AI-friendly websites
AI-friendly is nine checks. Nobody passes them yet.
A customer now asks an assistant before they search, and the assistant reads a plain text file at your address before it names anyone. We wrote down what a site has to do to be quoted from that file, as nine checks anyone can run in a minute, and ran them on 281 trades businesses in Iowa and fourteen other states. None passes all nine. Every site we build does, the day it launches.
Your file: what you sell, what it costs, how to reach you.
The answer, with your line in it, quoted from you.
The file
What the machine reads before it names you.
When somebody asks an assistant who does the thing you do, the assistant does not browse your website the way a person does. It goes looking for the shortest trustworthy account of your business it can find, and the address it tries first is yoursite.com/llms.txt: a page of plain text that states what you sell, what it costs, where you are and how to reach you.
If the file is there and it is yours, the answer carries your name, your price and your phone number, quoted from you. If it is not, the silence gets filled anyway, from whatever third-party page ranked that day. And if a platform generated it, the answer describes the shopping cart.
Google spent twenty years proving that the business that showed up first and stayed consistent became nearly impossible to unseat. That is what search compliance meant. The assistants are forming the same habits now, and the name they learn first is the name they keep.
# www.examplecompany.com ## Pages - [Location One](https://www. examplecompany.com/location-one)- [Thank You](https://www. examplecompany.com/thank-you)- [Blog](https://www.examplecompany .com/blog)- [Home](https://www.examplecompany .com/): Call today for a quote.
# Online StoreThis store is built on a hostedcommerce platform. ## For agentsUse the platform's checkout protocolto complete purchases. See theplatform documentation. ## Pages- /collections/all- /pages/about- /policies/refund-policy
# K&J Shine Works> Auto detailing in Cedar Rapids, Iowa. The shop is at 842 Vernon Valley DrSte 4, Cedar Rapids, IA 52403.Phone (319) 677-2890. ## Prices- Interior Deep Clean: $150 to $250- Ceramic Coating, 1 year: $950- Maintenance Detail Program: $60 to $120 per visit
What we found
Three markets, one dot each.
We took a whole market at a time, fetched the file on every site in it, and read every one we found. A pine dot is a business whose file is an account of the business. A grey dot is a file a plugin or a website builder generated, which is almost always a list of the site's own page links. A hollow one is nothing at all.
- 341
- businesses fetched, in three markets
- 19
- publish a file about the business
- 0
- pointed at it from robots.txt
01
Iowa trades: plumbers, electricians, HVAC, roofers, detailers and eighteen more
157 businesses, fetched 17 September 2026, as a market study of our own.
41% of the set answer at that address. Look twice: 5% of the set publish a file that is about the business. 1 of the 157 put a phone number, a place and a price or the hours in it. And 0 of the 64 files that exist are pointed at from the site's own robots.txt, so even the good ones are hard to find.
8 publish a file about the business
Eight files were an account of the business rather than an index of its own pages, and one of the 157 carried a phone number, a place and the opening hours together. It is an appliance repair shop in Des Moines.
56 carry a file a plugin or a platform generated
Forty-seven were a list of the site's own page links, generated by the website builder, opening with the domain name where the business name should be. Eleven still carried the builder's placeholder pages, so an assistant reads that the shop has a page called Location One. Seventeen carried a plugin's credit on the first line, and seven were over 50 KB, past what a model will read.
93 publish nothing at all
A not-found page, an HTML page, or no answer.
02
The same trades in fourteen other states, from Phoenix plumbers to Boston painters
124 businesses, fetched 17 September 2026, as a market study of our own.
36% of the set answer at that address. Look twice: 7% of the set publish a file that is about the business. 3 of the 124 put a phone number, a place and a price or the hours in it. And 0 of the 45 files that exist are pointed at from the site's own robots.txt, so even the good ones are hard to find.
9 publish a file about the business
Three of the 124 carry the facts: an HVAC contractor in Denver and a roofer in Tampa, each with a phone number, an address and real prices in the file, and an electrician in Atlanta with the address and the hours. They are the whole of what a national market has managed.
36 carry a file a plugin or a platform generated
Thirty were a page index. Twenty-one carried a plugin's credit. One file describes a different company from the one whose address it sits at, one lists a staging domain nobody can reach, and the four largest run to 107, 141, 234 and 305 KB.
79 publish nothing at all
A not-found page, an HTML page, or no answer.
