AI-assisted training development in the Ontario Public Service

The problem

The emergency preparedness team at the Government of Ontario needed a training presentation on Microsoft Planner, a task management application the team uses to create, assign, and track work. The presentation needed to walk staff through how and when to use the tool, and it later needed to be adapted for two different audiences: newcomers joining the team and existing staff who needed a refresher.

Developing training materials like this is time-consuming, particularly when the content needs to be clear, well-structured, accessible, and tailored to different audiences. The team had also experienced turnover, and much of the institutional knowledge that newer staff would previously have drawn on from longer-serving colleagues was no longer as readily available. Having a way to get structured feedback on clarity, tone, and presentation flow — without relying on other people's time — would make the process faster and the output stronger.

The solution

To put together the training presentation, Microsoft Copilot was used — the Artificial Intelligence (AI) tool approved for use within the Ontario Public Service. The process combined subject matter expertise with AI-assisted drafting and review.

The work began with a first draft developed independently. A second version was then generated using Copilot. The two were compared, and the strongest elements of each were merged into a single presentation. From there, Copilot was used repeatedly with increasingly specific prompts — first refining for clarity and brevity, then for tone, then for accessibility — progressively tightening the content with each round.

Copilot suggested structural improvements, such as reordering the presentation to lead with when to use Microsoft Planner before covering when not to use it. It converted lengthy prose into concise bullet points, generated slide titles and executive summaries, and added “so what” statements that clarified the purpose of each slide — articulating points the team wanted to make but had not as explicitly drawn out. When the presentation needed to become more operational in tone, Copilot was used slide by slide to adjust the content. When it later needed to be adapted for newcomers rather than existing staff, the tool was used again to adjust the structure, messaging, and level of detail.

Copilot was also used to review the presentation for accessibility, identifying issues such as missing alternative text, poor colour contrast, font choices that did not meet accessibility standards, and improper heading structures. This ensured that the presentation was not only informative but accessible for staff to effectively engage with.

The impact

1. Faster drafting and refinement
Copilot enabled the training presentation to be drafted and refined significantly faster than would have been possible manually. The iterative process transformed the content from something that resembled a script into concise, operational material.

2. One presentation adapted for multiple audiences
A single presentation was adapted for both newcomers to the team and existing staff without needing to start from scratch, saving time and ensuring consistency across both versions.

3. Accelerated accessibility review
Copilot identified accessibility issues such as missing alternative text, poor colour contrast, and improper heading structures, making the review process faster and more thorough than a manual check would have been.

4. Structured feedback without relying on colleagues' time
The tool partially filled a role that more experienced colleagues had previously played, serving as a source of feedback on clarity, content, and presentation flow. For newer team members who may not have longer-serving staff to consult informally, this kind of support makes a practical difference to both confidence and the quality of work produced.

Lessons learned

• Start with lower-stakes projects and build experience. Beginning with internal-facing, lower-complexity work allows staff to learn an AI tool's strengths and weaknesses before applying it to larger or more sensitive tasks.
• Be direct and specific with prompts. The more clearly the objective, audience, and desired output are defined, the better the result. Iterating with multiple rounds of increasingly specific prompts — rather than expecting a single prompt to produce a finished result — brings the output closer to what is needed.
• Combine AI drafting with subject matter expertise. Creating an independent draft first and then comparing it to an AI-generated version allows for merging the best of both. AI is most effective when used for first drafts and structural suggestions, with domain knowledge applied on top.
• Review all output before sharing. AI functions differently from humans and can produce language or formatting that does not suit the intended audience. Output should be reviewed carefully before use.
• Be transparent about where AI has been used. Colleagues were made aware of where Copilot had contributed to the training presentation before it was delivered. Openness about AI use supports trust and allows others to assess the material with the right expectations.
• AI can support accessibility compliance. Copilot identified accessibility issues that might otherwise have been missed and accelerated the review process against Ontario Public Service standards. For teams producing documents and presentations at volume, this is a practical and immediate way to check that content meets accessibility requirements.
• AI does not replace professional judgement. Final policy decisions, sensitive language, and anything requiring political or legal sign-off should not rely on AI. For example, updating joint legal agreements with partner jurisdictions on emergency preparedness involves considerations specific to partner organisations and internal requirements that AI would not be aware of. It is important to exercise judgement over the information being put into AI tools, and to leave critical decisions to human expertise. In conclusion, this case study’s lessons reiterate the importance of human judgement and critically underscore that human creativity and intelligence cannot be replaced by AI but can rather be complemented by it. Finally, while this case study’s experiences also highlight AI’s risks and disadvantages, at a fundamental level they underscore that as our human race proceeds into our new epoch with the rise of AI and its many unknowns, this case study’s fundamental lesson highlights that AI will not replace humans, however professionals without the knowledge and awareness of AI’s strengths, vulnerabilities, benefits and risks, as described above, may become anachronistic in our new world.