Using AI to build an Outstanding and Accessible dashboard

Week 2 ?? Power BI DataViz World Championships

We are currently in the middle of the 微软 Power BI DataViz World Championships. The theme for Week 2 was Travel. No dataset was provided, so participants had to find their own data within the theme as a starting point.

After selecting a suitable dataset and transforming it into a star model, the challenge was to identify a key issue, create initial visuals, and build a dashboard that stands out. It needed to be clean, easy to understand, and accessible, all to capture the judges' attention.

I use ChatGPT regularly, mostly to review my LinkedIn posts and articles, catch typos, check my English spelling and grammar, and improve the flow of my writing. But during the championships, I took it a step further. I provided ChatGPT with details about my dataset, the insights I was exploring, and the judging criteria.

I used the free ChatGPT-4o version, keeping the entire conversation in a single chat. This allowed the model to retain context, incorporating all previous discussions and advice into each new response.

In this article, I’ll walk through how I leveraged ChatGPT throughout the process. While I used ChatGPT, similar results could likely be achieved with other tools, such as Copilot.

Using Chat GPT to set the right color theme

The green and grey colors in my dashboard were chosen with the help of ChatGPT. I asked for two environment-related colors, one representing pollution and the other a good, low-CO2 option. Additionally, they needed to have high contrast to ensure accessibility.

After a few back-and-forth discussions, I finalized my selection and incorporated them into my design. Here you see the advice and the color theme I ultimately used, with the ChatGPT-suggested green and grey in positions 1 and 4:

Using the Judging Criteria to fine-tune the dashboard

While normalizing the dataset and splitting it into facts and dimensions, I had already shared a lot of details with ChatGPT about my dataset, focus points, and measures. So when I introduced the judging criteria, I immediately received valuable and actionable feedback.

Adding context with a text box

After reviewing the feedback and making improvements to the dashboard, I asked ChatGPT for help with a text box. I wanted to place it in the bottom right corner to provide context about CO? emissions and travel. I requested a short header and around six sentences.

What I got back was a bit more than six sentences, but it served as a great starting point.

Using dashboard screenprints for fine-tuning

You can upload a screenshot to ChatGPT to get feedback on possible improvements. Of course, always be mindful of the content you share. In my case, I only provided anonymized data.

The feedback I received was extremely useful, both in highlighting what worked well and suggesting improvements. Everything was to the point, easy to understand, and simple to implement.

What impressed me the most was that ChatGPT read all the text on my dashboard, even catching typos in titles and suggesting more user-friendly subtitles! ??

Refining the Dashboard with a Second Review

After implementing the initial feedback, I shared a second screenshot with ChatGPT. This time, I mentioned that I had worked on the previous advice and asked if there was anything I had overlooked.

Once again, the feedback was impressive, highlighting areas for better readability, balance, and consistency. Here’s the response I received:

Making your dashboard accessible

Before Week 1, I had little hands-on experience with accessibility in dashboards—aside from attending sessions on the topic. I knew where to find alt text and tab order settings, but I had never actually used them.

So, I turned to ChatGPT for guidance and asked:

"Now I need to make it accessible. How do I get the alt text in the right way in the dashboard, and what about the tab order? Do I need that for every element?"

Like before, the tips I received were very helpful. I applied most of them in the final report, ensuring the dashboard met accessibility standards.

Alt text for visuals

Since I had no prior experience writing Alt text and no time for extensive screen reader testing (yet), I asked ChatGPT to help me create the descriptions.

The first challenge I encountered was the top card visual. It contains four separate KPI cards, but Power BI only allows a single Alt text entry for the entire visual. So, I asked ChatGPT to combine the information from all four into one concise description.

Here’s the Alt text I received for each visual in my dashboard, which I used in the final version.

Final thoughts

Week 2 was a busy one! ChatGPT helped me make my dashboard both outstanding and accessible. As I write this article, my submission post has received 103 likes, and the short video has been viewed 1.919 times. One in three people clicked on the video, with an average watch time of 17 seconds, about 20% of the total duration. Pretty high engagement!

Using ChatGPT saved me a lot of time and taught me valuable lessons about accessibility and designing clearer dashboards for end users.

After submitting (I just couldn’t wait any longer!), I tested the dashboard with a screen reader and a contrast checker. The screen reader worked, but I need to dive deeper to make it truly useful. The contrast checker, however, failed. The green color I used didn’t meet accessibility standards, even the bright bars weren’t fully compliant, and my "nicer grey" (#b3b3b3) turned out to have too little contrast as well.

There's still a lot to learn, adapt to, and improve, but the journey has begun. I’ve now built two dashboards with accessibility in mind, learned how to use ChatGPT effectively, and next time, the process will be even smoother and better!