November 18, 2026
How Do I Track My Brand's AI Visibility by Product Line and Region?
Split it: track AI Overviews and citations per product line and per city in Semrush Position Tracking, with a tag for each product line and a city target for every market, and track the same fixed prompts in ChatGPT and Gemini per city and product line. One national score hides the product line the AI does not connect to you, and for the kitchen and bath remodeler in this test that was bathrooms: nationally, an AI Overview on 58 percent of its bathroom keyword-days over 13 days, and zero citations.
Why does one national AI visibility number mislead a multi-location brand?
Because it averages places that have nothing in common, and product lines the AI treats as different businesses.
A national score for a business that sells city by city is a forecast for the whole country. Partly cloudy. Somewhere. Average Salt Lake City and St. Louis and you land in northwestern Kansas, which is a very nice field that has never once asked for a quote.
AI Overviews are handed out per search, not per domain. Bathroom remodel near me in Lincoln and kitchen remodelers near me in Salt Lake City are separate contests with separate winners. A domain score adds them together into a number that describes neither.
I found the same thing with plain rankings in the Local Rank Gap test: nationally the brand ranked for one of eight generic keywords, and in Salt Lake City for three. AI visibility has the same problem plus one more axis. The city changes the answer. So does what you are selling.
How do I split my keywords by product line in Semrush?
One keyword list per product line, and one tag per line in Position Tracking, so each line reports on its own.
Build the lists in Keyword Magic Tool: one seed per thing you sell (bathroom remodel, kitchen remodel, and so on), keeping the searches a customer would actually type. Then a Brand list: your name, its misspellings, and your name with each city. Brand gets its own line because branded searches flatter you. In this test, 39 of the 48 national citations came from searches that already had the brand name in them.
In Position Tracking, give every keyword a tag (Kitchen, Bath, Brand) when you add it, or tick the rows in the Overview table and tag them from the Actions menu. Then filter the Overview by tag to read one product line at a time.
Confession: I had not tagged this campaign. For this post I split the export by the words in each keyword. Has the brand name: Brand. Says bath and not kitchen: Bath. Says kitchen and not bath: Kitchen. Says both: its own line. That system works right up until somebody searches tub guy near me. Tag yours.
How do I add city targets in Position Tracking?
Add a new target for every city that pays you, with the same keywords, device, and search engine, so the only thing that changes is the location.
In Position Tracking, open the campaign, click the location under the domain name, and choose Add new target. (Extra targets need a Guru or Business plan. Guru allows up to 10 per campaign.) Pick the city, keep the device and search engine, and use your existing target as the keyword source so the lists match. Each city counts its keywords against your tracking limit, so start with the five to ten keywords that make money and the cities that make the most of it.
Leave the city out of the generic keywords in a city target. The target does the locating. Kitchen remodelers near me, tracked from Lincoln, is the search a homeowner in Lincoln actually types.
The setup for this post: the brand's national campaign, 70 keywords on Google desktop in the United States, plus four city targets (Cincinnati, St. Louis, Lincoln, and Salt Lake City), each tracking the same eight generic remodeling keywords. The national numbers below run November 5 to 17. The city targets started November 9.
How do I count AI Overviews and citations per city and per product line?
Two numbers per city and per tag: how many searches showed an AI Overview, and how many of those cited you.
In the Overview report, filter the SERP features to AI Overview. Position Tracking keeps two facts apart: the feature showed up on the page, and your domain was one of its sources. The first number is the size of the room. The second is whether anybody in it said your name.
Then look at the local pack for the same searches. (Position Tracking only reports your own spot in it if the campaign has your business name exactly as it appears on your Google Business Profile.) When the map pack leaves a local search, look at what moved in. In the cities on November 11, it was usually an AI Overview. Nationally, it usually was not.
Pull every target into one table instead of clicking through them. I use the Position Tracking API, which treats each target as its own campaign, and the Semrush MCP can pull Position Tracking reports from a chat. All five targets, 13 days, every keyword: about 10,600 API units. Five targets in five browser tabs is not a reporting system. It is a hostage situation.
