Proof
We probe the major language models with a battery of canary queries about .. us — recognition, category, attribution, fact-checking — and scoring every answer. Below is the live result. And if you expected a 100% score ... that's not realistic, now is it!
Measured 5 Oct 2026 · 10 days ago
Movement vs previous run · no previous run yet
Circle of control · self 2% 2/90
Do the models know you by name, get the facts about you right, and credit your work to you?
Circle of influence · concepts 27% 20/73
Do the models use your published terminology as you meant it, and treat your framework as yours?
Circle of concern · category 9% 10/112
When the category is asked about, do the models bring you up alongside the alternatives?
Composite score 14/100
Weighted average across all axes. A zoom-out of the three spheres; the sparkline shows this composite over time.
275 scored responses · 40 active canary queries · 0 source citations
Across languages
We run the canary set in each market’s language. The same entity, probed in English and Dutch — and the models don’t always answer the same.
English ●
13%
17/135 recognised
Dutch
11%
15/140 recognised
Tested across dimensions
Each canary query targets a distinct way a model can get you right or wrong. The breakdown per dimension:
| Dimension | Responses | Recognised | Partial | Not | Cites source |
|---|---|---|---|---|---|
| Circle of control · self 2/90 | 2% | ||||
| Disambiguation Does the model distinguish between categories or similar entities? | 2 | 2 (100%) | 0 | 0 | 0 |
| Attribution Does the model credit terms, methods and citations to the correct entity? | 40 | 0 (0%) | 0 | 40 | 0 |
| Brand recognition Does the model know the entity by name and can it place it? | 24 | 0 (0%) | 0 | 24 | 0 |
| Fact accuracy Is a specific factual claim the model makes about the entity correct? | 24 | 0 (0%) | 0 | 24 | 0 |
| Circle of influence · concepts 20/73 | 27% | ||||
| Fair criteria Does the model use criteria that align with the entity's own positioning? | 21 | 20 (95%) | 1 | 0 | 0 |
| Term definition Does the model define a coined term the way its author meant it? | 52 | 0 (0%) | 2 | 50 | 0 |
| Circle of concern · category 10/112 | 9% | ||||
| Comparison How does the model position the entity against alternatives? | 24 | 9 (38%) | 7 | 8 | 0 |
| Relationships Does the model know the relationships between entities (person↔practice, org↔parent)? | 23 | 1 (4%) | 2 | 20 | 0 |
| Category presence Does the entity come up when the category or problem is asked about? | 65 | 0 (0%) | 0 | 65 | 0 |
Where AI finds the evidence
We ask the models who I am, why you would trust me, pay me and choose me. Each question twice: from memory and with web search. This is where the evidence in their answers comes from.
my own sitemy own profileswritten by me elsewhereothers
Share of the sources the answers with web search cite; every model counts once.
My own site has its largest share at “Why pay me”: 75% of the sources.
Others have their largest share at “Who I am”: 24% of the sources.
Share of the evidence from my own channels, site and profiles: ChatGPT 94% · Gemini 48%.
Memory against the web
In 34% of the questions, a model’s memory gives a different picture of me than the web does today.
samediffersmemory doesn’t know
Clients get the full map: which pages, which sites, and where the memory lags behind.
Method
We probe the major language models — ChatGPT, Gemini, Claude, Llama and others — with canary queries: the questions a real user would ask about the brand, its category, its claims and its relationships. Every answer is scored on whether the entity is recognised, described correctly, and whether the canonical source is cited.
The figures are the latest scored answer per query and provider, refreshed continuously. They measure association — whether the models carry the right story — not traffic.
A canary result is a strong signal, not proof of causation: it shows what the models currently say, not exactly why. Numbers update as new responses come in.
Last response: 2026-10-05 12:47
It starts with an evidence architecture, anchored by a True Root. Check whether your site qualifies.