How to Measure AI Search Visibility in a New Market

To measure AI search visibility in a new market, set that city, region, or country up as its own tracked location so its results never average into the home market, then rebuild three things inside that scope; the competitor list, because local rivals are usually different, the prompt set, because people phrase questions with local words, and the goal list, because local press and directories carry more weight there than national outlets. Run 15 to 25 prompts nightly across ChatGPT, Gemini, Claude, Perplexity, and Google AI Overviews for that market alone, and read its share of voice on its own line next to the home market. The gap is usually wide; a brand sitting at 31% share of voice nationally often lands in single digits in a market it entered last quarter. Expect 8 to 12 weeks before the numbers move, since a new market starts with almost no local sources naming the brand, and those sources have to exist before AI can repeat them.

Here's the conversation that starts this workflow. A client is opening in a new city or a new country, the launch budget is approved, and somebody asks what AI says about them there. The agency pulls the visibility dashboard and reads out a number that has almost nothing to do with the new market.

Why a blended number hides a new market

Brand-level AI visibility is an average. It's built from prompts that mostly reflect where the brand already has customers, coverage, and reviews. A regional chain that owns its home city can look strong at 31% share of voice and still be invisible three states over, because the home market is doing all the lifting.

That gap matters because AI answers with local intent really are different by place. When someone asks for the best option near them, the assistant runs web searches, and those searches pull local sources; city publications, regional directories, review sites with local density, "best in town" roundups. A brand with none of those doesn't appear, no matter how well known it is nationally.

So expansion measurement has one requirement above all others. The new market has to be its own line in the report. Once it is, three questions get answers instead of opinions. Where do we stand there today, who is AI naming instead of us, and is the local work moving anything.

The rest of this guide is the setup that makes those three answerable. It works with any AI visibility setup; the callouts show how each step runs in Spyglasses.

Step	What you're doing	What tells you it's working
1. Scope the market	Adding the market as its own tracked location	A separate results line from day one
2. Localize competitors	Replacing national rivals with local ones	Names you didn't expect showing up
3. Localize prompts	Rewriting inherited prompts in local language	A prompt set a local buyer would recognize
4. Set local goals	Naming the publishers and pages that matter there	Targets scored for AI value, not just reach
5. Read it separately	Charting the new market against the home one	A gap you can watch close

Set the new market up as its own scope

Start by deciding what the market actually is. A city, a metro area, a region, or a country. Pick the smallest unit the client's business really works in, because the smaller the unit, the more local the AI answers are, and the more useful the measurement gets.

A country-level scope is right for an international launch. A city or metro scope is right for anything with a physical footprint, a service area, or local competitors. Splitting one country into three cities gives you three clear reads; rolling three cities into "the Midwest" usually gives you one blurry one.

Then add it as a scope that sits under the existing brand rather than as a brand new setup. That inheritance saves real time. The brand description, the product framing, and the prompt structure all carry over, so you're editing rather than starting from an empty screen.

One thing worth deciding up front. Every tracked prompt in the new market runs on its own schedule and costs on its own, so a client with five target markets is committing to five times the prompt volume. Talk about that at setup, not at the invoice. Agencies running several markets per client usually cap each one at 15 to 20 prompts and spend the effort on picking the right ones.

Rebuild the competitor set for that market

The competitor list from head office is usually wrong for a new market, and it's wrong in a specific way. It lists the brands the client competes with on paper, and it misses the local operator that has run the same three neighborhoods for fifteen years and owns every local roundup.

Do this in two passes.

First, cut. Go through the inherited list and remove anyone who doesn't operate in the new market. A national rival that hasn't opened there yet isn't competition for a local question, and leaving them in makes your share of voice look worse than it is.

Second, add. The best source for local competitors isn't the client's opinion; it's the answers themselves. Run a handful of the local questions manually and write down every brand named. You'll usually find two or three you'd never have listed, and those are the ones the expansion is actually fighting for space with.

Keep aliases in mind here. Local businesses get referred to by short names, nicknames, and old names that stuck after a rebrand. If the tracking only matches the legal name, the local rival will look smaller than it is. The same logic applies when you're deliberately going after one of them, which displacing a competitor in AI answers covers step by step.

Rewrite the prompts so they sound local

Prompts inherited from the parent brand arrive with the geography added, and that's a starting point rather than a finished set. Read every one as if you were a buyer in that market. Some will be fine, some will read like a national question with a city glued on the end, and a few will make no sense at all.

Three kinds of edits do most of the work.

