Research · Methodology
How we measure
what changed.
A shared method for local search and AI visibility field reports. Each publication states its own dates, deviations, sources and limits.
01 · Source records
A number needs a traceable source.
For every published figure we keep its source, property or business, export date, measurement period, filters, calculation and reviewer. We maintain a dated log of website, profile, review and listing changes. Client permission determines which screenshots or extracts can be public.
Each field report identifies its first reliable measurement date. We mark earlier conditions unknown when analytics were installed after work began. We correct errors on the original page and note material revisions.
02 · Search visibility
Compare the same measurement.
Search Console observations specify the property, search type, dates, query rules and pages. We keep branded and unbranded query groups distinct. Its average position is an aggregate, not the exact position seen by a customer in one city. Query tables can omit anonymized searches, so summed query rows may differ from overall totals.
Where available, we record Google Business Profile performance separately from website search data. Profile calls and website clicks are interactions; they are not completed calls or qualified jobs. Manual local results record the query, date, location, device and result type. Localized and personalized results can differ.
03 · Traffic and leads
Actions and outcomes have different labels.
GA4 reporting states the property, acquisition dimension, period and event configuration. We distinguish a phone-link click, a verified form submission, a tracked completed call, an appointment and a won job. We check that events fire correctly before comparing periods. Referral sources from AI platforms describe visits the browser attributes to those sources; they do not count every time an AI system mentions a business.
When an owner reports more calls or estimates, we quote or summarize that as client-reported activity. We do not add those reports to analytics-confirmed totals or assign a channel without supporting tracking.
04 · AI visibility samples
Record answers, not a permanent rank.
Before each sampling round, we save the exact unbranded prompts and the service, area and intent they represent. For each answer we record the platform, date and time, available location and account conditions, whether the business appears, answer context, exposed citation URLs and an archived capture. We repeat the same prompts over time and report appearances as a count out of the tested set.
Google's dedicated generative AI performance report, when it contains data for a property, measures links shown in its AI search features. It is a different measure from our manual recommendation sample. We report it separately and do not infer which prompt caused an impression.
05 · Interpretation
Say what the data supports.
We separate observed changes from possible explanations. Short windows, seasonality, small samples, concurrent website changes, changing review counts and platform behavior can all affect results. An association between an optimization and an increase does not establish that the optimization caused it.
Burnworth Co. is the service provider in client studies, so readers should treat the commercial relationship as part of the context. A study can be useful without finding a positive result. Browse the research.