Ghosts of Baltimore is a sister site to Ghosts of DC. All three Ghosts sites run on the same research process, so one AI policy covers them all.
This page explains how AI fits into the work behind the Ghosts sites. For who I am and why this started, see About.
Why I use AI
It comes down to time. Pulling a good story out of the archives means hours of reading old newspaper pages, tracking down deeds and land surveys, and lining all of it up against census records and maps. Those weekends are hard to come by now, and for a while that meant a lot of good stories just sat there.
AI changed that for me. It helps me search the archives, pull the records, and put a draft together much faster than I could on my own. A story that used to eat a whole weekend can come together in an afternoon now.
These tools are changing fast, and I want to be clear about what they are. They’re tools, and they’re only as good as the person using them. Over the past couple of years I’ve built a set of workflows, really an engine, that pairs a person with AI to find these stories, check them, and tell them. I keep tuning it as the tools get better.
I believe in what that makes possible. Most of the stories on this site wouldn’t exist without it. Nobody else was going to write them, and without people using these tools to bring them to the surface, they’d likely stay buried in the archives. The goal hasn’t changed since 2012. Dig up lost and untold local history, starting in Washington, DC, and back it with sources you can check yourself.
What this isn’t
When people hear “AI” and “blog” together, they usually picture someone typing “write me a blog post about the Hawaiian prince who served in Congress” into ChatGPT and hitting publish. That’s not what’s happening here. A chatbot working from memory will hand you a smooth, confident story with the dates wrong and the quotes made up, which is the opposite of what this site is for.
What I’ve built is closer to a research engine, and it runs on a pretty rigid process. Its job is to find stories in the original records and then show, as best I can, that what we’re telling you is authentic, accurate, and traceable to documents anyone can look up.
The standards I work to
I won’t walk through the whole process here, but these are the standards it’s built around:
Original records first. The research starts with old newspapers, deeds and land records, surveys and plats, maps, census records and government documents.
Facts tied to documents. The facts in a post should trace back to a specific document, and quotes come from the original pages rather than somebody else’s retelling.
Conflicts noted. When sources disagree, I note it and go with the stronger one.
Unproven claims labeled or cut. If something can’t be backed up, it comes out or gets labeled as uncertain. When an old newspaper accuses someone of something, we say which paper said it instead of stating it as fact.
No easy superlatives. I try to steer clear of phrases like “the most notorious” or “the first,” because they’re usually wrong.
Credited images. Images come from public archives and fellow history sites, credited and linked in the caption.
Checked before publishing. Posts get checked against their sources before they go live, and then I read them.
Sources you can see. Research posts end with a list of sources linking back to the originals, so you can see where everything came from.
Where the history comes from
Most of what you read here comes from public records and public domain collections: the Library of Congress and its digitized newspapers and maps, the National Archives, the Smithsonian, local libraries and historical societies, state and county land records, census records, historical newspaper archives, the Internet Archive, and Wikimedia Commons.
Who does what
AI helps with the research, the drafting and editing, the images, and the behind-the-scenes work of getting a post ready. It also helps with the newsletter, social posts and emails, and it voices the podcast episodes I’m experimenting with. Most of this runs on Claude, made by Anthropic. It also helped build The Atlas, our map of stories: AI read each post and pulled out the addresses and places it names, and those were matched to official address records where possible. The pins it couldn’t match exactly are flagged, and I review them.
I pick the stories, decide where to look, check the facts against the sources, and read the posts before they go live.
Images, and an honest note about the experiment
These sites have been an experiment from the start, and I’ve made some calls along the way that I might make differently today. For a while I created images for stories where no historical photo existed, colorized black-and-white photos, and animated some stills. At the time I went with what seemed acceptable to most readers.
These days I work hard to use real archival images, credited and linked back to where they came from. When an image has been generated or altered, I do my best to say so in the caption.
Older posts
Some older posts on Ghosts of DC were edited with AI tools back in 2022 and 2023, mostly their titles and search descriptions. I’m working my way back through them.
Corrections and reader contributions
Reader corrections and additional sources help improve the stories on our sites. If you spot a possible error or have a record that adds context, use the “Suggest a correction” email link at the bottom of a post, or email us. Include the article link, the detail in question, and any supporting source.
These submissions can inform further research, corrections or clarifications. A request does not guarantee a reply or a change to a post; any revision depends on the available evidence.
