🪤 Dataviz traps

Good morning! ☀️

Last week, I shared the results of the Pacific Dataviz Challenge and how impressed I was by the technical level of the entries.

Note: if you still feel limited by technical issues in your work, check out my D3 course. It solves them all!

Today, I suggest we go through the most common mistakes I spotted across the 40 projects I reviewed as a judge, so you can steer clear of them in your own work.

Let’s go!

1️⃣ Too much Claude kills the Claude

A striking number of projects were built with Claude as an AI assistant.

They’re easy to spot, because Claude leaves very recognizable design fingerprints: warm off-white backgrounds, big serif headlines, a terracotta accent color, rounded cards with soft shadows, and a row of “key stat” boxes with giant numbers at the top.

For instance:

It’s not bad design at all!

But if you stick with the defaults, your project will look generic, because half the web looks like this now.

If you're using AI, start by building a personal design system. And even better, create a skill you can apply each time!

2️⃣ Tell a story, don’t just show the data

AI makes it easy to build 12 pages with 40 charts each, covering every possible angle of the dataset. But that doesn’t help your reader at all.

Instead they want to understand your main discovery. Your main point. That’s it.

As a judge, this was the most painful mistake. Some entries would have taken a full hour to read. They used every single dataset provided, and offered no real takeaway.

Take a look at the winning entry, it's not too long and uses only a sample of the available datasets!

3️⃣ Less is more

I’ve written about this before, and I’ll say it again: in my opinion, this is the new challenge in dataviz.

The technical barrier is so low that anyone can churn out endless flashy, gimmicky content. And all that noise loses your audience instead of telling them a story.

Here is one example among many others. A button allowing to zoom in and out the whole page. 100% redundant with the view -> zoom in native feature of your browser:

Not the worst thing, but add a few useless buttons like this and the app becomes cluttered and hard to follow.

If you only remember one point from this email, make it this one. Your reader has a very, very limited amount of attention to give you. Many of the points below follow from this one.

4️⃣ People don’t read

Your reader is lazy and in a hurry. Assume they will skip the text around your charts. Your charts must explain themselves.

I saw many entries with dedicated “Explanation” sections. In my opinion, they’re useless and even counterproductive. The better fix is a simpler chart that speaks for itself.

5️⃣ Using the wrong chart type

AI is great at building whatever you ask for. It’s much worse at telling you you’re heading in the wrong direction.

Combine that with our natural urge to pick unusual chart types to look original and exciting, and you end up with a lot charts that don’t serve the story.

In the example above, the bubble map with a slider on top doesn’t work well. It's very hard to read timeseries through colors changing when slider moves!

Perhaps a simple line chart would have been far more readable.

Check my decision tree for hints on the proper chart type!

6️⃣ Paragraphs that are too long

When you build a data-driven web page, think like a newspaper.

Lines longer than about 60 to 75 characters (roughly 10 to 15 words) get hard to read. When a paragraph runs long, use formatting to highlight what matters most.

Structure helps people follow along.

7️⃣ Keep animations light

Building animations is fun, and it’s tempting to stretch them out for maximum wow effect.

But a long overwhelming animation gets annoying fast. Treat it like salt: a little enhances the dish, too much ruins it.

8️⃣ Direct labels beat legends

A classic, but it can’t be said enough.

Whenever you can label your data directly instead of using a legend, do it. You’ll save your reader’s brain a lot of back-and-forth.

The bottom line

Once again, I was blown away by the quality of the entries. 👏👏👏

This list is not about mocking anyone: congratulations to every participant, I had a great time going through your work.

I just hope reading about what a judge feels can help you avoid a few common traps on your next viz project!

Wishing you the very best ❤️

Yan

PS: If you enjoy dataviz caveats, I have a much bigger collection here. And I’m preparing a set of flashcards on the topic, stay tuned!

PPS: wonderful newspaper dataviz page and about the love for coding

Yan Holtz

Find me on X, LinkedIn, or check my Homepage