role owns analysis and assessment or primarily production, facilitation, LMS administration, technical writing, or enablement. Current Synthesia and Harvey roles demonstrate several real combinations around AI products, video, interactive learning, customer adoption, and technical audiences. Verify each role on the employer's official careers page. On AIMovieJobs, combine those titles with AI video, generative media, creative software, customer education, learning platform, video production, and technical enablement. Tailor the application to the stated audience and learning product. Lead with one complete case, not a reel alone. The strongest portfolio proves that people can perform something valuable, accurately, accessibly, and responsibly after experiencing your design.

Sources and further reading