ChatGPT and Gemini do not show up in a Google rank report, so give them a fixed exam. In Semrush that is Prompt Tracking in the AI Visibility Toolkit: the same prompts, rerun every day in ChatGPT, Google AI Mode, and Gemini, with tags for city or product line. Write the prompts per city and per product line (who does bathroom remodeling in Lincoln, Nebraska; best kitchen remodeler in Salt Lake City) and then leave the wording alone. Rewrite the questions every month and you are not tracking AI visibility. You are writing fan fiction about it.
Then split what comes back by page, the way I did in the page-not-domain test.
One caveat before the numbers. Position Tracking captures each search once a day, and AI Overviews change during the day. A zero below means not cited when Semrush looked. It also means that 13 days in a row. A keyword-day is one keyword checked on one day.
What did the split show for a real kitchen and bath remodeler?
Bathrooms: the product line with the most AI Overviews got zero national citations and one city citation that lasted a day.
Nationally, November 5 to 17, by product line:
Bath, 29 keywords without the brand name: an AI Overview on 218 of 377 keyword-days, 58 percent. Cited: zero.
Kitchen, 11 keywords: an AI Overview on 64 of 143 keyword-days, 45 percent. Cited: zero.
Kitchen and bath, 9 keywords that say both: an AI Overview on 52 of 117 keyword-days. Cited 9 times, every one of them for kitchen and bath franchise, a search with 50 a month from people shopping for a franchise, not a bathroom. The AI is confident this company sells franchises. Whether it remodels anything remains, as far as the robot knows, an open question.
Brand, 17 keywords with the name in them: an AI Overview on 76 of 221 keyword-days. Cited 39 times. (Four leftovers, like bedroom renovation near me, fit none of the lines and were not cited either.)
Add it up. 48 national citations in 13 days. 39 from people who already typed the name, 9 from people who want to own a franchise, and zero from a homeowner who had not heard of the brand yet. A national AI visibility score would call this brand cited. It is. By its own fans.
Under those 218 bathroom AI Overviews, the brand ranked in the top 10 on 3 keyword-days, all for one search, bathroom remodel logan utah, two of them with its Northern Utah bathroom page. So the bathroom pages exist. They are just not doing the work. The Rank Gap test showed the same hole in plain rankings: no bathroom page ranked for a bathroom keyword in any of the four cities.
The cities said it in a smaller voice. 32 searches a day across four markets. AI Overviews on zero of them on November 9, on 19 by November 12, and on 11 on November 17. Over those nine days the brand sat in the top three, in the map pack or organic, on the same page as an AI Overview 20 times. Cited in none of them.
The one city citation came on a search where the brand was nowhere in the organic results or the map pack: Cincinnati, bathroom remodel near me, November 14, and it went to the brand's national bathroom page. On the 15th it was gone. A mayfly lives about a day. So did this. Sitting near the top did not get the brand cited, and being nowhere did not stop the one citation it got. The AI Overview is not reading your rank report.
Lincoln is the cleanest example. The brand ranks first for one day kitchen and bath every single day, in the map pack on all but one of them. From November 14 an AI Overview showed up on the same page, four days running, and did not cite the business in first place. Like a waiter who will not make eye contact.
And the map pack moves. On November 11, and again on November 17, the map pack dropped off close to half the searches that had one, in the national and city campaigns on the same day (national 62 to 35, then 50 to 29; cities 31 to 17, then 26 to 13). It hit both campaigns at once, so it is not about this brand. But only the city view shows what it cost: in Salt Lake City the brand was third in the map pack for kitchen remodelers near me every day the pack showed up from November 9 to 16, and on the 17th the map pack was gone, an AI Overview was on top, and the brand was nowhere in the top 100.
So what do I do with the split?
Fix the bathroom pages the AI is not picking, because bathrooms have the most AI Overviews and the national view and all four cities failed them.
This brand already has bathroom remodel pages for St. Louis and Lincoln, and neither ranked for a bathroom keyword in its own city. Make each one a real bathroom remodeling page for that city, not a bullet on a services page. It is the one fix that should move every market at once.
Keep the brand line separate so it stops flattering you. Keep the city targets on. Keep the prompts fixed. I will update this post in December with 30 days of data.
Position Tracking is part of the SEO Toolkit in Semrush One. Tag the product lines, add the cities, count two numbers. Then ask the question a national score is built to avoid: which product line, in which city, is it hiding?