How Do I Do Keyword Research for Multiple Locations?

How do I do keyword research for multiple locations?

Not by copying the St. Louis list and finding-and-replacing the city name. That is not research. That is a mail merge with ambitions.

Last week I did this for a real estate brokerage that wants to rank in five suburbs on the west side of the metro. Same service, five towns, five different people typing five different things. Ballwin has schools people move for. Lake St. Louis has a lake. Chesterfield has a valley that floods, ask anyone who was here in 1993. The only word those towns share is "homes," and "homes" is not a keyword, it is a category of building.

Why the template page never ranks anywhere

A location page built from a template is the same page five times with the city swapped. Google noticed that years ago. AI notices in about a second, because it is comparing your five pages against each other and against every other page that answers the question better.

The part people miss is that the searches themselves differ by town. A commuter suburb searches around the highway and the train. A lake community searches around the lake. A town with a famous school district searches around the school district. Your five lists should look different because your five customers are different. If the lists come out identical, you have measured your template, not your market.

The per-location process, in order

One. One seed per location, not one seed for the brand. "[city] homes for sale," "[city] real estate," "[city] realtor," and then the things only a local would say: the neighborhoods, the school district by name, the lake, the highway everyone complains about.

Two. Pull the real volume per phrase. I use the Keyword Magic Tool in Semrush One, filtered to the city terms and sorted by intent, and I run it once per town. Five towns, five lists. The lists do not look alike, which is the entire point.

Three. Keep the question phrases. "Is [city] a good place to live," "what is [city] known for," "how are the schools in [city]." The brokerage had never written a sentence answering any of them, because everyone in the office already lives there. Those questions are exactly what an AI assistant fans a search out into, and the volume behind them is real. On a home services list I pulled the same week, the questions report alone came back with the top six phrases running from 4,400 searches a month down to 320.

Four. Check who ranks now, per town. If Zillow owns the top ten for the money term in one suburb and a local blog owns it in the next, those are two different pages with two different jobs, not one template.

Five. One primary keyword per page, written for the human who typed it. The school district paragraph goes on the school district town. The lake paragraph goes on the lake town. Revolutionary.

Six. Track it per location. Position Tracking in Semrush lets you set the location down to the city, so each page is measured where its customers actually search, not from a data center in another state.

The rollup: what a franchise owner and a location manager both need

At Local Howl we run the same problem at a bigger scale for franchises. One brand, six locations, six managers who each care about one town, and one owner who cares about all of them. We keep a keyword list per location and one rollup above it, so the owner sees the brand moving and each manager sees their own street. Nobody has to squint at a spreadsheet that averages Denver with Des Moines. (An average of two cities is a city that does not exist. It ranks accordingly.)

The Semrush MCP server is how we automate the pulls now, one location at a time, straight into our own dashboard. The research did not change. Only the typing did.

Three ways this goes wrong

The brand list masquerading as the location list. Everything is "[brand] [city]" and nothing is what a stranger types.

Ignoring the neighborhood level. In a bigger suburb the neighborhood name outranks the town name for the people who live there, and they are the ones buying.

Pages that exist only because a keyword exists. If the page would embarrass you in front of a resident of that town, it will embarrass you in front of the model too.

One town, one list, one page. Then do it again.

The hidden questions an assistant breaks a search into are the subject of the fan-out field guide, and why each page has to agree with the profile it represents is in the entity corroboration piece.

If you run more than one location, count your keyword lists. One is the wrong answer.