I came in as a developer, so I thought I knew what I was doing. I was confused for a long time,
asking the same questions again and again.
A mentor would tell me to fix this, then fix that, and I'd fix it. But the thing that really cost me
time was my own habit: leaning on AI before I understood Geo myself.
You search a name, you don't see it, you create it. Now the graph has two of them.
This is not rare. Some well-known things exist three times over, and every copy is in use — one of
them with more than a thousand links pointing at it. Topics are the easiest to double up, because
the same idea gets written a slightly different way each time.
Search before you create. Search the name on its own, not the name plus the type you have in
mind, because the one that exists may be filed differently than you expect. If you find more than
one, use the one with more links.
I kept being told "the ontology has to be right" and I had no idea what that meant.
It just means: the space has a list of the kinds of things it keeps, and how they connect. A Person
works at a Company. A Claim has Sources. That's it.
Two things I got wrong for weeks:
A thing can be several kinds at once. I kept asking whether something was a project, an
exchange or a company. It can be all three, and carry all three.
The kind decides what the thing can hold. Give it the wrong kind and the fields you wanted
aren't there, so the page looks empty even though you filled it in.
The move that made Geo click for me, and it takes two minutes:
Open an entity like the one you want to add, and copy its shape. Look at the kinds it carries
and the properties it fills in, then match them. Yours will fit everything around it instead of
being a lonely one-off.
Searching a type by name doesn't teach you this. One real example does.
I'm a developer and I still got this wrong. I handed the AI a task before I understood the job, and
it confidently did the wrong thing, fast.
Your AI can only do what you already understand well enough to ask for. So:
Look around Geo yourself first. Open real pages. It is not wasted time; it is the thing that makes
your instructions worth following.
Tell it to check Geo first and show you what already exists, before it creates anything.
Make it show you what it will publish, and the id of whatever it says it found.
Ask it why it chose a kind. If the answer is a guess, it is a guess.
AI makes you faster at whatever you already are. If you're guessing, it makes you guess faster.
The main spaces are Crypto, AI, Health, World affairs and US Politics. Every one of them holds tens
of thousands of entities, so none of them is short of work, and none of them is "the easy one".
So don't pick by size. Pick the subject you already follow, because you can only spot a wrong or
missing fact in something you already understand. On a subject you don't know, you can't tell your
good edit from your bad one, and you'll wait on other people to tell you which it was.
If nothing pulls you, start in the Geo Curator Program space, where the bounties say plainly what is
needed.
Start by playing in your own space, or a datasets space where nothing is at stake. Add a few things,
organise them, and see how pages, entities, collections, query blocks, data blocks and tabs actually
behave. It costs you an afternoon and it removes most of the confusion.
Then take what you learned to a main space. That's when a small, correct edit is worth making.
Everything big can wait: large imports, elaborate pages, your own perfect structure. Ten correct
entities teach you more than one big upload.
No editor minds a small edit. Everybody notices a duplicate or a mistake.