How to Use AI to Create Maps
Can I create a map with little time, no Python, and only some text data? Well . . .
A question in a master’s level class this week about the use of maps by historians got me thinking about how we historians might use AI to generate our own maps, saving us time and money but also allowing those of us who are highly visual to “see” our data easily and quickly.
That led me down a too many hours long rabbit hole and the answer is yes and no—it can be done but not easily AND accurately right now. Do I think this will change? Yes. I’d say try again in six months and it will likely be doable. And by doable, I mean, that I can enter text data into a chatbot and get a visual image of a map that’s accurate.
I’ll share my process and findings—in trying to do this thing, I’ve used ChaptGPT, Claude, Perplexity, Carto, Tableau, DAll-E, and Giasai with varying success.
I started out comparing what ChatGPT, Claude, and Perplexity could do because all three are text-based. I didn’t need any geocoding, for example, only text-based descriptions of the data, which in my case were descriptions of land grant locations held by women in 1833 Texas.
Chatgpt gave me the best and closest thing to a map where I can visually see my data, though it couldn’t overlay it on a map from 1835. Here’s what I achieved with Chatgpt:
Very useful for me to visualize the data but not at all something I would include with an academic article or book.
Claude was NOT helpful. Here is its attempt, and this is after several prompts to regenerate. Claude simply could not grasp what I meant by a visual map of Texas:
Another Claude attempt. What the heck is this?
Perplexity couldn’t create a map BUT it gave me solid instructions on how to do it, which included taking the text descriptions of these locations and geocoding them into latitude and longitudinal points. I went back to Chatgpt and had it do that for me and tried again with Chaptgpt and Claude and got basically the same things. BUT perplextity sent me to some more sophisticated non-text map-making tools like Tableau and Carto. I fiddled with both, but they would require a lot of time just to learn to use and are not text oriented.
I tried Dall-E (backed by Chatgpt) and it gave me that crazy, inaccurate map at the top of this post. It’s lovely, appears to be “historical” in the sense that someone who glanced at it would think “old,” and is absolutely useless to a historian.
Finally I tried another Chatgpt backed program, Gisai.
Map? How about graph. And there’s no context as for location. I can see the grants relationally, but I can do that much better with the above Chatgpt map.
Here are my findings:
A text-based map creation service is not readily available to historians *today* but that will likely change very, very soon.
Chatgpt 4o did the best job of taking my text prompts and turning it into something visually useful, though not something I would include in an article or book.
Perplexity gave me the best instructions in terms of how I might go about actually creating a map for a book but it would take some time and effort for me to do it myself. Because I’m not needing a map like that today, I’ll wait a bit and see what develops in the next few months.
As always, the prompt makes all the difference.
I have attached below a copy of the prompts and output from Chaptgpt, Claude, and Perplexity. I think I should have begun by asking these AI bots what the *best* prompts would be to use if I wanted to create a visually accurate map of the data. I’ll try that next time as a starting point.
If you find this helpful, let me know in the comments. I’m thinking about doing more posts on researching in the history profession using AI. If that’s something of interest to you, again, let me know in the comments.