What is an agent-readable website?
An agent-readable website is a website structured so humans, search engines and AI agents can all understand what the business does, what it sells, what proof exists and which actions are safe to take. The design stays premium for people. Behind it sits a machine-readable layer: clear service pages, schema, llms.txt, markdown versions of key content, canonical business facts, a service catalog and an action map.
That layer matters because the next visitor may not be a person clicking through a menu. It may be an AI assistant comparing providers, summarising services, checking proof, or helping a customer decide who to speak to. An agent-readable website hands that assistant a clean source of truth instead of a visual layout it has to guess at. We build agent-readable websites for South African businesses from Cape Town, and we have delivered systems like this for 35+ companies over 3+ years.
How does an agent-readable website work?
An agent-readable website works as a set of small, consistent files sitting alongside the pages people see. llms.txt comes first: a concise public index pointing AI systems to the most important service pages, documents and next steps. Markdown versions of those pages follow, stripping visual complexity down to summaries, problems solved, workflows, integrations, FAQs and next steps.
Structured data does the classification work. JSON-LD for Organization, Service, FAQPage, BreadcrumbList, ContactPoint and WebPage tells search engines and AI systems what each page is for. A business facts file holds the canonical name, location, services, service areas, target customers and approved descriptions, so every surface says the same thing. An action map defines the safe paths: book a call, request a quote, request a demo, submit a brief, ask a question, start WhatsApp. Nothing in an agent-readable website is AI-only spam. Every file mirrors content a person can read.
What does an agent-readable website replace?
An agent-readable website replaces the guesswork AI systems do when a site is vague, visual-only or loosely structured. On most sites, important services hide inside design sections and vague marketing copy. There is no llms.txt, no markdown library, no service catalog and no canonical facts file. Schema is missing or incomplete. AI assistants cannot tell how to book a call or request a quote.
The result is misunderstanding. AI systems skip pages, describe services wrongly, fail to cite the business, or recommend a clearer competitor instead. An agent-readable website removes that ambiguity. Service pages state the problem solved, outcomes, integrations, process, guardrails and next step. Structured files carry the same facts in a form machines parse without effort. Agent readability does not replace SEO or design. SEO helps people find the business. Agent readability helps AI systems understand, cite and act on the business correctly.
Which tools and files does an agent-readable website connect to?
An agent-readable website connects to the systems a business already runs rather than replacing them. The content layer covers website pages, sitemap, robots, llms.txt, markdown pages, business facts and the service catalog. Those files can be generated from an existing CMS or maintained directly, and served behind Cloudflare so crawler traffic stays visible and controllable.
From there the action layer reaches outward. Enquiry and booking forms feed the CRM, whether that is HubSpot or GoHighLevel. Conversations start through WhatsApp Business Cloud API. Calendars and mailboxes sit in Google Workspace or Microsoft 365. Content and schema jobs run through n8n or Make.com, with OpenAI, Anthropic Claude or Google Gemini drafting summaries a person approves before publishing. Supabase or PostgreSQL stores the service catalog and monitoring data, and Shopify product content plugs in the same way. Monitoring closes the loop, tracking llms.txt requests, AI crawler visits, schema errors and agent-submitted leads.
Is an agent-readable website POPIA safe, and what stays private?
An agent-readable website is a public layer, so the rule is simple: agent-readable never means everything becomes public. Client data, internal strategy, credentials, unpublished pricing, private workflows, CRM records, support tickets and contracts stay out of it. Public content gets structured and useful. Private data stays behind authentication, access control and server-side security.
POPIA shapes the same decisions. Personal information lives in systems the business owns, consent is recorded, and opt-outs are honoured immediately. robots.txt is not security, so sensitive areas get real access controls instead of a polite request. An agent policy layer states what is public, what is off limits, which crawlers are welcome and which actions an agent may trigger. Sensitive actions wait for human approval. Forms are rate-limited, inputs validated and submissions logged. On an agent-readable website the business keeps the final say on every action an AI agent asks for.
How do we start with an agent-readable website?
Starting with an agent-readable website means adding the machine-readable layer, not rebuilding the site from scratch. The first step is an agent-readability audit: we review service clarity, schema, sitemap, robots, FAQs, contact paths, forms, page structure and source-of-truth gaps. That review costs nothing beyond a conversation, and briefing it usually takes under an hour.
The first build is an Agent-Readable Website Pack: llms.txt, roughly ten markdown service pages, a schema pack, a business facts file, a structured service catalog, an AI action map and a monitoring checklist. Everything runs on the business's own domain and accounts, so there is no lock-in and no migration later. Agent readability is not a once-off file, so the pack includes ongoing checks. We have worked this way with 35+ companies across South Africa, and the 4.9/5 Google rating from 17 verified reviews comes from exactly this approach.
Related capabilities. The same parts, your business.
Keep reading. Pages close to this one.
Tell us what runs slow. We build what fixes it.
Send one message describing how your website is structured today. We reply with an honest read on what an agent-readable website can fix, what it will take, and whether it is worth building.