A growing share of the people who find us never see our home page first. They ask an assistant, and the assistant reads the web on their behalf. If it can't read our site cleanly, it guesses, and guesses about a studio's clients, dates and work are exactly the kind of thing that goes wrong.
So we treated AI assistants as a first-class audience and made vuild.com easy for them to read, cite and talk to. Everything below is generated from the same content as the site itself, so it can't drift out of date.
A map written for language models
The first step was llms.txt, a short plain-text summary at the root of the site: who we are, what we do, our case studies and projects with dates, and links to everything else. Beside it sits llms-full.txt, the long version, with every case study and project in full, the FAQ and what clients say about us.
Both are built from the site's content on every change, so they always match what's on the pages.
Markdown twins
Every main page now has a markdown twin: the same facts as the HTML, without the layout. The AmpUp case study lives at /ampup for people and at /ampup.md for agents that prefer plain text. Each page links to its twin in its head, and llms.txt lists them all.
Structured facts, not just prose
Search engines and assistants lean heavily on structured data, so we added more of it:
- FAQ markup on the About page, answering the questions people actually ask about us.
- Reviews: client testimonials are marked up as named reviews.
- Real relationships: case studies link clients to their sites and, where there is one, their Wikipedia page, and only name genuine backers as backers.
- Active periods on every case study and project, so "when did Vuild work on this?" has a clear answer.
Telling crawlers what changed
Freshness matters. Our sitemap now dates each page by its last real change, taken from its version history, rather than stamping everything with today's date. Our robots file names the AI crawlers we welcome, so the intent is unambiguous.
And on every deploy, the site pings IndexNow with the pages that changed that day. IndexNow is an open protocol supported by Bing and others, which in turn feeds several AI search products.
Letting other agents talk to ours
The last step goes further than being readable. Our site has a concierge, the assistant behind the Ask Vuild button, and it now speaks A2A, the open Agent2Agent protocol. Another company's agent can find it through our agent card at /.well-known/agent-card.json, see what it can help with, and send it a message.
POST /api/a2a
Content-Type: application/json
{"jsonrpc": "2.0", "id": 1, "method": "message/send",
"params": {"message": {"role": "user", "parts": [{"kind": "text", "text": "What does Vuild do?"}]}}}The concierge answers from the same site content as llms-full.txt, and remembers the thread of a conversation. It sits behind the same protections as the chat on the site: rate limits, strict request checks and clear boundaries on what it will and won't do.
Why it matters for our clients
This is the work we now do for clients, so doing it for ourselves first was the obvious move. The questions are the same for any business: can an assistant find you, understand what you do, and describe you accurately? Can other agents do business with yours?
If you'd like the same for your product, start a project, or just ask the concierge on any page. It knows where to send you.