Brand or branch: which one AI assistants actually name in local answers
Local AI visibility is not a single brand-versus-branch score. Measure what the answer names, at what level, and under which location conditions.
A local answer can name a chain, a specific outlet, both, or neither. Those are different outcomes. If a report collapses them into “brand mentioned,” it may flatter national awareness while hiding whether a customer can identify the nearby location. If it counts only branch names, it can miss a useful brand-level recommendation that sends the reader toward a locator or booking flow.
The practical question is not which level assistants prefer in general. It is which level appears for each kind of local intent, in a specified market and test setup. Treat brand and branch as separate units of observation, then report how often each is named and whether the answer makes the connection between them clear.
First define what counts as a name
A brand mention is an explicit parent or trading brand name. A branch mention is an identifiable local outlet: its location-specific name, address, neighborhood, or another unambiguous identifier. In this hypothetical example, “There’s a Citelytics clinic nearby” is not a branch mention unless the answer identifies which clinic. “Citelytics on King Street” is. A linked result can support identification, but record it separately from text in the answer.
Use mutually intelligible labels rather than a single yes/no field. For each response, code: brand only; branch only; both; neither; and ambiguous. Add a separate field for whether the branch is correctly matched to the requested area. This prevents a chain name from being treated as evidence that the assistant selected the right outlet.
Why the query changes the answer
Local intent varies. “What does Brand X offer?” invites a corporate answer. “Is there a Brand X near Northgate?” tests the connection between the brand and a local outlet. “Find a late-opening pharmacy near Northgate” is category-led and may produce competitors without naming the requested brand at all. Don’t mix these into one denominator and call the result local visibility.
OpenAI’s ChatGPT Search help page describes location-related features, including approximate location based on IP address and optional device location sharing. It also advises users to specify a location for more relevant local results. Google describes local results in terms of relevance, distance, and prominence. These are platform-specific statements, not proof that every assistant uses the same signals. For measurement, make location inputs explicit and keep them consistent.
Build a comparison that can be defended
As a methodological choice, you can use paired prompts: keep the task constant and change only the location or brand context. For example, compare “Which Brand X locations offer weekend appointments in Northgate?” with the same question for Southbank. Include a separate category-led set such as “Which providers near Northgate offer weekend appointments?” The first tests brand-to-branch resolution; the second tests whether the brand is surfaced in a local choice set.
As part of your test method, set the assistant and version or product mode, date, language, market, location context, browsing or search state where controllable, and exact prompt. Use a fixed set of prompts across locations. Don’t silently retry until a preferred result appears. If responses vary, preserve every run and report the variation rather than selecting the best answer.
Choose a sample size suited to the claim. A small diagnostic set can reveal examples and coding problems, but it cannot establish a stable network-wide rate. For a directional benchmark, consider including multiple locations, intent types, and repeated runs; publish the counts alongside percentages. A useful results table shows eligible responses, number of branch mentions, number of brand mentions, both, ambiguous cases, and correct-location matches. State the sampling frame: which branches were selected, why, and whether the sample represents all markets or only priority markets.
Separate naming from usefulness
A branch name is not automatically a good answer. It may be outside the requested area, closed, or irrelevant to the stated need. Likewise, a brand-only answer may still be useful if it provides a working location link, but that is not equivalent to naming the branch. Keep distinct measures for textual naming, correct local match, and useful next step. An answer can score on one and fail on another.
You can audit coding with a second reviewer on a subset, especially where outlet names resemble the parent brand or where citations point to a branch not named in the prose. Resolve disagreements using written rules, and retain the original response and source links. That makes later comparisons auditable when the model, interface, or local information changes.
What the result can—and cannot—support
A measured rate supports a bounded statement: in this prompt set, with these locations, settings, and dates, the assistant named branches at a particular frequency. It does not establish what every user sees, prove that the assistant’s answer caused visits, or guarantee future recommendations. Nor does a brand mention establish that the right branch was selected.
The operational output is a gap map, not a promise: where the chain is named without an outlet; where an outlet appears but is mislocated; where category prompts omit the brand; and where responses do not identify a useful next step. Those distinctions tell brand, local SEO, and analytics teams what to investigate—and give reviewers a result they can reproduce and challenge.
Sources
Measuring a brand or a network?
Every scan keeps its raw answers and sources, so a number can be re-checked months later instead of taken on trust.
See plans