Summarize this article with AI
Send it straight to your assistant and get the short version first. The link travels with the prompt, so the model reads the real page.
Opens in a new tab with the prompt already written. Nothing is sent until you press enter.
The short answer
You measure it by asking. Twenty-five buyer questions, several repetitions each, across four engines, recording how often you are cited and which sources get used instead of you.
Step one: build the query set
Twenty-five questions a homeowner or developer would genuinely type before hiring in your trade. Not keywords. Questions, in their words.
- Who should I hire to build a custom home in Henderson?
- Best kitchen remodeler in Summerlin for a 1990s tract home
- How much does a home addition cost in Las Vegas?
- Which Las Vegas contractors handle tile roof replacement?
- Is [your company name] a good builder?
Include a few branded questions. The gap between how well you do on branded versus unbranded questions is one of the most diagnostic numbers in the whole exercise.
Step two: repeat, because one ask is an anecdote
Run each query at least five times per engine, ideally seven, with web search enabled. Retrieval is not deterministic and the same prompt returns different sources on different runs.
Step three: record properly
For every ask: the engine, the model, the date, whether you were cited, and every citation URL returned. The URLs are the part most people skip and the part that makes the whole thing auditable later.
Exclude asks that failed upstream rather than counting them as misses. A failed request is not evidence you were not cited, and folding it in quietly inflates the miss rate.
What an honest result looks like
A full run is twenty-five queries repeated seven times against each of four engines, and costs only a few dollars in API spend. Rates come back per engine with the denominator attached.
| What to report | What not to report |
|---|---|
| Cited in 18 of 140 asks, engine and date named | We rank well in AI search |
| One engine: 0 of 140, while citing other sources freely | That engine is broken |
| Denominator reduced by upstream failures, stated | A rounder number with the failures hidden |
| Top cited domains about us, ranked | We are the top source about ourselves |
The half everyone ignores
Your own rate is the less useful output. The more useful one is the ranked list of domains cited instead of you, because that tells you where the work actually is.
Questions builders actually ask
Can I do this without APIs?
Yes, manually, with a spreadsheet and patience. It is tedious but entirely doable and it will still tell you more than guessing.
How often should I re-measure?
Ninety days, with the identical method. More often than that and you are measuring noise rather than change.
What if the results embarrass me?
That is the normal first result and it is the point of a baseline. You cannot demonstrate improvement without a starting position you were willing to write down.
Share this article with AI
Finished it? Hand the whole thing to a model and ask what it means for your business specifically, including anything in it you think is wrong.
Opens in a new tab with the prompt already written. Nothing is sent until you press enter.