03
US sushi, sashimi and fresh seafood restaurants and sellers
60 businesses, fetched 15 September 2026, as a market study of our own.
35% of the set answer at that address. Look twice: 3% of the set publish a file that is about the business. And 0 of the 21 files that exist are pointed at from the site's own robots.txt, so even the good ones are hard to find.
2 wrote their own file
Both were national e-commerce sellers. Not one restaurant in the set had written one, and the set included the names a diner would recognise.
19 carry a file a plugin or a platform generated
Eleven Shopify, four Wix, two from a WordPress SEO plugin that opens by crediting itself, one raw blog feed, and one 246 KB catalog export.
39 publish nothing at all
A not-found page, an HTML page, or no answer.
The score
0 of 281 are AI-friendly.
AI-friendly has a definition here: a site passes all nine checks below. Not one of the 281 trades businesses does. Most sit at three or four, which is a robots.txt that answers, nothing blocked and a sitemap declared: the checklist of 2015, met by accident. The best score in Iowa is six. Nationwide two sites reach eight, and both miss the same check, the one nobody passes.
Every site we ship is 9 of 9 on the day it launches, and the build fails if it would not be.
The file is now a box somebody else ticks for you.
Sort the same 281 by what built the site and the audit explains itself. Every Duda site has a file. Every Wix site has one. Not one of those businesses asked for it, and one of the fifty-one reads as an account of the business. What the builder writes is the site's own page list, opening with the domain name where the business name should be, some still carrying a page called Location One.
An assistant handed that file learns that you have a Thank You page. It does not learn your phone number, and it goes and finds one somewhere else.
| Built with | Sites | Have a file | About the business |
|---|---|---|---|
| WordPress, with an SEO plugin | 137 | 36 | 4 |
| Duda | 41 | 41all | 1 |
| Wix | 10 | 10all | 0 |
| GoDaddy Website Builder | 9 | 6 | 3 |
| Squarespace | 11 | 1 | 1 |
| Custom, or could not tell | 66 | 12 | 7 |
| Shopify, Webflow and others | 7 | 3 | 1 |
| All trades sites | 281 | 109 | 17 |
Platform read from the generator tag and the asset hosts on each home page, 17 September 2026. “About the business” is the count whose file passes check seven below.
The standard
Nine checks. The trades, and us.
Being readable by a machine is not an opinion, so we wrote it down as nine checks a stranger can run on any address in about a minute. Every one of them is a fetch and a read: no judgement, no score out of ten, nothing that depends on liking the site.
The two market columns are the counts from 17 September 2026, 281 businesses in all. The third column is every site we build, and it is not a promise: the build fails if a check would not pass, and the live address is fetched and read after each deploy.
- 01
robots.txt answers, in plain text
The address every crawler tries first returns a text file rather than an error or a styled page. Twenty-eight sites across the two markets have no usable one: eleven answer 404 with a full HTML error page, seven answer 200 with an empty body, and the rest refuse the request or never connect.
- Iowa trades
- 141 of 157
- US trades
- 112 of 124
- Ours
- Every site
- 02
The assistants are not blocked
Nothing in robots.txt shuts out the crawlers that read for ChatGPT, Claude, Perplexity, Gemini and the rest, whether by name or by a blanket rule. Three sites block one by name. One of them shuts out the two agents that fetch a page because a person just asked a question about it, which is a customer standing at the door.
- Iowa trades
- 156 of 157
- US trades
- 122 of 124
- Ours
- Every site
- 03
They are admitted by name
The file names them: GPTBot, ClaudeBot, PerplexityBot, Google-Extended and the others, each with its own group. A wildcard admits them too, so this changes nothing about what is fetched. It states a position that can be read, and its absence reads as not having thought about it.
- Iowa trades
- 9 of 157
- US trades
- 8 of 124
- Ours
- Every site
- 04
The sitemap is declared
A Sitemap line in robots.txt, so a crawler on a budget is handed the list of pages instead of guessing at them. This is the oldest rule here and still the one most often missed.
- Iowa trades
- 119 of 157
- US trades
- 99 of 124
- Ours
- Every site
- 05
robots.txt points at the llms.txt
There is no directive for it, so it is a comment on its own line after the sitemap. Not one site in either market carries it, including every site that has gone to the trouble of publishing a file.
