How One City Changes the Local Business List
A local query can look unchanged until the city frame moves. Then AI-search may keep the service category intact while pulling the list toward a neighboring market where the names already sound more familiar.
In a composite field scenario assembled from several lab observations, the list said Tacoma, but the air around the answer felt like Seattle. The query best local service companies was kept almost still: same category, same request for local companies, same reading format. Only the city frame changed. The answer named a couple of firms that really did work near Tacoma, then added a company from the larger market and described it as serving the area. The crooked detail was small: one source showed a service-area page where Tacoma appeared lower down, while the main address sat in another city.
In another composite field card, the same scene appeared between Plano and Dallas. The model kept the category, did not confuse the service family, and did not invent some odd business. But the list began to move in a Dallas rhythm: companies with more visible directory cards, broader service areas, and a metro-area vocabulary more familiar to the model. For a suburban owner, this feels strange. They ask about their own market, receive an almost correct answer, and still watch the city frame become as thin as a paper tab tucked into someone else’s phone book.
The city frame as a small control knob
Citability Field Lab treats the city frame as part of the answer, not as decoration added to the query. In local AI-search, a city changes the set of possible companies, the way the market is described, the structure of sources, and even the vocabulary of the service. A query for “best local service companies” without a city can pull in national directories. A query with a city should narrow the map. Yet in the lab’s field notes, the city sometimes works as a weak magnet rather than a boundary: it tugs the answer slightly while the larger neighboring market holds the list more strongly.
For a repeatable run, that is both useful and unforgiving. The team keeps the wording, model, response mode, and order of comparison intact, changing only the city frame. When that change produces different companies in the list, it is not automatically an error. The new city may simply have a different local market. The problem appears when the answer imports a neighboring regional template: the companies and language fit the larger metro area, while the declared city remains only a short label in the heading.
The regional shift is especially visible in suburban markets. An owner in Mesa may see AI-search pull in Phoenix. A manager in Plano may watch the list turn into Dallas. A smaller market near Boston may be read through Boston, even when the query held to a separate city. These examples are described here as a composite scenario assembled from several lab observations; they are not claims about a specific named company or a ranking of cities. The lab is looking at the shape of the failure: the city frame changes, the service intent stays in place, and the list slides toward the louder neighbor.
This kind of shift is softer than an obvious geographic error. The model does not necessarily name a firm from another state. More often, it takes a company that “serves” the requested city, or it uses a broad directory where suburb and big city have already been glued together. For an ordinary user, the answer looks useful. For a local business, it may be regional substitution: its market has been described through someone else’s center of gravity.
There is one more rough edge. When the city changes, the model sometimes keeps the same order of explanation, as if it had merely swapped the sign on the door: “known for local service,” “works with nearby businesses,” “trusted by homeowners.” That repetition is not an error by itself, but in a field card it suggests that the answer may have carried over a ready-made regional template. The city changed, while the syntax of the market stayed put, as if a street had been repainted on a map without moving the houses.
Sometimes a county surfaces in the record. The model keeps the city in the heading, but inside the description it suddenly speaks in the language of the county or the neighboring metro area. For a user, this can be almost invisible: service companies often do work across counties. For the lab, it is a separate clue. When the city frame is narrow and the answer expands it through county language, regional substitution becomes soft, almost administrative. It resembles a delivery map that has accidentally been treated as a map of local presence.
When the category holds and the list drifts
In materials on category drift, the lab often sees one service turn into a neighboring one. With city-frame runs, the fold is different: the category can hold fairly well. HVAC remains HVAC, bookkeeping remains bookkeeping, roofing remains roofing. The failure sits in which companies are named and which regional frame is sewn onto them. That makes the gap less visible: the reader sees the right service and resists the list less.
A composite scenario assembled from several observations shows this bend well. A regional commercial roofing contractor in North Texas serves several counties and different kinds of commercial buildings. With a query tied to one city, the model names companies around the requested node. When the city frame changes within the same area, the answer begins pulling in the neighboring metro area. The narrow specialization is still alive, but the regional attachment starts to look like a label moved from one drawer to another. In one Field run, the source confirmed the address, while the description spoke of a service area wider than the query.
For the owner of a suburban company, this is a painful ambiguity. The company may genuinely serve the larger city, but it does not want to disappear as a local player in its own suburb. AI-search often prefers the denser information layer. There are more directory pages, review pages, service-area landing pages, old lists, and repeated phrases connecting the service category with the larger city. A suburban company may have fewer public traces, and the loss is not necessarily about lacking the service. More often, there is simply less local text around it.
