We gave ourselves a Wikidata item (and closed the loop)
An earlier entry taught our site who we are with a JSON-LD graph. This one is about teaching the rest of the web — the off-site corroboration that actually moves the needle for AI citations. The plain-English what-and-why is in our article, Get cited by AI search; this is the implementation detail.
The goal: make the same entity verifiable from every direction — our site, an open knowledge graph, and a founder entity, all pointing at each other.
Scripting the Wikidata writes
Editing Wikidata is an authenticated write to a public, permanent graph — so we didn't want to
hand-click it, and we didn't want a brittle one-off. We wrote two small Node scripts (in docs/wikidata/)
that log in with a bot password and call the Wikibase wbeditentity API. Both default to a
dry run — they print the exact payload and post nothing until you pass --commit.
add-statements.mjs— added 10 sourced statements to the org item (instance of, industry, country, HQ, official website, Crunchbase / LinkedIn / GitHub IDs) plus the Indonesian label.create-harry.mjs— created the founder's Person item and setfounded byon the org, in one run.
Wiring it back, both ways
The off-site profiles only count if our site points back at them. We made ENTITY_PROFILES the single
source for the corroboration anchors — Crunchbase, Wikidata, Clutch — so orgGraph() builds the
Organization sameAs from it and the /proof page renders the same list. Then we added the founder's
Wikidata Q to his Person sameAs too. Now the org, the founder and the site all agree, in both
directions.
Do
- Dry-run the writes first; source every statement
- Single-source the profile list (JSON-LD + page can't drift)
- Make every off-site link reciprocal
Don't
- Commit the bot password, or any credential
- Fabricate facts to look bigger than you are
- Leave a profile pointing at you with nothing pointing back
We also surfaced the whole set on /proof — a "Verified across the web" row — so humans see the corroboration too, not just crawlers.
The payoff
Across these moves, our internal Citation Possibility Score went from ~46% to ~60%. The mechanism is simple, and it's the whole point of the work:
Corroboration turns you into a trusted source — and trusted sources get cited with higher probability. You stop being a string the model isn't sure about, and become a thing it can name.
What's left is off-site and not ours to code: Clutch reviews, and — eventually — Wikipedia.
Sources