- Iowa trades
- 0 of 157
- US trades
- 0 of 124
- Ours
- Every site
- 06
/llms.txt answers with a real body
A 200, plain text, and something in it. Two Iowa sites answer 200 with zero bytes, which passes every check that stops at the status line, and twelve more hand back a web page with a stylesheet on it.
- Iowa trades
- 64 of 157
- US trades
- 45 of 124
- Ours
- Every site
- 07
The file is the business, not the sitemap
It reads as an account of the business rather than a list of its own page links: no generator's credit on the first line, no builder placeholders, no staging domain, and under 50 KB, which is the point past which a model stops reading. This is where nearly every file that exists falls over.
- Iowa trades
- 8 of 157
- US trades
- 9 of 124
- Ours
- Every site
- 08
It carries the facts a buyer asks for
A phone number, a place, and either a price or the opening hours: the three things somebody is asking the assistant for when your name comes up. Four of the 281 businesses have all three in the file.
- Iowa trades
- 1 of 157
- US trades
- 3 of 124
- Ours
- Every site
- 09
The page says the same thing in structured data
JSON-LD on the home page describing the business, with the telephone in it, so the older reader and the newer one are handed the same record. A quarter of the set publishes no structured data of any kind.
- Iowa trades
- 62 of 157
- US trades
- 67 of 124
- Ours
- Every site
of the trades sites pass all nine
pass the four that decide a citation: let the crawler in, publish a file, make it about the business, put the phone number and the price in it
of nine on every site we build, checked on the live address after each deploy
What it takes
Seven things that are true of every site we ship.
None of this is an upgrade or a package. It is how a site is built here, and the last one is the line between a file written for your customers and a file written for a platform.
- 01
Every fact lives in one place
Your phone number, your hours, your prices and your address are written once, in the site's own content source, and every page reads them from there. A change is one edit. Nothing on the site can disagree with anything else on the site, which is the whole reason a machine can trust it.
- 02
An llms.txt that is a route, not a file
The plain-text summary an assistant reads is generated from that same source on request, so the morning a price changes it changes with it. It is written for the question a buyer actually asks: what you do, what it costs, how to reach you, and what each page holds. It carries no claim the pages do not make.
- 03
Pointed at from robots.txt, listed in the sitemap
A file nobody is told about is a file nobody reads. The site's robots.txt names the sitemap, points at the llms.txt, and admits the assistants' crawlers by name: OpenAI's, Anthropic's, xAI's, Perplexity's, Google's and the rest, each listed rather than left to a wildcard. Every public page is in the sitemap, and the pages that should never be quoted, a private preview or a signed invoice link, are kept out of all three.
- 04
Structured data from the same source
Search engines and assistants both read schema.org markup: the business, its address and hours, each service, each product with its price as an Offer, the questions a page answers, and the trail of pages that leads to it. On our sites that markup is generated from the same records the page renders, so it is never a second copy that quietly goes stale.
- 05
Pages a machine can read without rendering them
Every page is prerendered to real HTML with real headings, real text and described images. A crawler on a budget reads it in one request. A page that only exists after a script runs, or a heading that is a picture of words, is a page that does not exist to half the things that would quote it.
- 06
Verified on the live address, not in the code
Our build fails if a site does not publish the file. After every deploy we fetch the live URL and read the body, because a file that answers 200 with zero bytes passes every check that stops at the status line. We have found exactly that on a live site, and it is why the check reads the body.
- 07
Nothing about AI on your site
The file is a business publishing facts about the business, in plain text, the way a printed menu does. Your pages never mention AI, models or assistants, and neither does the file. That restraint is what makes it read as a source rather than as marketing.
- +
And a strong website underneath
An AI-friendly site is a fast, well-built site with one more reader. Everything the older reader wants is still here:
- Every page prerendered and served from a CDN, so it opens before a visitor decides to leave
- Lighthouse 90 or better on performance, accessibility, best practices and SEO, on a phone, before launch
- A share card on every route, so a link pasted into a text or a feed shows the page and not a blank
- A rebuild keeps the old addresses, with redirects mapped, so the pages you already rank for stay ranked
- Search Console and analytics set up at launch, with the sitemap submitted, so the first crawl is on purpose
- A real 404 in the site's own design that hands a lost visitor the next best page
Why it is the future of search
First in line stays first.
Across the 281 trades businesses in the two audits above, 4 have a file carrying a phone number, a place and a price or the hours. That is the strongest position a business can hold, and in almost every market it is still open.
- At launch
You are the only answer.
