Smart websites · AI-readable build
Built so an AI can read it, quote it, and get it right.
More of your customers now ask an assistant before they open a search box. We build the site so the assistant can find your business, read the facts, and quote them correctly — the same way we built this one.
Priced to scope · Published mechanics · Audited on our own site · No third-party requests
The question moved. It gets asked to a machine now.
Same question your customers always had. Different place they type it.
Your bathtub won't drain. Ten years ago you typed plumber near me and picked from a list. Now people open an assistant and type it the way they'd say it — "my bathtub won't drain, who near me can fix it today", or "who's the best lawyer in my city for my case" — and get one paragraph, not ten links.
That paragraph comes from somewhere. The assistant either reads a page it can understand, or fills the gap with a stale listing, a competitor's write-up, or a guess at your prices.
Before the pitch, the caveat
We ran that exact question ourselves on 29 August 2026 and read what came back. Not one small local plumber ranked — national DIY articles, a drain-product brand, a few large multi-location firms. Generic questions get generic answers assembled from big aggregations, so we won't tell you we can put you at the top of one.
The buildable promise is narrower: when someone asks an assistant about your business, your service, your prices or your area, it finds a page it can read and quotes it correctly instead of guessing.
What AI-readable actually means.
Four things, each one checkable on the finished site.
One
Every fact ships in the HTML
Services, prices, phone, hours, area and your common answers — all in the bytes the server sends, not assembled by JavaScript after load. A crawler that never runs a script still gets the page.
Two
Answers in the words people ask
Your FAQ written in your customers' phrasing, not your brochure's, each answer short enough to lift whole. An assistant quotes a sentence; it does not interpret a table.
Three
Structured data, sold as what it is
Schema so your name, number and address resolve to one business across every page. Ahrefs tested the citation claim directly — 1,885 pages that added JSON-LD against about 4,000 controls, August 2025 to March 2026 — and found no uplift. Schema is for clean extraction, not for citations.
Source: Ahrefs, 11 May 2026 — ahrefs.com/blog/schema-ai-citations
Four
Crawler access made explicit
A robots.txt that names AI crawlers instead of leaving them to a wildcard, a sitemap with real URLs, and an llms.txt because it costs an hour — a convenience, not a channel. The crawlers that matter read your HTML.
What we measured on our own site.
Not a case study. An audit of this domain, run 28 August 2026, with numbers we produced.
We have no smart-website client to show you — AOS AI has three clients, all phone-agent work. What we have is this site, built the way we'd build yours, then measured.
- The full text ships before JavaScript runs — the raw HTML carries more words than the rendered page does. The script is net-subtractive.
- Nothing is JavaScript-only: zero rendered-only content on the private-AI page, and every apparent case on the homepage traced back to the raw HTML.
- With JavaScript off — a crawler's failure mode — the pages get longer, not shorter, with every answer visible.
- Seven requests on the homepage: the document, five self-hosted fonts, one inline image. Zero external hosts, zero failed requests, zero console errors, zero full-viewport overlays.
- Sixteen AI crawlers named one at a time in robots.txt, search and training, each explicitly allowed — policy on the record, not a wildcard's accident.
The part nobody publishes
Three defects, ours, from the same audit: the private-AI page shipped no structured data, the share image 404'd on the live domain, and that domain's sitemap was empty. We publish the failures — an audit that only finds good news is not an audit. We run the same audit on our own site, and on yours before handoff.
Still being read — measured, not claimed
We record every request an AI system makes to this site, at the edge, before any JavaScript runs. It is the same measurement we install for you.
Live figures load with JavaScript. Without it, this block stays honest and empty rather than showing a number we cannot verify for you here.
What we build for you.
Plain steps, in the order they happen.
The pages. Rebuilt as static, self-contained HTML, with content, prices, contact details and service area in the served bytes.
The answers. Your FAQ written from the questions customers ask on the phone, in their words, one liftable answer each.
The structure. Schema and entity linkage so every page resolves to one business, plus robots.txt, sitemap and llms.txt.
The audit. The same parity and AI-crawler pass we ran here, run on yours, with the report handed over rather than summarized into a score.
