How I use AI for Data Storytelling
Pere Rovira
Data storytelling and visualization focused on research, public and international institutions.
AI tools like ChatGPT are changing the way we work, but how effective are they when it comes to data storytelling? In this article, I’ll share my experiences with what works well—and what doesn’t—when using AI for data-driven storytelling projects.
In particular, I’ll cover the following:
What works well
What doesn’t work so well
This is quite a long read, but I wanted to be comprehensive, such that you get an idea of the different tasks that I undertake when I create a new storytelling project.
My goal with this article is to introduce you to new ideas and possibilities you may not have considered before. In upcoming posts, I’ll dive deeper into specific points and provide examples to illustrate them.
What Works Well? ?
???Understanding the Problem Domain
As data storytellers, we often have to explain concepts we aren’t deeply familiar with. For instance, I might work on a project about SDGs one day, then switch to another about adolescent challenges or innovation the next.
ChatGPT serves as a helpful teacher ??. It breaks down complex topics, helping me grasp the essential ideas behind the data. This foundational understanding is crucial for building meaningful stories around data.
ChatGPT also excels at summarizing long reports and articles from clients, quickly surfacing key insights. And when I need a quick self-assessment, I even ask ChatGPT to create quizzes to test my knowledge!
??Brainstorming Stories
Sometimes, you’re just given some data and asked to create a narrative without much further indications. Where to start, particularly if you are not familiar with the data? ChatGPT can help.
First, I need it to know about our dataset. For private data, I describe the key indicators and dimensions; for public data, I upload the dataset directly.
Second, I provide ChatGPT with context—such as the industry, audience, and problem domain—.
After completing these two steps, I ask questions like:
Starting with open questions, and then going through some back-and-forth, I generate specific storytelling ideas that I can later test against the data to see if they lead to meaningful insights.
????Copywriting
A common challenge when presenting data is that some people may struggle to interpret a chart. That’s why I believe in clearly stating the main insight directly in the chart title. However, this can be more difficult than it seems—often requiring you to convey a lot of meaning in just a few words.
Instead of settling for generic, boring titles like “Sales Trends in 2024,” I ask ChatGPT to generate several title options, focusing on the communication of insights. This ensures I land on a title that resonates with my audience ?? .
Chart annotations is another useful technique to make the data message clearer and to guide the audience’s attention to key points. I upload my charts to ChatGPT and ask it to suggest annotations, as well as ideas on how to best integrate them within the chart layout.
???Color Palette
While there are great tools for creating color palettes, ChatGPT can assist as well. By providing a couple of base colors (such as a client’s brand colors), I can ask it to generate complementary palettes. In particular, I ask for:
领英推荐
???Emotional Touch
ChatGPT helps me brainstorm personal stories or case studies that resonate with the data. For example, when presenting on data about poverty, it’s more impactful to begin with a human story rather than a statistic.
Data storytelling isn’t just about numbers—it’s also about creating an emotional connection with the audience.
Related to this, you can also try and use emoticons to spice up ????a piece of text. Just copy and paste your text, and ask ChatGPT to suggest emoticons and where to use them.
What Doesn’t Work So Well? ?
???Dependance on previous knowledge
When discussing topics like data visualization and storytelling—subjects I’m very familiar with—I can recognize when ChatGPT’s suggestions are off-track. This allows me to guide the conversation, ask for revisions, and eventually arrive at better solutions.
The more knowledge I have about a particular area, the more useful my interactions with ChatGPT become.
Hence, the advantages I’ve listed earlier may not apply to everyone. The quality of the final output depends heavily on how well you can identify what’s useful and what’s not, which is easier when you deeply understand the topic.
For example, while ChatGPT can suggest color palettes, it may not consider essential aspects like accessibility ??? for colorblind users or proper contrast for text. If I don’t specifically ask it to factor in these concerns, the color suggestions may be suboptimal.
???Creating Charts
While ChatGPT can generate basic charts, they’re not up to the standards required for high-quality presentations or storytelling.
For quick data exploration, it might work, but for polished storytelling not so much.
Plus, if you are fluent with tools like Tableau or Observable, you’ll be much faster at data exploration than using ChatGPT.
????Creating Mockups
I’ve found that using ChatGPT for image generation, such as creating basic website mockups, just doesn’t deliver. The results are often awkward and impractical.
Paper and pencil remain my go-to for quick mockups.
While I wish there were an efficient way to turn paper prototypes into digital ones, I’ve yet to find a reliable solution. Do you know of any?
???Closing thoughts
In summary, ChatGPT has proven to be a valuable tool in many aspects of data storytelling—especially in helping me grasp complex topics, brainstorm creative ideas, and craft compelling texts.
I’ve learnt that success with ChatGPT depends largely on how well you can refine its outputs and guide it with clear, detailed follow-up prompts. In the context of this article, it means that the more you know about data storytelling and visualization, the better your results will be.
AI isn’t just a tool for boosting productivity; it’s about improving outcomes. I don’t use AI to write faster—I use it to write better.
As Steve Jobs famously said, "computers are like bicycles for the mind," but they don’t replace the mind altogether.
I’d love to hear how you’ve used AI tools like ChatGPT in your data storytelling and visualization projects. What challenges have you encountered? Have you found any unexpected ways to use AI creatively? Please share your ideas ???in the comments!
Data Science Team Lead @Boehringer-Ingelheim // M.Sc. in Data Science // Go-to-market
5 个月I think it's a matter of prompt engineering and creating multi-agent systems with specialized skills that have in context what matters in terms of data visualization. An orchestrator that actually requests specifically what he wants to see. A pipeline specialist that transforms data, a dataviz expert that translates that into python code of javascript that can generate plots. It won't be as good as something made by a specialist, but better than most things I'm seeing around in terms of data analytics... Probably.