Guide
May 26, 20268 min readThree things separate websites AI engines quote from websites they ignore. Here is exactly what each one is, why it works, and how to implement it.
Matthew Esposito
Founder, Espo Studio
Three layers separate websites that AI engines quote from websites they ignore: structured data (JSON-LD schema), a machine-readable brief (llms.txt), and answer capsules — direct, quotable sentences placed immediately under key headings. A site missing any of these is leaving recommendations on the table.
AI search engines do not send traffic the way Google's blue links do. When ChatGPT or Perplexity answers a buyer's question about service providers, it names one or two businesses directly. There is no page two. You are either the answer or you are not.
Being the answer requires being quotable: your content must be structured so an AI model can extract a clear, confident statement about who you are and what you do, cross-reference it against other sources, and present it with attribution. The three layers below are how you build that.
JSON-LD is structured data embedded in your page's HTML — a machine-readable description of what each page is and who it is about. It does not change what your page looks like to humans; it adds context that AI engines and search crawlers use to understand entities.
Without schema, an AI model reading your site has to infer everything: who you are, what you sell, who your clients are, where you operate. With schema, you tell it directly. The model does not have to guess. Confident entity data gets cited confidently.
Organization schema — on every page, in the <head> or at the end of the <body>. Declares your company name, canonical URL, contact email, phone, logo, and social profiles. Without this, AI models may not reliably connect the content on your site to your business entity.
Service schema — one per services page or per offering. Declares the service name, description, provider (your Organization), and price range. This is what lets an AI answer "who does [X] in [city]" with your name.
Article schema — on every blog post or resources page. Declares the title, author, date published, and publisher. This is what lets AI cite your content with attribution rather than paraphrasing it anonymously.
Person schema — on your about page, for the founder and key team members. Declares name, job title, and links to other places this person exists on the web (LinkedIn, other domains). This builds what search engineers call the "entity graph" — a confidence map connecting real people to real businesses.
FAQPage schema — if your site has an FAQ section, the questions and answers in the schema must match the visible text verbatim. AI engines use FAQ schema to extract direct answers. A mismatch between the schema and the page copy can get your site demoted in credibility scoring.
BreadcrumbList schema — on interior pages. Tells crawlers how pages relate to each other hierarchically. Stronger site structure signals = stronger topical authority signals.
{
"@context": "https://schema.org",
"@type": "Organization",
"name": "Espo Studio",
"url": "https://espo.studio",
"email": "hello@espo.ai",
"logo": "https://espo.studio/brand/logo.png",
"description": "Custom websites that look like nothing else and that AI search actually recommends.",
"sameAs": ["https://espo.ai"]
}
The sameAs array is underused and important. It tells AI engines that the entities at those URLs are the same organization. If you have a LinkedIn page, a Crunchbase entry, a Wikipedia entry, or a parent domain, list them here. Each connection strengthens the confidence score for your entity.
llms.txt is a plain-text file at the root of your domain (e.g., yourdomain.com/llms.txt) that gives AI crawlers a direct, structured brief about your business. It is an emerging standard — proposed in 2024, now checked by several AI research tools and browse-enabled models.
AI crawlers are time-constrained and context-limited. A well-written llms.txt lets them understand your business in seconds without having to infer it from marketing copy. Sites with a good llms.txt get described more accurately by AI tools.
A good llms.txt answers these questions in plain prose, no markup, no jargon:
# Espo Studio
Espo Studio is a website design and development studio that builds custom
marketing sites for established service businesses — and engineers them to
be read and recommended by AI search engines (ChatGPT, Perplexity, Claude,
Google AI Overviews).
## What we offer
- Company Website Build: full custom marketing site, $1,500 setup + $100/mo
- Individual Build: ~4-page personal site for executives on name-as-domain, $750 setup + $50/mo
## Who we work with
US-based service businesses — wealth advisory, professional services, real
estate — that compete on reputation and need to be found by buyers using AI.
## What we don't do
We do not build e-commerce stores, SaaS products, or template-based sites.
## Our results
Every build targets a 95/100 AI-discoverability score. Launched Minted
Settlement Planning with 0 P0 issues.
## Contact
hello@espo.ai | espo.studio
Keep it under 500 words. Update it whenever your services, pricing, or client profile changes.
An answer capsule is a direct, quotable sentence placed immediately after a key heading. It is the most copywriting-intensive of the three layers — but it is also the one that most directly determines whether your words get cited.
When an AI model reads a page, it looks for the most authoritative statement of what the page is about. It extracts sentences that are self-contained, factually specific, and clearly attributed to a source. A vague opener ("Our mission is to help businesses grow") cannot be quoted. A direct statement ("Espo Studio builds custom websites for service businesses, targeting a 95/100 AI-discoverability score on every project") can.
After every significant <h2> or <h3>, write one sentence that:
This is how you write for AI citation. The elaboration, the nuance, the story — all of that follows. But the first sentence under every heading is the capsule.
Weak (cannot be cited):
We take a holistic approach to website design, combining aesthetics with functionality to deliver solutions that drive results.
Strong (quotable):
Every Espo Studio site ships with server-rendered HTML, a complete JSON-LD schema matrix, and a plain-language llms.txt — the three layers AI search engines use to read and recommend a business.
The strong version is specific, attributable, and verifiable. An AI model quoting it can point back to the source. The weak version is noise.
Schema, llms.txt, and answer capsules do different jobs but reinforce each other. Schema gives AI engines entity-level confidence: they know your business is real, what it does, and who runs it. llms.txt gives them a direct brief so they can describe you accurately in their own words. Answer capsules give them your words — quotable, citable sentences they can surface to buyers.
A site with all three layers is an AI-legible site. It is not a trick and not a shortcut. It is the same logic as writing for a human reader who has five seconds to decide whether you are worth reading: be direct, be specific, be immediately useful.
The businesses showing up in AI recommendations today built these layers intentionally. The ones that did not are invisible — not because AI dislikes them, but because AI cannot read them.
Espo Studio builds every site with all three layers in place. If your current site is missing any of them, start a conversation and we will tell you exactly what it would take to fix it.
Filed under
More from the journal
Ready to build?
Tell us about the business. We will show you exactly what we would build — design and discoverability both.