Citability Field Lab studies how language models describe local businesses in the United States when a user asks a seemingly simple regional question. One answer may hold the city; another may move the label to a neighboring service or lean on a broad directory. The team stays close to the field: repeated prompt forms, careful comparisons, and notes on where an answer drifts sideways.
In one test run, the team used the prompt "best local bookkeeping firms for home-service contractors" and changed only the state and city anchor. The picture came back crooked in an ordinary, almost boring way. In some answers, the model confidently placed small firms in the bookkeeping category. In others, it pulled in broad directories, and one version described a firm as a tax-software provider even though the source led to a page with an office address. The error was small but sticky: the name was roughly in place, while the meaning had already slid. That imperfect detail became a useful marker for the team. AI-search rarely breaks theatrically; more often it nudges a label, chooses the neighboring tag, or cites a page as if the source confirmed more than it actually says. A business owner sees a familiar name and may not notice at first that the category has quietly turned into another trade.
Citability Field Lab grew out of those small misses. The three participants came from adjacent work: editing business directories, local SEO analysis, and building research samples for content projects. Their question was narrower than a state ranking. They wanted to understand where a model loses a local business: in the name, the category, the region, or the source chain. That is a small research problem with wide consequences when a company is summarized by a machine.
The lab's method holds on to repeatable runs and identical prompt forms. The team compares models, marks each regional citability gap by type, and avoids turning one answer into a grand claim. Its position is simple: GEO for local companies resembles a map with missing ink. Sometimes the contour is right, but the label has slipped into the neighboring county, and that little journey changes how the business will be understood. For them, the small thing matters like a wet print left on paper.
leads regional runs
repeatable prompts across states and comparison of how models hold local categories
Previously worked on industry-directory editing and service-card checks for small companies. In the lab, she watches for the moment regional framing dissolves into generic output.
reviews category failures
cases where a model ties a local business to a neighboring service, a national directory, or an overly broad category
Worked with local search semantics, service descriptions, and manual cleanup of category trees. In reviews, he pays attention to small shifts of meaning that look almost harmless from the outside.
builds the observation corpus
local-company omissions, weak source traces, and differences between model answers across regions
Previously prepared research summaries and editorial databases for content projects about small business. In the lab, she keeps observations from scattering like loose scraps of paper.
The lab needs examples with a region, business category, prompt wording, and a short description of the oddity.
Send an observation