Study design

Long Island AI Visibility Index

The study design, published before any data exists, so the method is fixed in advance.

Status

Methodology published. No data collected yet. Baseline collection has not begun. There are no results on this page and nothing here should be read as a finding.

The question

When somebody asks an AI assistant to recommend a local business on Long Island, which businesses does it name, and what does it cite when it names them?

This is asked constantly and answered mostly with speculation. The honest position is that nobody outside these companies knows the selection criteria, and the useful position is to measure the output rather than guess at the mechanism.

Design

A fixed prompt set across a fixed set of Long Island industries and geographies, run at fixed intervals across multiple assistants, recording which businesses are named, in what order, and which URLs are cited.

Industries will be chosen to span the range that matters locally: a home trade, a food business, a health practice and a professional service behave very differently in local search and there is no reason to assume they behave the same here.

Geographies will span a dense Nassau town, a dense Suffolk town, and somewhere on the East End where business density is much lower, because the interesting question is whether these systems have anything useful to say where the data is thin.

Every run records whether the assistant named any local business at all. We expect a meaningful proportion of runs to return no local business, and a study that only records hits would hide that.

What will be published, and what will not

Aggregate patterns will be published: how often local businesses are named at all, what kinds of source get cited, whether the same names recur, and how much answers vary between repeated runs of an identical prompt.

Named businesses will not be ranked or scored publicly. Publishing a league table of which local competitors an assistant prefers would turn a study into a weapon, and it would also be irresponsible given how unstable these answers are between runs.

No business will be contacted, and no result will be shared with a business as a sales prompt. If that changes the study becomes marketing and stops being research.

Honest position on feasibility

This is manual work. Each interval is a person running prompts and recording answers, so the realistic cadence is monthly or quarterly, not continuous, and the sample will be modest.

We are stating that here rather than describing an automated pipeline we do not have. A study design that overstates its own instrumentation is the first thing a careful reader should distrust.

What this study cannot tell you

Published alongside the method, not buried after the results, because a study that hides its limits is advertising.

  • No data has been collected. Everything on this page is design.

  • Manual collection limits sample size and frequency, and the sample will be small enough that individual results carry wide uncertainty.

  • These systems are non-deterministic and change without notice, so results are dated snapshots rather than stable measurements.

  • The publisher is a Long Island agency measuring a market it competes in. Aggregate-only publication and a fixed pre-published prompt set are the mitigations; they reduce the conflict rather than eliminate it.