When a buyer asks who can do the thing you do, in the place you do it, you are named because there is nobody else to name. Every question becomes a lead with your price already in it.
- Year one
The habit sets.
Every answer that carries your name teaches the machine to carry it again. A customer who found you once finds you every time, and your competitors watch orders go somewhere they cannot see.
- Years two and three
You are the incumbent.
When the rest of the market catches on, they find an established, consistently cited source sitting where they want to be. Unseating that is the hardest job in search, and now it is their job rather than yours.
Go and read one
Anybody can put a claim on a page. Here are the files.
Each of these is a live site we build and host, and each address opens the file an assistant reads. They are plain text, written for the question, and generated from the same content the pages are. Open one and read the first line.
- K&J Shine WorksAuto detailing, Cedar Rapidskjshineworks.com/llms.txt
- Jerome PrintshopPrint shop, Jerome, Idahojeromeprintshop.com/llms.txt
- Deafinitely DogsService dog charity, Hiawatha, Iowadeafinitelydogs.org/llms.txt
- Poo PatrolPet waste removal, Iowathe-poo-patrol.com/llms.txt
- Pineland TechnologiesThis sitepinelandtech.com/llms.txt
Straight answers
The seven questions this page gets.
- What does AI-friendly mean?
- Nine checks, each a fetch and a read, that decide whether an assistant can quote your business: robots.txt answers in plain text, the assistants' crawlers are not blocked and are named, the sitemap is declared, robots.txt points at the llms.txt, the llms.txt answers with a real body, the file is about the business rather than a list of its pages, it carries a phone number, a place and a price or the hours, and the home page says the same thing in structured data. A site is AI-friendly when it passes all nine. Of 281 trades businesses we measured on 17 September 2026, none does. Every site we build does on the day it launches, and the build fails if it would not.
- How do I check whether my website is AI-friendly, or AI-searchable?
- Type the address into the free checker at pinelandtech.com/ai-friendly#check. Our server fetches your home page, robots.txt and /llms.txt, runs the nine checks, and shows you each one with what it found, the first lines of whatever an assistant reads at your address today, and where the score sits against the 281 trades businesses in the study. It takes about ten seconds, stores nothing, and asks for no sign-up. If you would rather read it by hand: open yoursite.com/llms.txt and ask whether a stranger could quote your phone number and a price from it, then open yoursite.com/robots.txt and look for the assistants' crawlers by name and a line pointing at the file.
- What is an llms.txt?
- A plain text page at yoursite.com/llms.txt that states, in order, what the business is, how to reach it, what it sells and for how much, and what each page of the site holds. Assistants read it before they name a business in an answer. On our sites it is generated from the site's own content on request, so it can never disagree with the pages.
- Does this replace search engine optimisation?
- No. It is the same discipline with one more reader. Everything a search engine wanted still applies: fast prerendered pages, real headings, structured data, a sitemap, redirects kept on a rebuild. The file and the crawler policy are added on top, and the same single source feeds all of it.
- Which assistants read it?
- Any that fetches the address. Our robots.txt admits the crawlers of OpenAI, Anthropic, Perplexity, xAI, Google, Apple, Amazon and Meta by name, and points at the file. A wildcard would admit them too; naming them states a position that can be read.
- Will my website mention AI?
- No. The file is a business publishing facts about the business, the way a printed menu does, and the pages never mention AI, models or assistants. That restraint is what makes the file read as a source rather than as marketing.
- I already have a website. Can this be added to it?
- If the site is one of ours, it already has it. If it is not, the honest answer is usually a rebuild, because the file has to be generated from a single content source and a site built any other way does not have one. The free check above tells you where yours stands today.
Free check · no charge, no sign-up
Score your site against the nine checks.
Is my website AI-friendly? Is it AI-searchable? Does ChatGPT see it? Same question, one answer. Type the address and our server fetches the home page, robots.txt and /llms.txt, reads what came back, and scores it the same way the 281 were scored. You see every check, what was found, and the first lines of whatever an assistant reads at your address today.
Three fetches from our server, nothing stored, no sign-up. The same nine checks we ran on 281 trades businesses.
Be the name in the answer.
Tell us what the business does and where. Within a day you will have a straight answer on what a site that gets quoted takes to build, what it costs to run, and whether your market is still open. In most of them it is.
Or skip the form entirely. hello@pinelandtech.com reaches a person, not a queue.