How we did it.

The findings are in the report. The data is public: download it as a CSV file, one row per business (324 rows), with no names, addresses, website addresses or Google IDs.

1. The businesses

Source. Google Maps (Google Places Text Search), on October 19, 2026. We kept every result Google gave (up to 60 per search), not only the first ones.

The 130 searches. 13 search words, each one in each of 10 towns, as “search word in town, AR” (for example “plumber in Conway, AR”):

Trade	Search words
Restaurants	restaurant
Salons and barbers	hair salon · barber shop
Home services	plumber · roofing contractor · HVAC contractor · remodeling contractor
Auto repair	auto repair shop
Health practices	chiropractor · dentist · physical therapy clinic
Local shops	boutique · florist

Area	Towns
Little Rock area	Little Rock · North Little Rock
Northwest Arkansas	Fayetteville · Springdale · Rogers · Bentonville
Conway	Conway
Jonesboro	Jonesboro
Fort Smith	Fort Smith
Hot Springs	Hot Springs

6 trades × 6 areas = 36 groups. A business found by more than one search belongs to the group of its first search.

Who is left out.

• Chains and franchises: a national brand on our list (for example Great Clips, Midas, Aspen Dental), or the same business name or the same website at 3 or more places in all our searches.
• Places Google marks as closed (none in these results).

The searches found 5,046 businesses. 4,471 independent businesses were left.

The pick. From each of the 36 groups we picked 9 businesses at random: 36 × 9 = 324. The random pick uses a fixed seed (“xiobo-ar-2026”), so the same scan always gives the same 324 businesses.

Businesses without a website count. When the Google listing has no website, or links only to Facebook, Instagram, Yelp, Linktree, a booking or ordering platform page, or a directory, the business counts as “no website of its own”.

2. The check

One automated check of each home page, on October 19, 2026, with the same rules for every site. Only the home page, once.

Check	Counts as “yes” when
Website of its own	The Google listing links to a website on its own domain.
Works	The website opens, in our check or in a normal browser, and shows the business (not an error, a “not found” page or a domain for sale).
Secure	The page opens on https with a valid certificate.
Set up for phones	The page has the setting that tells a phone to fit the page to its screen (the viewport tag).
Phone speed	Google PageSpeed mobile performance score of 90 or more, Google’s “good” band. 50 to 89 is “needs work”, 0 to 49 is “poor”. This is a lab test: it loads the page as a mid-range phone on a slow 4G connection does. We use Google’s score bands for this test, not the 2.5-second line for real visits.
Tap to call	The phone number is a link that starts a call (a tel: link) with a full phone number, 10 digits or more.
Hours	Opening hours are in the page text (for example “Mon–Fri 8am–5pm”, “M-F: 8AM - 5PM”, “Open today 8:30 am – 5:00 pm”) or in the page data that Google reads.
Address	A street address is in the page text or in the page data that Google reads.
Way to reach them	A contact form (a message box, or 3 or more fields; a lone email box is a newsletter sign-up and does not count), an email link, or a link to a contact page.
PDF menu	Restaurants only: the menu link opens a PDF file.
Copyright year	The latest year after the © sign is 2024 or older.
Spanish	The page is marked as Spanish, points to a Spanish version, or has a link named Español or Spanish.

3. How we checked our own work

A second look at every website, in a real browser. We opened every website that opened in the first check again, in Chrome, at phone size, with JavaScript on. For tap to call, hours, address, way to reach them and Spanish, the browser’s answer replaces the first check’s answer, both ways: a wrong “yes” counts as much as a wrong “no”. This second look also caught 3 websites that open on a “not found” page or a domain for sale. They count as “does not work”.

Every website that did not open, opened again. 21 websites did not open in the first check (an error from the server, or the address does not exist). We opened each one again in a normal browser. 12 of them opened. Most of these servers had refused our automated check, not a customer. They count as “works”. Because we have no first-check answers for those 12, they are not in the 180 websites the other checks are based on.

Two random samples of 30, read page by page. We picked 30 websites at random and read each home page, page by page, against five answers: tap to call, hours, address, way to reach them, Spanish. We used what we found to fix our rules (the browser look above). Then we picked 30 more at random and read them the same way, before we changed any rule for them. On these 30, 148 of our 150 answers matched what the page shows (98.7%). The reading was done with an AI model that saw the page text, the links and a phone screenshot of each page.

4. The counts behind each number

• 324 businesses: 207 with a website of their own, 27 with only a Facebook, directory or platform page, 90 with no website on the Google listing.
• 204 listings with a website of their own, in the “does not work” number (3 websites ask automated checks to stay out, so we left them out and did not open them).
• 180 websites that opened and that we could check. Every check on the home page uses this base.
• 168 websites that Google PageSpeed could test (the speed number).
• 38 restaurant websites (the PDF menu number).

5. Limits

• This is a sample of independent businesses that are on Google Maps, in 6 trades and 6 areas, on one date. It is not a sample of every Arkansas business: most Arkansas small businesses have no employees, and many are not on Google Maps.
• A few businesses in the sample are local branches of larger organizations (for example a hospital system, or a company with offices in several states). The sample rule keeps them.
• We read the home page only. Hours or an address on another page do not count.
• Hours and addresses count when they are in the page text or in the page data for Google. A customer cannot see the page data, so the share of customers who cannot find them can be higher.
• Speed is a lab test, not each customer’s real visit.
• Margins of error (95%): about ±5 points for all 324 businesses, about ±7 points for the 180 websites, and ±13 to ±22 points for one trade or one area. A difference of less than 20 points between two trades or two areas can be chance.
• Spanish at home: people age 5 and older, U.S. Census Bureau, American Community Survey 2020–2024 5-year estimates, table C16001 (Arkansas 6.1% ±0.1, Springdale 31.8% ±2.1, Rogers 22.7% ±2.6, Fort Smith 15.6% ±0.9, Little Rock 9.0% ±0.7, Conway 5.6% ±0.9; 90% margins).

6. The columns in the data file

Column	Meaning
trade, area	The group the business was picked from.
website	What the Google listing links to: own, none, facebook, instagram, linktree, platform_page (booking or ordering platform), directory.
opened	True = the website opened and we could check it (one of the 180). The other columns are empty when it is False.
https, phone_ready	Secure (https) · has the mobile viewport tag.
lcp_seconds_lab, speed_score_lab	Google PageSpeed mobile lab test: time to show the main content (Largest Contentful Paint), and the performance score 0–100.
speed	good (90–100), slow (needs work, 50–89), poor (0–49). Empty when PageSpeed could not test the page.
tap_to_call, hours, address, contact, spanish	The checks above, after the second look in a browser.
pdf_menu	Restaurants only.
booking_or_ordering	A booking or ordering link or button (Toast, OpenTable, Vagaro and others). Not used in the report.
copyright_year	The latest year after ©, empty when there is none.
platform	What the website is built with, when we can tell (WordPress, Wix, Shopify, Squarespace…).

7. Check our work

• The data: every business we checked is a row in the CSV file, without names.
• The code: public, linked at the end of this page.
• The Google scan: it holds Google listing content, which Google’s terms let us keep for 30 days only, so we cannot publish it. Anyone can run the same 130 searches and compare.
• Questions about a number: write to us at hello@xiobocreative.com.

8. Who we are

Xiobo Creative is a small design studio based in Arkansas. We build websites for small businesses, so we have an interest in this topic. That is why the method, the code and the data are public.

For researchers: the code. study.py picks the sample and computes every number in the report (Python 3.9 or newer, standard library only). browser-all.mjs is the second look at every website in a real browser (Node.js, Playwright and Chrome).