The half that is not your website. Google Business Profile, Bing Webmaster Tools, and the same name, address and phone across every listing — a site that only cites itself is weak evidence. This part needs you; we tell you what to do.
What we will not promise.
Most of the difference between us and the rest of this market is here.
No rankings, no positions, no citations.
There is no ranking to buy inside an assistant's answer, no submission form, no schema tag that promotes you. Nobody can guarantee an AI citation — we don't either, and neither can the vendor promising one.
Generic category questions are not the goal.
"Best plumber near me", "best lawyer in town" — assistants answer those from large aggregations, and a small firm will not displace them. We won't take your money to chase one.
No percentage we cannot show the working for.
This market is full of pages quoting a precise lift in AI citations with no study attached. We publish the mechanism and the measurements — and when the evidence says a thing doesn't work, as Ahrefs' does of schema, we publish that too.
Retrieval moves in weeks. A model's memory does not.
We build for retrieval: the assistant searching, fetching and reading your page, which follows indexation and moves in weeks. What a model already knows from training moves over months to years, and is not for sale.
Same rule elsewhere on this site: our private AI page refuses to claim compliance, for exactly this reason.
What you can check yourself.
Every claim above is falsifiable. Here is how.
View source on any page we build and search it for your phone number and your prices. If they're there, they ship before JavaScript. Turn JavaScript off and reload: the page should still read start to finish. Check Search Console and Bing Webmaster Tools for indexation — that's the door most assistants come through.
Then ask. Put ten questions a real customer would ask to ChatGPT, Perplexity, Claude and Gemini, write down what comes back, and repeat every 90 days. There is no rank tracker for assistant answers. A vendor selling you an AI visibility score is selling a number they invented.
What it costs.
Priced to scope after the 21-minute call, before you commit. No public number for this one, and we won't invent a range: it depends on how many pages you have, what they run on now, and how much of the off-site half you want us to handle. Our two off-the-shelf agents do carry published prices, in the pricing section. This one gets quoted once we've read your site.
Eight questions people actually ask.
Be findable by the crawlers and indexes ChatGPT reads, and give it a page it can quote: facts in the served HTML, answers in your customers' phrasing, schema, an explicit robots.txt, indexation in Google and Bing. There is no submission form.
A site whose facts — services, prices, phone, hours, area — ship in the HTML the server sends, in short quotable sentences, marked up as one entity, and open to AI crawlers by an explicit rule.
No, on the best evidence available. Ahrefs tested 1,885 pages that added JSON-LD and found no citation uplift (ahrefs.com/blog/schema-ai-citations). We write schema for extraction, never as a ranking lever.
Very little on its own. It is a proposed convention, no major assistant has confirmed it consumes one, and the crawlers that matter fetch your HTML directly. We ship it because it takes an hour, and never sell it as the mechanism.
No. Nobody can. Assistant answers are generated fresh from sources picked at answer time, and there is no placement to buy. What we control is whether your page is findable, readable and quotable when it's picked.
Retrieval — the assistant searching and reading your page — follows indexation, so it moves in weeks once Google and Bing have the site. What a model knows from training moves over months to years.
Mostly the same work for a different reader. Being in the index still decides everything. What changes is emphasis: facts shipped before JavaScript, answers in the customer's phrasing, sentences short enough to quote whole.
Often we can fix it. If your site already ships content in the HTML, the work is structure, answers and access. If it renders everything with JavaScript after load, rebuilding is usually cheaper than patching.
Something not answered here? Call 646-846-9369 · help@aosai.co
One idea a week.
One useful AI idea for your business. Weekly. Unsubscribe anytime.
That didn't go through. Email help@aosai.co and we'll add you by hand.
JavaScript is off. If this doesn't go through, email help@aosai.co and we'll add you by hand.
No third-party embed. Your address is stored by us and used only for the newsletter. See privacy.
21 minutes. Bring the site you already have.
We'll read it on the call, tell you what an assistant can and cannot see in it, and price the work to scope. If it's already fine, that is the answer.
The site is where your AI receptionist lives · Priced to scope