Great chart titles succinctly tell the reader an insight about the data. In a basic form, this is how we tell a data story.
In some ways, great chart titles are easy – you just type a few more words and add a period. In some ways, it’s hard – you have to know the insight. I wondered if AI tools could make insight-seeking a breeze, so I tapped ChatGPT and Gemini (which has been called Bard).
I started with a very simple and fake data set.
Here was my prompt:
write a title for this dataset that provides the reader with a clear takeaway point. The dataset is about the number of animals at a rescue
dogs 30
cats 45
rabbits 20
snakes 2
Gemini’s first title suggestion was: Canine Captives: Dogs Take Top Spot at Animal Rescue.
Gemini… there seems to be a fundamental misunderstanding about how numbers work. Dogs may be first in the list but there are more cats in this dataset. Though, you get a point for having an insight in this title, even if it’s incorrect.
Gemini also suggested: Unwanted Souls Take Refuge: Number of Animals Awaiting Forever Homes. Cuter than your average academic title, but this doesn’t contain a takeaway insight.
ChatGPT provided: Diverse Rescues: Cats Abound, Snakes Minimal. And Rescue Harmony: Cats Thrive, Snakes Coexist.
Which got me thinking WHY DO ALL THESE TITLES INCLUDE COLONS? Good lord make it stop.
While ChatGPT is technically correct – these sound like insights – I struggled with their choice of “diverse” and “harmony.” By what measure is this diverse? Who said anything about there being harmony?
The real work in telling a story with data is in the analysis. Exploring the data is when we generate the insights. And just like humans, AI seems to be struggling at this step.
I tried a slightly more complicated dataset, but one that has a clear story to it.
I gave AI the data behind this chart:
I wrote about this data over here, where I say it took me about two minutes to turn this from a table into an insightful visual that inspired a big rethink for my client. It has an obvious pattern in the data and I’d created a succinct 5 word insight-driven title.
The way it’s been sold to me, AI should be able to locate that same insight in way less time (because my two minutes included making the graph too).
Here was my prompt: make a title for this dataset that communicates the takeaway message in a single sentence. The sentence should be sentence case, with a capital only on the first word and a period at the end.
Calls Handled Calls Abandoned
Jan 14007 2877
Feb 12727 1633
Mar 15060 485
Apr 14203 365
May 18086 972
Jun 17480 907
Jul 19106 2817
Aug 17498 4729
Sep 11096 5951
Oct 9845 6208
Nov 8433 5039
Dec 2950 315
ChatGPT said: Efficient Operations: Handling Increased Calls with Minimal Abandonment.
👎🏼 Not an insight. Not a sentence. Not sentence case.
Me: Try again. Make the title a full sentence, not a two part sentence. Use sentence case.
ChatGPT: Effective Call Management: Handling increased calls with minimal abandonment demonstrates operational efficiency.
👍🏼 Almost sentence case. I’ll give it to ya.
👍🏼 Appears to be an insight.
👎🏼 That insight doesn’t actually seem to be supported by the data………… uh oh. It would be dangerous to draw conclusions about operational efficiency.
👎🏼 Why the freakin colon??
Same prompt went into Gemini. Did it do any better?
Gemini said:
Calls Handled and Abandoned by Month.
Gemini explained: This title is sentence case, with a capital only on the first word and a period at the end. It clearly communicates the content of the dataset, which is a record of calls handled and abandoned for each month of the year.
Where’s my face palm emoji? Oh here it is. 🤦🏼♀️
Me: Try again. The title, in sentence case, should communicate an insight about the data.
Gemini: Highest Call Abandonment in 10th Month (63.06%)
Y’all, this is pretty close to the insight I have for the chart title. The big difference is that mine is bringing attention to the season, rather than simply the highest month. I saw context where it saw single data points.
People who are new to writing insightful chart titles do exactly what Gemini did – they go directly to the highest value and make that the point. Sometimes it is! Only sometimes, though. It takes some experience with data to draw out larger stories.
So I asked Gemini to try this one more time, using sentence case.
Gemini: A Tale of Two Seasons: Call Center Traffic Peaks in Summer, Plummets at Month-End.
At this point, the colon is making me cry.
And apparently sentence case is just too big of an ask.
Other than that, did Gemini do ok?
Well, it’s true that traffic peaked in summer. But the plummet at month end? Huh? That conclusion can’t be drawn from this dataset.
Can AI write chart titles? Yes.
Can AI write great chart titles? Nope. At least, not yet.
You bring situational awareness, nuance, context, and subject matter knowledge to your datasets, which gives you a vantage point AI doesn’t have right now.
Surely it will learn soon. As you know, it learns from us.
Which means we need to know how to tell clear, insightful stories about our data.
I show you how to do that in my Evergreen Data Four Step Visualization Process. You need to check it out. This is the first lesson in the Data Visualization Academy, my online course.
Chris will tell you, that very first lesson changes e-v-e-r-y-t-h-i-n-g.
I only open the Academy twice a year – and this is one of those moments. Good timing!
Your ability to…
win over clients
move projects ahead
get more attention on your work
bring clarity to team conversations
weave equity into your data stories
become The One who knows what to do
…hinges on your ability to tell clear stories about your data.
The Data Viz Academy is your always-on resource center for figuring out which chart to pick. You’ll learn how to message your data. And exactly how to package it so it lands with your audience.
Your last chance to join us is March 27.
Your ability to…
win over clients
move projects ahead
get more attention on your work
bring clarity to team conversations
weave equity into your data stories
become The One who knows what to do
…hinges on your ability to tell clear stories about your data.
The Data Viz Academy is your always-on resource center for figuring out which chart to pick. You’ll learn how to message your data. And exactly how to package it so it lands with your audience.
Join the VIP list to get first notice about the next time I open up enrollment. I’ll let you in before anyone else and gives you a heads up about possible discounts.
You’ll be able to better partner with AI. Your boss. MarComm. Your team. Because you’ll know how to communicate the insights.
