Country Profile - Canada

The United Nations Development Programme (UNDP) and Elections Canada recently held an AI + Elections conference in Ottawa. Behind the scenes, the organizing team relied on AI tools to help pull the event together.

Osama Aljaber, a Global Digital Democracy Specialist at UNDP, said the decision to use AI was partly deliberate and partly practical, given the subject of the conference.

"I would say it was both," Aljaber said. "The conference itself was about AI and elections, so we also wanted to demonstrate some practical ways AI could support the organization of the event."

Capacity was the other factor. Deadlines were tight, and materials had to be produced in English and French, Canada's two official languages, with a smaller portion also required in Spanish and Arabic.

"Some of these things would have taken significantly more time and resources without AI, particularly given the size of the team and the timeline we were working with," Aljaber said. In some cases, he added, AI made work possible that otherwise would have been cut.

"I think some of this would not have happened without AI," he said. "We would have skipped it."

AI tools used

The team used AI in four main areas, and every output was reviewed by a person before it was made public.

Building the conference website with Webflow and Codex

The team built the site in Webflow using its standard drag-and-drop tools.

Aljaber shared that he has a software engineering background, although coding is no longer part of his day-to-day work. He used an AI coding assistant connected to Webflow through an MCP connector, which lets an AI tool work directly inside another application.

The AI-assisted workflow handled several tasks:

• A countdown clock to the event.
• The site’s colors and theme. The team shared the conference design and asked for a matching website. “It required a little back and forth to refine it, but in 30 minutes it was complete,” Aljaber said.
• Pop-up speaker bios inside the agenda. Webflow’s default sends users to a separate page. “If you’re looking through the agenda and want to learn more about a speaker, you shouldn’t have to leave the page and then navigate your way back,” he said. “It’s a small thing, but we wanted to make that experience as seamless as possible.”
• Uploading more than 100 participating organizations from an Excel registration list into the content management system. Doing it by hand would have meant creating each profile one at a time.
• Sorting speakers alphabetically.

No sensitive information was provided to the tools. Information published on the website was limited to content intended for public display, with participant information published in accordance with the conference consent process.

“All the data we used with these tools was intended for public display,” Aljaber said. “Nothing sensitive went into any of these tools.”

Translating slides and materials with Copilot

Conference communications and materials needed to be available in both English and French. Speakers often sent slides close to the start of the conference. The team used Microsoft Copilot to translate slides into French. Colleagues from Elections Canada then checked the translations.

Preparing slides with Claude

Aljaber used Google Gemini to help build and improve the visuals in his own presentations on the first and last days.

Producing an animated film with AI and Higgsfield

(inset video here) 

Aljaber opened the conference with a short animated film. To my surprise, I learned he made it on a one-hour flight.

“I was actually on the airplane flying from New York to Ottawa,” he said.

He used generative AI to develop the concept and draft the script. He set the rules up front: the output needed to be animated, and no humans should appear in it. The tool offered several options, and he picked one. He then tagged Higgsfield, a video generation tool, to produce the scenes and sound.

The decision not to depict people reflected UNDP safeguards around the use of generative AI and synthetic representations of people.

The animated video didn’t come without cost, because the Higgsfield generation used almost all of his subscription’s monthly token allowance.

“The video used almost all of my monthly token allowance, so there was still a real resource constraint even when using AI,” he said.

He still needed French subtitles. With no Higgsfield tokens left, he used an AI-assisted workflow to add them to the final version. Because of this, the final version was lower quality but met the language requirement without extra spending.

Policy and safeguards

UNDP has internal guidance and safeguards governing the use of generative AI, including for text, images and video, as well as requirements around which tools can be used.

“I was working within UNDP’s guidance on the use of AI,” he said. “For me, the important thing is understanding what you can use AI for, what you shouldn’t use it for and where human oversight is needed.”

Most output had more than one reviewer. That included the video, the website and the translation. A colleague tested the site after changes were made. The team switched between English and French to check every feature.

“Sometimes we would add a feature and it would work in English but not in French, or a change would affect something else on the site,” Aljaber said, highlighting that testing caught these problems.

The team also drew a clear line between making things with AI and handing things over to AI. The website, the translations and the film were all made with AI help. But none of it had AI inside it. There was no chatbot on the website, and no AI agent answered emails or managed registration. The agenda was a plain agenda. UNDP used AI to build the conference – it did not use AI to run it.

Lessons for election offices

Election offices plan conferences, train poll workers and run voter education. Aljaber’s experience offers four lessons.

• Start with policy. Know which tools your organization approves and where the limits are before you begin.
• Start with low-risk uses. “I would encourage election management bodies to start with low-risk use cases, learn from them, and build from there rather than trying to introduce AI everywhere at once.”
• Always verify. “Human oversight is essential. AI can help produce something quickly, but people still need to review and take responsibility for the final output,” he said.
• Separate building from running. AI can help make a website, a training video or a translated handout. That does not mean the finished product needs AI in it. For offices still writing their policies, this is a practical place to start. The tool does the drafting, while people own the result.