Aquariums List is one of the directories we run ourselves on Directories.ai. On October 3 it went from empty to a researched list of every open public aquarium in the US that we could verify. An AI assistant did the research and the data entry through our MCP server, and a person reviewed the result. It took one afternoon. Today 220 aquariums are published at directories.ai/aquariumslist.
This post walks through how it went, including the parts that didn't go to plan.
Step 1: Pick a niche you can actually finish
Aquariums are a good first directory because the universe is small and well defined. Our first estimate was about 250 public aquariums in the US. The research showed the real number of open ones is closer to 200. Several well-known names have closed, including Miami Seaquarium, Gulf World, SeaQuest Fort Worth and SeaQuest Roseville, and the Children's Aquarium at Fair Park. A directory that lists closed places loses visitors' trust on the first click, so every closed aquarium was left out.
Step 2: Set up categories before importing
Before any listings went in, the assistant created six type categories. They run from aquariums, zoo aquariums and marine parks to marine science centers, science museums, and nature centers and hatcheries. Categories become your browse pages and your search landing pages, so it's worth getting them right before the data arrives.
Step 3: Research from official sources, never guess
No Google Maps import was connected for this directory, so the assistant researched state by state from each aquarium's own website or another official source. The rule was simple: a phone number or email goes in only if a source shows it. Otherwise the field stays blank. A blank field can be filled later. A wrong phone number makes the directory look careless.
The research covered all 50 states and DC. West Virginia and Wyoming have no public aquarium. DC's National Aquarium closed in 2013. That leaves 48 states with at least one listing, led by Florida, California and Texas.
Step 4: Import as drafts, then check quality
Everything went in as drafts, using ingest_listings for researched records and bulk_create_listings for chain locations (more on that below). Then two checks ran before anything was published:
score_listings rated each listing for completeness. Most of the weaker ones just had short descriptions, which is an easy fix.find_duplicate_listings looked for the same aquarium entered twice. It found no real duplicates.Anything uncertain went into a review sheet for a person to check. Examples are a phone number that only appeared on a review site, seasonal hours, and a facility closed for renovation.
The bug we found along the way
The import tool treated any two listings on the same website as the same business. That's usually right, but it silently dropped chain locations. SeaWorld Orlando and SeaWorld San Antonio share a domain, and so do the SEA LIFE and Ripley's aquariums. We added those 30 chain locations a different way at the time, then fixed the import. A shared website or phone number now only counts as a duplicate when the two businesses are in the same city, ZIP and state. Building our own directories on the platform is how we find bugs like this before customers do.
What the assistant did, and what the person did
The assistant:
The person:
That split is the point of Directories.ai. The repetitive work runs through your own AI assistant, whether that's Claude, ChatGPT, Gemini or a local model, and the decisions stay with you.
Try it on your own niche
If you have a niche in mind, connect your AI assistant to https://directories.ai/mcp and ask it to plan the categories and research the first state. You can see the finished directory at directories.ai/aquariumslist, and the Claude conversation that published it on the live demo.