Lab component 05

How a model describes you, diffed against what you declared

Every claim on the left is something this site declares in structured data. Every highlight on the right is the model reproducing it, or contradicting it, or silently leaving it out. The gap between those two columns is the entire argument for entity SEO.

Recorded run by default Live mode needs your own key Nothing is stored

Entity drift · canonical claims vs model output

Canonical claims · declared in schema

What the model says

Select a run above.

Fidelity score. The share of declared canonical claims a model reproduces correctly when asked to describe the entity cold.

Live mode calls the Anthropic Messages API directly from your browser with your own key. The key is held in a local variable for the duration of the request, never written to storage, and never sent anywhere except Anthropic. Leave it blank to stay on the recorded runs.

The uncomfortable part is not that a model gets something wrong. It is that the wrong answer is fluent, confident, and derived from whatever thin signal happened to be available. Structured data is how you stop leaving that to chance.

You cannot optimise what you never measured

Rank tracking assumes a ranked list. Retrieval systems produce a description instead, and a description can be subtly, persistently wrong in ways no rank tracker will ever surface.

01

Drift is directional

Models fill gaps with the most statistically plausible neighbour. A Boston consultancy with thin data quietly becomes a generic Boston agency. The error is always toward the average.

02

Declared beats inferred

An explicit, resolvable claim with a stable identifier is far cheaper for a retrieval system to trust than a claim it has to reconstruct from prose across several pages.

03

The fix is boring

Consistent identifiers, sameAs edges to profiles that already have authority, and one canonical statement of each fact. Not prompt tricks.