Using AI to create bilingual onboarding videos

We are currently creating onboarding videos from policies and procedures for a client. The videos are in both English and Arabic.

We’ve found AI is helping us produce the videos faster, but it hasn’t removed the need for instructional design, visual design, translation review, or stakeholder approval. The people involved still make the important decisions. AI works far better when we gave it a defined job, approved source material, and clear project rules.

What the videos need to achieve

The purpose of the videos is to introduce staff to the policy, explain why it matters, and prompt them to think about how they would act in realistic situations.

The finished format combines voiceover, motion graphics, and b-roll footage. We use motion graphics to explain topics such as a process, a potential scenario, or a decision. We also use B-roll to give some topics context and prevent the video becoming just a sequence of text panels.

The production workflow

We start with the approved policy.

Our workflow is:

1. Write an English video script based on the policy and obtain stakeholder approval.
2. Create an Arabic translation and send it for review and approval.
3. Generate the English and Arabic voiceovers. We used AI-generated voices for this project, then asked relevant stakeholders to review the Arabic voiceover.
4. Ask Codex to create a visual plan. The plan identifies what appears on screen, when it appears, and how long it remains visible. We used the same visual plan for both languages.
5. Review and edit the visual plan before making the video.
6. Generate a preview, refine it through focused instructions, and produce the final video once the preview is approved.

We use Camtasia, Audiate, and Codex. The approach should also work with other editing tools and local AI systems that can follow reusable project instructions.

You’ll notice we don’t use PowerPoint slides. This saves a great deal of time, and it simplifies the localisation stages.

AI needs direction from people who understand the work

The most important decisions happen before AI generates anything.

An Instructional Designer decides whether the video should summarise the policy, explore realistic scenarios, or persuade staff to change their behaviour. That shapes the script, visuals, and learning outcome.

A designer decides how the video should look. We’ve found without that direction, AI-generated motion graphics tend to become very generic looking. They might be technically competent, but visuals don’t have any specific relevance to the staff or organisation.

A moodboard and a small set of sample videos help at this stage. They give stakeholders something concrete to approve before the production team creates a series of videos.

Some changes are also quicker to make in an editing tool such as Camtasia. Asking AI to fix a small visual detail through several rounds of prompting can take longer than just fixing it directly.

Creating an approved library of assets

We created a visual system that AI could use when building the videos. It included templates, animation files, fonts, logos, screenshots, and other approved assets.

The library included approximately 20 motion-graphic templates using the client’s brand colours.

For this project, the templates also had to work in both languages. English uses left-to-right text and Latin typefaces. Arabic uses right-to-left text and Arabic typefaces. Creating both versions of the templates meant the system could produce consistent videos in either language.

Below are some initial proof of concepts:

Here is a template that uses a richer visual design:

Use b-roll where it adds meaning

The visual plan identifies moments where video footage would explain the point better than animation or motion graphics.

Stock footage can look generic, irrelevant, or odd. And AI-generated footage can sometimes contain inaccurate details or unrealistic movement.

For this reason, we create b-roll separately using tools such as Google Flow or Seedance, and then add the approved clips to the video in Camtasia.

Reference images made the footage more specific to the organisation. For example, a photograph of the company’s real lobby could become the opening frame of a generated clip. The resulting video is more relevant to staff than a generic office scene.

Give AI a bounded job

The approach that worked was to give AI a limited, well-prepared part of the process.

• For motion graphics, provide reusable templates for common visual types, such as workflows and organisational charts. Base them on a moodboard that defines the intended visual style.
• For animation, provide a folder of approved Lottie assets, or add the selected assets separately.
• For b-roll, ask AI to identify suitable moments and draft prompts. Source or generate the final clips using approved reference images where appropriate.
• For the project as a whole, provide a skill file or equivalent set of instructions covering colours, fonts, available assets, and the expected structure of the work.

It’s important also to address any governance issues. Before sharing policies, screenshots, logos, or reference images with an AI tool, check what the tool is permitted to receive and how it handles the material. And there might be licensing issues to consider, depending on the software and assets you use.

AI can reduce repetitive production work. But you still need some with instructional design skills to decide what staff need to understand, and whether the policy is accurately represented.

For more information, see Turn your policies and procedures into learning materials.