AI names local businesses more often in the nearby towns their website names

We asked Google's Gemini who to call for the same trade in neighbouring towns, 2,640 answers across 55 local markets. Then we read the homepage and the web addresses of the pages of 569 businesses it named, to see which of those towns each website names. 419 of them state an address we could place, and they are the study.

The same business, asked about in several towns

Comparing one business with another is close to worthless here. A company with a page address for every town in the county may be a bigger company, and AI may name bigger companies more anyway.

So each business is compared against itself. We asked AI the same kind of question in neighbouring towns: who should I call for this job, in this town. A business whose website names some of those towns and not others gives two numbers. The share of the towns its website names where AI named it at least once, and the same share of the towns it does not. Same business, same website.

The town a business gives as its own address is left out of both numbers. A plumber is usually named in the town it sits in and writes that town on its site, and that would look like the result we are testing for.

177 businesses could be compared this way. On average AI named them in 52 percentage points more of the towns their website names (90% interval 46 to 58). 123 were named more in the towns they name, 7 less, 47 the same.

The same split showed up when the town a website names was the farther one

Distance could explain this. A business names the towns nearest to it, and AI may name a business in a town because it is near, whatever the website says.

We measured how far each town is from the town the business gives as its address, using the Census Bureau's location for each town. Then we kept only the awkward cases: a business whose website names a town farther away and leaves out a nearer one.

The same split showed up where distance pointed the other way. For 67 businesses the website names a town at least 3 km farther from the business's own town than a town it leaves out, measured between town centres. AI named 32 of them in more of the farther, named towns, 4 in fewer, 31 the same.

Where the named town was the nearer one, so that distance and the website point the same way, 94 of 129 businesses were named in more of the named towns, 3 in fewer, 32 the same.

15 of 25 businesses were named in more of their page-address towns than their homepage-only towns

We looked for each town in four places: the address of any of the site's own pages, like /plumber-glendale, whatever that page holds, the homepage's title, description or a heading, the words of a homepage link, and the rest of the homepage's text. A town found in more than one place is counted once, under the first of them in that order.

Taking the four places together, AI named the business in 48% of the towns its site names (302 of 629) and in 14% of the towns it names nowhere we looked (88 of 614). Most of the named ones, 497 of 629, are in a page address. A town only in the homepage's other text: 20 of 103 (19%).

Those figures combine every business, and a company with a page address for every town is a different company from one with a line of text. So we looked at one business at a time again.

144 businesses name one town in a page address and another nowhere we looked. AI named 107 of them in more of the page-address towns, 4 in fewer, 33 the same.

25 businesses name one town in a page address (the web address of one of the site's pages) and another only on the homepage. AI named 15 of them in more of the towns that are in a page address, 2 in fewer, 8 the same.

43 businesses name one town only on the homepage and another nowhere we looked. AI named 16 of them in more of the homepage towns, 3 in fewer, 24 the same.

A town in a page address means the site links to or lists a page whose web address has the town's name in it. We did not open those pages, so we cannot say what on them counts, or how much of a page it takes.

AI named these businesses in 48% of the other towns their site names and 14% of the ones it does not

Those two figures combine different businesses, so they do not hold the business constant. The comparison that does is the 52 points at the top, which covers the 177 businesses that name some towns and not others. These cover every business in the study, over towns other than its own: 302 of 629 business-and-town pairs where the site names the town, 88 of 614 where it does not.

In 237 of the 264 cases where a business's own town was one we asked about, AI named it there (90%). That figure may be high partly because of how a business enters the study: AI has to have named it in at least two answers in one town.

Among towns where AI did name the business, 12 businesses were named in more of the answers in the towns their site names and 7 in fewer. That is not a clear difference. What we measured is being named in a town at all.

On 145 of 417 websites we found none of the neighbouring towns we asked about

Set cause aside, because this count does not depend on it. Of the 417 businesses with a neighbouring town to check, 145 name none of them in any of the places we looked. 95 name every one.

The towns are the neighbouring towns we asked about. They are not a list of where each business works, and we did not check that.

What we would do with this

Worth doing

• Write down the towns you actually take work in. By name, the way a customer would type them.
• Check your homepage and your page addresses for each one. Most of the towns we counted as named, 497 of 629, were in a page address, and a town only in the homepage's other text was named 19% of the time against 14% for a town named nowhere we looked. 145 of 417 sites here named no neighbouring town in any of them.
• Use the town's own name. We counted a town only where the site wrote that town's name, so that is the wording this result is about. We did not test region names such as "Greater Los Angeles".

Worth being careful about

• Do not list towns you will not drive to. This study says nothing about what happens after AI names you, and a call from a town you do not serve is a wasted call for both sides.
• Do not build fifty thin town pages on the strength of this. We read page addresses and never the pages, so nothing here says a stack of near-identical pages pays.