• Reword for local phrasing. People search with the words their market uses. A neighborhood name, a regional term for the product, a landmark instead of a zip code. Prompts that match how buyers really ask get matched to the pages that answer that way.
• Cut what doesn't apply. National comparison prompts, prompts about products that aren't sold there, prompts about services that launch next year. Each one you drop is a nightly run you spend somewhere better.
• Add the local-only questions. Every market has a few. Availability, delivery radius, local regulation, whether you're open on a day that matters there. These are often where a new entrant can win first, because national brands never answer them.

For a new country, add language to the list. If buyers ask in the local language, track prompts in that language, not translated English. The sources AI retrieves for a Spanish-language question are different sources, and measuring the English version tells you about a market that isn't the one you entered.

Tag the whole set with the market name. That tag is what makes filtered client reporting possible later without re-deriving which prompts belong to which market every month.

Pick local goals worth measuring

A new market has a shorter list of things that would actually change the numbers, and the list looks different from the national one.

Local publishers usually beat national ones here. A city business journal, a regional trade site, or a well-read local blog gets pulled into local answers far more often than a national outlet does, because the assistant's searches are looking for local sources. That's a friendlier target list than the national one too, since local press is easier to earn.

Score the list before you commit outreach hours. Two things decide whether a placement can do anything for this channel; whether AI can read the site at all, and whether AI already cites it for questions in your category. Plenty of well-known publications block AI crawlers, and a placement there won't move a local answer no matter how good the story is.

Then there's the client's own local pages. A city landing page, a location page, a local FAQ. These are frequently the fastest win in a new market, because they're the only pages that can answer a local-intent question directly and the client controls all of them.

Local goal	Why it works in a new market	Typical time to effect
City or region landing page	The only page that answers local questions directly	3 to 6 weeks
Local business directories	Dense local signals AI leans on for "near me" answers	4 to 8 weeks
Regional press placement	Local sources get retrieved for local questions	6 to 10 weeks
Review site presence	Feeds comparison and recommendation answers	8 to 12 weeks
Local partner or association pages	Third-party confirmation the brand operates there	6 to 12 weeks

Attach the shortlist to a project as goals with the market's date range on it. That turns "the expansion is going well" into a number the client can see, and it gives the monthly report something to show besides a share of voice line. Building monthly AI visibility client reports covers how that packaging works across a book of clients.

Watch share of voice per market, not blended

Now the reading, which is where most expansion measurement goes wrong.

Put the new market on its own line, next to the home market, on the same chart. That comparison is the whole point. The home market shows what "good" looks like for this brand with mature local signals behind it, and the new market shows how far there is to go.

Expect the first read to be low. Single digits is normal, and zero happens. That's information, not failure; it says nothing local points at this brand yet, so AI has nothing to retrieve. What you're watching for over the following months is the slope, not the level.

Three patterns come up often enough to name.

Slow and steady climb. Local pages went live, a directory or two picked the brand up, and each week adds a little. This is what the work looks like when it's working.

Flat at zero for months. Usually means nothing local exists yet. Check whether the location pages actually shipped and whether the brand is in the obvious local directories before touching anything else.

A jump with no cause. Someone did something. A local placement ran, a partner listed the brand, a review site crossed a threshold. Find it and mark it, because it's the most repeatable thing you'll learn all quarter.

Read the citations underneath the numbers too, not just the share of voice. In a new market the source list is more useful than the score, because it tells you exactly which local sites AI trusts for these questions. That list is your outreach plan for the next quarter, and you didn't have to guess at it.

What the first quarter should produce

Set expectations before the first report goes out, because the first month of a new market rarely looks impressive and the reason is structural rather than a problem with the work.

By the end of a quarter, three deliverables are reasonable to promise.

1. A real starting position. Share of voice in the market, the local competitors AI names, and the sources it cites. Even if the numbers are small, this is the baseline every later claim gets measured against. The approach is the same one in baselining a brand's AI visibility, scoped to one market.
2. A local source list. The sites AI leans on for these questions, ranked by whether they're readable and already cited. That's a targeting document worth more than most of what a launch plan usually contains.
3. An early slope. Not a win, a direction. Whether the line is moving off the floor tells you if the local work is landing, and it does that months before the share of voice number is worth showing to anyone senior.

Keep the home market on every chart, all the way through. It's the control. When both lines move together, something changed about the brand overall; when only the new market moves, the expansion work did it. That distinction is hard to argue with, and it's the difference between reporting activity and reporting effect.

One last habit. Set the next market up before it opens, not after. Two or three weeks of pre-launch data costs almost nothing and turns the launch report into a before and after instead of a single number with no context behind it.