Buyer-intent prompts
Distinct local questions a customer might ask.
METHOD / DEFINED SCOPE
Measure a pattern.
Not a single answer.
PresenceCurrent repeats buyer-intent searches, saves the observations, then separates recurring appearances from one-offs. The result is a dated, bounded account of what was seen.
01 / WHAT WE MEASURE
One defined search set. Three passes.
Prompts reflect the business, its services, and location. Repeating each prompt helps distinguish an isolated answer from a pattern within the same tested scope.
Independent repetitions
Separate observations per prompt.
Planned observations
The canonical measurement set for one Full Audit.
A CONCRETE FIXTURE EXAMPLE
“best roofing in Raleigh”
In the local roofing fixture, this question is planned for three separate runs. Each saved result retains its prompt, run identity, mentioned businesses, and source references. Inspect the saved example ↗
02 / USABLE EVIDENCE
A failed attempt is not a negative result.
A usable observation needs a successful response and a recorded target-appearance decision. A successful absence counts as evidence; a failed, timed-out, skipped, or malformed attempt does not.
All 75 canonical observations must be usable before a Full Audit can be published. Validated analysis, actions, and publication artifacts are separate requirements. The 70/75 local example is therefore partial, not a completed Full Audit.
03 / FROM OBSERVATION TO FINDING
Trace the conclusion back.
Counts describe different things. We keep distinct prompts, individual observations, and citation records separate.
Save the response context
Prompt, repetition, status, mentioned businesses, and citation domains stay together.
Check prompt breadth
A business recurring across at least two distinct prompts is separate from repeated mentions within one prompt.
Propose a next step
Actions are hypotheses tied to evidence, not proven causes of an AI answer.
04 / LIMITS AND RECHECKS
A documented baseline, not a prediction.
Results belong to the tested prompts, provider, and date. A citation supplies context, not endorsement. The report cannot prove why a model mentioned a business, guarantee future recommendations, or represent every AI system.
A later user-started audit can create another saved snapshot for comparison. It does not by itself prove a website change caused a different AI response. Automated monitoring is not part of the current workflow.
See the beta packages ↗See the structure in practice.
Open the local, clearly partial measurement example and inspect individual observations.
Explore sample evidence ↗