The KEO playbook
How we build the verified, canonical, readable, citable record AI models learn from — the work behind Knowledge Engine Optimization.
First, we read you the way a model does
Before we change anything, we audit what the web already says about your business as an entity: what's accurate, what's wrong, what's missing, and where sources contradict each other. A model can only learn a clean story if a clean story exists to be read — so we start by finding where yours is muddy.
Build one clean entity (readable)
We give your business a single, machine-readable identity: a structured JSON-LD @graph with one stable @id per entity (your organization, its people, its products), so a crawler parses who-is-who without guessing. One entity, one description, repeated consistently everywhere a model might read it — not five versions that contradict.
Anchor it to public knowledge (canonical, verifiable)
We anchor that entity to the sources knowledge graphs and training sets trust: a sourced Wikidata item with real completeness (what it is, its field, where it operates, official site, external identifiers), consistent sameAs links across your authoritative profiles, and clean, consistent details across the web. The aim is that whatever a model cross-checks resolves back to the same canonical you.
Earn authority, never fake it
Some of the highest-weight surfaces are notability-gated — Wikipedia, for one — and they're earned through real, independent coverage, not self-published. We never plant fake mentions, never manufacture consensus, and never astroturf. We do this with you, in the open: every change is one you can see and verify. It's slower than faking it; it's the only version that survives scrutiny — and the only version worth a model's trust.
What this playbook doesn't promise
No one controls what an AI lab puts into its training data, or when. We can't promise your record lands in a specific model on a specific date. What we can do is make your business the best version available to be learned — verified, canonical, readable, citable, consistent — and report honestly whether the machine's memory of you is moving. Signals with a range, not guarantees.