How much does which source weigh?
Four methods, four figures, and they are published rather than hidden. That is not transparency for its own sake: a weighting you cannot see is one you cannot argue with — and then every number above it is a matter of trust instead of proof.
| Where it comes from | What it rests on |
|---|---|
| human | the owner confirmed it themselves, during intake |
| structured | it sits as schema on your own site — machine-readable and yours |
| thirdparty | an independent source says so: reviews, GBP, a directory, Wikidata |
| inferred | an AI engine said it. That counts as unproven until someone confirms it |
Why does an AI answer only count for 0.40?
Because it is not a source but an outcome. What an engine says about you is interesting — it is exactly what we measure — but it is not proof that it is true. Treating it as proof is how an error multiplies: the engine reads the error, repeats it, and the repetition becomes the new source.
What happens when two sources contradict each other?
Then the graph says which source claims what, and how heavily each of the two weighs. Nothing is decided in silence. A contradiction between your site and a directory is a fix on the directory; a contradiction between your site and an AI answer is usually a fix on your site, because the engine got that answer somewhere.
Frequently asked questions
Is this the same as structured data on my site?
Schema is one of the four methods, and a strong one — 0.80 — because it comes from you and is machine-readable. But it is not the graph itself. Schema says what you claim; the Evidence Graph puts alongside it what others claim and what AI makes of it, and keeps track of the gap.
Do I have to confirm everything before it works?
No. The graph works with what is there and states how much of it is proven. Confirming raises that share and changes which fixes come first — so it is not a condition but a lever, and the shortest one available.
What if an AI invents something about me?
Then it stands as a claim at 0.40 with the engine named, and the moment it contradicts a confirmed claim it becomes a conflict. Underneath sits the source the engine fetched, and that is where the correction belongs — not with the engine, which is only repeating what it read somewhere.
Does this stay private?
The graph is yours. What leaves it is what you publish: schema on your own site, a profile you update, a correction at a directory. Nothing is published on your behalf without approval, and every publication has a way back.
Read on
Free, no account and no card.