The Metro Atlanta Invisibility Count

26.1%

have no working website of their own

21.9–30.2%

the 95% confidence interval

n=202

businesses checked by hand

~2,300

businesses, within 25 miles of downtown

The finding

26.1% of mapped independent businesses within 25 miles of downtown Atlanta have no working website of their own. The 95% confidence interval is 21.9% to 30.2%. That is roughly 2,300 businesses, somewhere between 1,950 and 2,700.

The estimate held steady as the sample tripled. Within the unmapped half of the frame it read between 43% and 48% at every checkpoint from n=30 onward, and landed at 41.6% at n=202. A number that doesn’t move when you triple the sample is a number that is measuring something real. That stability is the strongest evidence in this study.

74 own site, live 2 site on file, dead 14 page on a platform 10 nothing findable

Two more findings came out of the same work, and each stands on its own.

Roughly 1 in 20 local business websites on record is already dead. Of businesses with a website on file, 4.7% return nothing. The domain expired, the DNS is gone, or the page was a Google Business Profile site, a free product Google shut down in March 2024.

About a quarter of the businesses that “have a website” don’t own it. Their entire web presence is a page on somebody else’s platform. A booking tool, a point-of-sale vendor, an unbranded shop subdomain. They rent an address they can’t move, can’t rank, and lose outright if the platform changes its terms.

Eight businesses in the corpus still hand out a .business.site address. Google deleted those sites in 2024 and, as far as the record shows, nobody told the owners.

Why the number is lower than it could be

Three scarier headlines were available. All three were thrown out.

1. “58% of local businesses have no website.” That is what the raw map data says, and it is wrong. Website coverage in the map was 42.2% and phone-number coverage was 42.3%. Near-identical figures like that are the signature of how thoroughly a volunteer filled in a record, not of whether a business has a website.
2. “22% of local business websites are broken.” That was the raw HTTP failure rate, and it is also wrong. Twenty-two of the 33 failures were bot protection blocking an automated request. Those businesses are open. Only 7 failures survived a second, patient look.
3. “Half of them are invisible.” That was the running estimate at small n. It fell as the sample grew, and the stratified combination brought the population figure to 26%.

A number that survives a reporter checking it is worth more than one that doesn’t. Every claim on this page is reproducible from the method below.

Method

The frame

All business points of interest in OpenStreetMap within 40 km, or 25 miles, of downtown Atlanta (33.7490, -84.3880), pulled through the Overpass API: shops, crafts, offices, healthcare, and business-like amenities.

Step	Count
Raw points in radius	18,914
With a name	17,467
Independent, chains removed	9,254
Clean frame, government and institutions removed	8,937

Chains were removed two ways. A brand tag in the map record, or a business name appearing three or more times across the metro. Government offices, public health, libraries, schools, banks and one Fortune 500 company were filtered by category and name.

Two strata, two kinds of checking

Stratum	Definition	Size	Share
A	The map records a website	3,766	42.1%
B	No website in the map	5,171	57.9%

Stratum A was checked exhaustively. Every one of the 148 URLs in the drawn sample was fetched over HTTP with a browser user agent, redirects followed, twenty-second timeout. Failures were re-tested and checked for DNS records, to separate a dead site from a site that blocks robots.

Stratum B was checked by search. A random subsample of 202 businesses, each searched by name plus city, and coded by hand. About 236 were searched to reach 202 usable observations.

How stratum B was coded

Code	Meaning	n	Share
SITE	Own website on its own domain, live	118	58.4%
SOCIAL	A social, booking, ordering or point-of-sale page standing in for a website	48	23.8%
NONE	No web presence findable beyond directory scrape pages	36	17.8%

“No working website of their own” means SOCIAL plus NONE: 84 of 202, or 41.6%, plus or minus 6.8 points.

Thirty-four further businesses were excluded and not counted: chains that slipped the filter, businesses found closed mid-check, and names too generic to identify with confidence. Excluding them keeps the number conservative.

Putting the strata together

p = 0.421 × 0.047  +  0.579 × 0.416  =  0.261

The 95% interval, from the stratum standard errors, is 21.9% to 30.2%.

Limits, stated plainly

1. The map is not a census. OpenStreetMap is a volunteer map. The frame is businesses mapped in it, not all businesses in metro Atlanta. Coverage skews urban and skews toward retail and food. The right phrasing is always of mapped independent businesses, never of all businesses.
2. Search has a floor. A business could own a site that ranks so poorly a search for its own name and city doesn’t find it. But that is close to the definition of invisible. If that search can’t find it, a customer can’t either.
3. Chain filtering is imperfect. A regional chain with two locations reads as independent. A few slipped through and were counted as having sites, which pushes the estimate down, not up.
4. This measures existence, not quality. A live site from 2011 that is unreadable on a phone counts as a site here. The number is a floor.
5. “Own website” is a judgment call at the margin. A business whose only page is on a booking or point-of-sale vendor’s domain was coded SOCIAL. It has a page, but not one it owns, ranks, or can take with it. Coding those as SITE instead would drop the headline to roughly 20%.

Reproducing it

The frame comes from the Overpass API, which needs no key. Filter chains and government by tag and name, stratify on whether a website is recorded, and draw a seeded random sample. For stratum A, fetch every URL with a browser user agent and look up DNS for the failures. For stratum B, search each business by name plus city and code it SITE, SOCIAL or NONE.

Always separate a bot-protection refusal from a dead site. It is the single easiest way to produce a wrong and embarrassing number.

If you want the same count run for a specific neighborhood or county, ask. That is the kind of follow-up this was built to take.

If you’re writing about this

The sentence to cite. About one in four independent businesses with a physical location in metro Atlanta, 26.1% with a 95% confidence interval of 21.9% to 30.2%, have no working website of their own, according to a September 2026 stratified sample of the 8,937 independent businesses mapped within 25 miles of downtown, run by Brandon George of Curb to Click.

Copy the sentence to cite

The version for a caption. 1 in 4 metro Atlanta businesses can’t be found online.

What it is not. Not a census. Not all businesses. Of mapped independent businesses, and the map is a volunteer one. Say mapped and you’ll be right.

How to check it. The frame, the strata, the coding and the arithmetic are all on this page, and the tools are free. A reporter with an afternoon can re-run stratum A from scratch. Stratum B takes longer. The coding rules are above.

Who to call. brandon.george@gmail.com. Atlanta. Available for a walk to a specific storefront if that helps the story.

Boilerplate. Curb to Click builds websites and landing pages for independent businesses with a real address within 25 miles of downtown Atlanta. Brandon George drives out, stands where the customers stand, writes a plan the owner keeps either way, and builds the front door. The method behind the work, Good Press, is published free at writethegoodfight.com.