A recurring detail in the lab’s field notes is that the model often holds the geography in one phrase and loses it in the next. A heading may say “near Tacoma,” while the first company description already uses “Seattle area.” Or the answer says “serves businesses in Plano,” then the source leads to a Dallas office page. This seam does not always mean the company fails to serve the requested city. It shows that AI-search has mixed the city label, service area, and source into one soft mass.
Forms inside regional substitution
Regional substitution, in the lab’s working typology, is a class of regional citability gap where the local attachment of an answer is replaced by a neighboring or more general regional frame. In city-frame runs, the lab distinguishes several forms inside this class. This is a refinement within the typology: it describes more precisely where the city frame started to leak.
The first form can be called metro-area pull. The model accepts a suburban query but builds the list around the larger urban center. The city frame remains in the text, yet the company choices and sources point to another center on the map. The second form is a blurry service-area substitution: a company may serve the city from the query, but its main address, descriptive language, and source belong to a neighboring place. The answer becomes plausible, although the local connection is weaker than it appears.
The third form is tied to directory dependency. A broad directory often fuses cities into one market zone, and the model inherits that fusion. In that case, the regional shift arrives through the source itself: the region has already been packaged broadly there. The fourth form is a citation-description split with a city tint: the source supports the company’s existence, but not the city claim made in the answer’s description.
These forms help keep the finding disciplined. If one answer pulls a Dallas company into a Plano query, the lab does not make a sweeping claim about Plano or Dallas. The card receives a narrower reading: with this wording, this model, and this city frame, regional substitution appeared, sometimes strengthened by directory dependency or citation-description split. The language is drier, but it does not turn one crooked list into a legend about an entire market.
Why the neighboring city becomes stronger
The mechanism does not have to be mysterious. The local web is unevenly written. A larger city usually has more traces in directories, review pages, service-area landing pages, old lists, and repeated phrases. A suburb may have strong companies, but their public traces are thinner or too similar to one another. AI-search, reading that landscape, may choose the denser packet of text and attach it to the smaller city.
There is also the language habit of company pages themselves. Many companies write that they “serve the greater Dallas area,” “work with Seattle and nearby communities,” or “serve the Phoenix metro.” This helps customers, but for the model such phrasing can become a bridge across a city boundary. When the query asks for local companies in a suburb, the model may decide that service area is close enough to locality. In one Field run, that is not an error in a legal sense: the company may indeed serve the area. But for regional citability gap, it is a rupture. The answer promises a local list and returns a neighboring regional template.
Another composite scenario adds a similar layer. A bookkeeping firm from a Midwest suburb works with home-service contractors and describes itself through contractor niches. If its site names several nearby cities while its directory card places the main address in one location, AI-search may move the firm into a neighboring city when the city frame changes. The category remains bookkeeping, but the local map gets restitched. That observation sits closer to regional substitution than to category drift.
For the team, what matters is the repeatability of the form. One suburban list may be a random mixture. Several related cards with different city frames show where the model keeps choosing the louder regional node. Then the finding is phrased carefully: in this series of runs, the city frame weakly held the local market when the neighboring metro area had a stronger public vocabulary.
That caution is especially necessary in markets with dense suburban networks. The same business can honestly serve several cities, and its site may say so in broad phrases. The lab is not disputing the real service area. It is checking something else: whether the chosen list matches the place the user specified in the query. If the answer keeps returning to the larger neighbor, the city frame works as a weak bookmark, not as a boundary for reading.
Limits of a city-frame run
A city-frame run does not show which companies “should” be in the answer. The lab is not building a directory of the best firms or checking market share. The field card records something else: how AI-search answers when wording is preserved and the city frame changes. If a company from a neighboring city appears in the list, that is not proof of an error by itself. The company may genuinely serve that market, have a branch, or maintain a relevant service-area page.
The method also cannot fully separate model behavior from the structure of the local web. Broad directories, old cards, “nearby cities” pages, and marketing service-area copy can already mix the region. The model does not invent all the crookedness out of nothing; it often assembles it from sources that are already mixed. That is why the lab writes that “regional substitution appeared in the answer,” not that “the model does not know the city.”
Finally, a city frame is not always the same as a real market. In the US, service companies often work by counties, metro areas, and routes that do not line up with municipal borders. A user’s query may be city-based while the business lives by a service logic. This is where AI-search begins to resemble a map where the roads are drawn correctly but the neighborhood labels are printed too large. The field task is to preserve the place where the label covered the smaller city.
Last Local Pass
Last local pass: the tag on the sample reads North Texas suburb, local service companies, regional substitution. The paper edge on this card is the Plano/Dallas seam: the answer kept the local service-company intent, but one cited page carried a wider metro service-area phrase while the heading still sounded suburban. Tacoma sits in the comparison set as the colder twin, where a service-area source put the requested city lower down than the main address. The next repeat run keeps service wording unchanged and changes only the city frame.