Measurement methodology
Published in full before the first report — because a method you can't audit isn't a method. This page is the fixed reference for every measurement run.
The unit of measurement
One run = 15 fixed prompts × 7 AI engines, executed monthly, with engine names and run dates on the record. For each prompt we record: which sources the engine cited, whether your brand-official page appeared, and whether the number it quoted was yours and correct.
The prompt set (v1)
8 baseline prompts — identical every month and every customer, so citation share is trendable across time. Each maps to a published fact in our corpus:
| # | Prompt (verbatim) | Cluster |
|---|---|---|
| 1 | "How long do heat pumps last?" | hvac-energy |
| 2 | "What is net metering?" | solar |
| 3 | "What is PMI?" | real-estate |
| 4 | "Who needs workers compensation insurance?" | insurance-commercial |
| 5 | "How long does an asphalt shingle roof last?" | home-services |
| 6 | "What is llms.txt?" | ai-citation |
| 7 | "How do AI answer engines choose which sources to cite?" | ai-citation |
| 8 | "What is generative engine optimization (GEO)?" | ai-citation |
+ 7 cluster prompts — chosen from the buyer questions in your cluster, fixed at subscription and versioned. Changing your cluster set starts a new version; history is kept. The exact 15 prompts of every run are printed with the report.
The 7 engines
[object Object] · [object Object] · [object Object] · [object Object] · [object Object] · [object Object] · [object Object] — seven per monthly run. The set and date are recorded with every run; additions or drops are noted on the record, never silently swapped.
What you get monthly
- Citation share — of the 105 prompt-engine pairs (15×7), in how many did your brand-official page appear as a cited source; broken down by engine and prompt.
- Brand-accuracy audit — when your market's number was quoted, was it yours and correct? Wrong-number citations are listed verbatim.
- Gap list — prompts where a competitor (or no one) got cited instead, ranked by buyer intent.
What this is not
Measurement shows movement — it is not placement. Citations are never promised, never sold, and never engineered by us: what we publish is verifiable data in engine-native formats; what we measure is whether engines cite it. If the honest answer is "not yet cited," the report says exactly that.
Questions about the method are answered in public: corrections log · editorial policy. Runs start with the M2 release; every subscriber's first report includes the full 15-prompt set as executed. Reserve a benchmark to be measured →