Research · October 2026
How ready are local-business websites for AI answers?
We scanned 1,902 local-business websites in 20 cities across the USA and UK on 12 October 2026, using the same free scan anyone can run on this site. 1,355 could be read and scored. This page gives every finding with its count, the method, and what the scan cannot tell you.
Check registry version 7 · scoring version 1 · 730 websites in the USA and 625 in the UK. No business is named anywhere in this report.
The findings in eight numbers
1 in 4 local-business websites already has an llms.txt file.
Has an llms.txt file: 338 of 1,355 (24.9%)
97% of Wix sites have one, because the builder writes it. On Squarespace it is 4%.
Wix sites with an llms.txt file: 118 of 121 (97.5%)
Under 1% block an AI answer engine in their robots.txt.
Blocks an AI answer engine in robots.txt: 9 of 1,355 (0.7%)
46% say nothing about the business in a form machines can read: no name, address, phone or hours in structured data.
No business facts in structured data: 621 of 1,355 (45.8%)
59% do not give their opening hours in structured data.
No opening hours in structured data: 796 of 1,355 (58.7%)
7 in 10 have no questions and answers on the pages we read.
No questions and answers: 970 of 1,355 (71.6%)
58 / 100 is the typical readiness score. One site in three scores under 50.
Scores under 50 out of 100: 468 of 1,355 (34.5%)
1 in 6 websites listed for these businesses no longer works.
Listed website no longer works: 328 of 1,902 (17.2%)
The typical score is the median, 58.1 out of 100. A quarter of websites score below 45.3 and a quarter above 72.1. 213 of 1,355 (15.7%) score 80 or more.
The findings as one image
You may republish this image with a link back to this page. Download the image.
1. What was scanned
The sample is independent local businesses that someone has mapped in OpenStreetMap with a website. Chains were set aside, and so were listings whose website is a social or booking-platform page.
| Step | Websites |
|---|---|
| Mapped places in 11 trades and 20 cities that list a website | 12,570 |
| Pointing at a social or booking-platform page instead | 280 |
| Belonging to a chain (one domain on three or more places, 291 domains) | 2,855 |
| Independent websites left | 9,044 |
| Drawn at random and scanned, at most 100 per trade per country | 1,902 |
| No working website at the listed address | 328 of 1,902 (17.2%) |
| Answered our reader with a security check, so nothing could be read | 172 |
| Redirects from https back to plain http, which the scan will not follow | 47 |
| Read and scored | 1,355 |
Trades: hairdressers and barbers, beauty salons, dentists, vets, car repair, gyms, estate agents, plumbers, electricians, roofers and accountants. Cities: New York, Los Angeles, Chicago, Houston, Phoenix, Philadelphia, San Antonio, San Diego, Dallas and San Jose; London, Birmingham, Manchester, Glasgow, Leeds, Liverpool, Newcastle, Sheffield, Bristol and Edinburgh. Unless a line says otherwise, every share below is of the 1,355 websites that were read.
2. One in four has an llms.txt file
An llms.txt file is a plain summary of a website written for AI tools. A quarter of the websites we read already serve one, and for most of them the owner did not write it. On Wix, Shopify and Duda nearly every website has the file, because the platform adds it. On Squarespace almost none does.
| Platform | Websites with an llms.txt file |
|---|---|
| WordPress | 109 of 576 (18.9%) |
| Custom-built or not recognised | 38 of 342 (11.1%) |
| Wix | 118 of 121 (97.5%) |
| Squarespace | 4 of 110 (3.6%) |
| Next.js | 13 of 46 (28.3%) |
| Shopify | 27 of 28 (96.4%) |
| Joomla | 1 of 24 (4.2%) |
| Duda | 19 of 19 (100.0%) |
The platform is the scan’s own reading of the page source. We did not check which WordPress plugin wrote a file, or whether an owner had edited one. You can build an llms.txt file for your own website free.
3. Almost nobody blocks AI crawlers in robots.txt
A robots.txt file is where a website owner tells crawlers what they may read. We counted a block only when the file names the crawler and disallows the whole website. The answer fetchers are OAI-SearchBot, ChatGPT-User, PerplexityBot and Perplexity-User. The training crawlers are GPTBot, ClaudeBot, Google-Extended and CCBot.
| A robots.txt rule that | Websites |
|---|---|
| Shuts out an AI answer fetcher | 9 of 1,355 (0.7%) |
| Shuts out an AI training crawler | 14 of 1,355 (1.0%) |
A firewall can still turn a crawler away with no rule in this file. That is not counted here. You can check your own robots.txt for AI crawlers free.
4. Nearly half state nothing about the business in structured data
Structured data is the block of code that states a business’s name, address, phone number and opening hours in a form a machine does not have to guess at. A bare tag that only names the website, which some builders add by default, is not counted as describing the business.
| Structured data | Websites |
|---|---|
| No structured data describing the business | 621 of 1,355 (45.8%) |
| Some, with expected facts missing | 499 of 1,355 (36.8%) |
| Every expected fact stated | 235 of 1,355 (17.3%) |
482 websites describe themselves as a local business. This is what those leave out.
| Left out of the local business entry | Websites |
|---|---|
| Opening hours | 175 of 482 (36.3%) |
| Phone number | 94 of 482 (19.5%) |
| Website address | 77 of 482 (16.0%) |
| Address or service area | 36 of 482 (7.5%) |
| Business name | 27 of 482 (5.6%) |
You can generate the code for your own business free.
5. Scores by platform and by trade
The readiness score runs from 0 to 100 and is worked out exactly as described on how the score works. The bars show the median for each group, and the tables give the exact figures.
| Platform | Websites | Median score | No business data | llms.txt | Q and A |
|---|---|---|---|---|---|
| Shopify | 28 | 69.8 | 17.9% | 96.4% | 42.9% |
| Next.js | 46 | 68.6 | 26.1% | 28.3% | 41.3% |
| Wix | 121 | 65.1 | 21.5% | 97.5% | 15.7% |
| WordPress | 576 | 61.6 | 37.2% | 18.9% | 27.8% |
| Squarespace | 110 | 58.1 | 26.4% | 3.6% | 20.0% |
| Custom-built | 342 | 47.7 | 74.3% | 11.1% | 16.4% |
| Trade | Websites | Median score | No business data | llms.txt | Q and A |
|---|---|---|---|---|---|
| Roofers | 63 | 69.8 | 23.8% | 34.9% | 46.0% |
| Plumbers | 89 | 66.3 | 31.5% | 31.5% | 36.0% |
| Vets | 155 | 62.8 | 42.6% | 17.4% | 16.8% |
| Gyms | 153 | 61.6 | 37.9% | 30.7% | 30.1% |
| Dentists | 149 | 60.5 | 43.6% | 28.9% | 29.5% |
| Car repair | 115 | 58.1 | 47.0% | 25.2% | 27.0% |
| Accountants | 151 | 54.7 | 55.6% | 18.5% | 13.9% |
| Hairdressers and barbers | 146 | 53.5 | 44.5% | 25.3% | 13.0% |
| Beauty salons | 133 | 52.3 | 55.6% | 30.8% | 19.5% |
| Electricians | 61 | 52.3 | 55.7% | 24.6% | 19.7% |
| Estate agents | 140 | 50.0 | 55.7% | 15.0% | 14.3% |
No business data is the share with no structured data describing the business. Q and A is the share with three or more questions and answers on the pages read. Platforms with fewer than 20 websites are left out. Roofers, plumbers and electricians are small groups, so one website moves their figures by a point or more.
6. Smaller findings
| Finding | Websites |
|---|---|
| Has a sitemap | 1,135 of 1,355 (83.8%) |
| Passes the https and mobile basics | 1,249 of 1,355 (92.2%) |
| Three or more questions and answers | 306 of 1,355 (22.6%) |
| No Open Graph share tags on any page read | 306 of 1,339 (22.9%) |
| Titles or descriptions missing or too thin | 461 of 1,339 (34.4%) |
| One main heading per page, images described, language declared | 403 of 1,339 (30.1%) |
Sixteen websites were read but returned no page the newer checks could measure, which is why some rows are out of 1,339.
7. Method
- Sample. Businesses were taken from OpenStreetMap on 12 October 2026: every place in the 11 trades, inside a box drawn around each of the 20 cities, that lists a website. Up to 100 were drawn at random for each trade in each country.
- Scan. Each website went through the free scan. It reads the home page and up to four more pages over plain HTTP, with robots.txt, the sitemap and llms.txt. No browser and no AI model is involved.
- Score. Eleven checks, each a pass, a warning or a fail, combined as the methodology page sets out.
- What was kept. The address of the website, each check’s result and the counts a figure needs. Nothing that could hold a phone number or an email address was stored.
- Checks on the data. Twenty websites were read a second way by hand and nineteen agreed; the twentieth is why section 4 says what it does about a bare website tag. Fifty were scanned again: 49 gave the same result for whether they could be read, and 32 of the 33 read both times gave the same result on every check.
8. What this does not tell you
- It says nothing about whether an AI engine recommends any of these businesses. It measures whether a website can be read and what it states, which comes first.
- At most five pages were read, with no browser. A fact on a sixth page, or one that only appears after scripts run, was not seen.
- The sample leans towards established premises in big cities. Trades that work from a van are thin in it.
- The scan ran from one connection in Thailand on one day. That does not change what a robots.txt file, a page’s code or an llms.txt file says, but websites that turn foreign visitors away are missing from every figure after section 1.
- Anthropic’s answer fetchers and the crawlers of Google and Bing are not among the eight the scan names.
Use these figures
You may quote any figure on this page with a link back to it. Please give the count with the percentage. Suggested citation:
MentionMachine (2026). How ready are local-business websites for AI answers? A scan of 1,902 websites in the USA and UK, 12 October 2026.
Download the figures as a CSV file. Contains data © OpenStreetMap contributors, available under the Open Database Licence.
Run the same scan on your own website. It is free and needs no account.