How we make sites AI-agent-ready (audit & setup)

AI agents already browse the web. Is your site ready?

ChatGPT with tool-use, Perplexity, Claude, Gemini-in-Chrome — all of them already read sites and take actions "on the user's behalf." Next week a prospect will tell an AI: "find a marketing agency in Chișinău and put a call on my calendar." If the agency's site isn't agent-ready, the request dies at step 2-3.

In March we ran smmart.md through [isitagentready.com](https://isitagentready.com) — a free audit covering 10+ standards for AI agents. We were at ~40%. Here's what we shipped to clear 90%+.

What an agent-ready audit checks

- `llms.txt` — the plain-text site map agents read first - `robots.txt` with an explicit allowlist for GPTBot, ClaudeBot, PerplexityBot, Google-Extended - JSON-LD schemas (Organization, BlogPosting, BreadcrumbList) — agents parse them as structured data - `/.well-known/agent-skills/index.json` (RFC v0.2.0) — a catalog of actions agents can call - Markdown content negotiation — `Accept: text/markdown` returns markdown instead of HTML - WebMCP — `navigator.modelContext.provideContext()` in the browser - `/.well-known/security.txt` — disclosure path for researchers

What we shipped on smmart.md

Agent-readable content. `/llms.txt` lists 21 marketing pages + 7 articles across three locales. Sitemap.xml with 39 URLs. Every page carries schema.org JSON-LD (Organization, Service, BreadcrumbList, BlogPosting).

Markdown for Agents. Send `Accept: text/markdown` to any page and you get the markdown source instead of HTML. `curl -H "Accept: text/markdown" https://smmart.md/en/ideas/playbook-smm-moldova-2026` returns the markdown body directly.

Agent Skills. `/.well-known/agent-skills/index.json` declares two skills: `submit-lead` (POST /api/lead with a JSON schema) and `browse-articles`. Agents know which actions they can take on the site.

WebMCP. In Chrome with an active AI helper, `navigator.modelContext` sees `submit-lead` as an inline tool. The agent can file a brief directly, without filling in the visual form.

Why it matters for a local brand

Today: ~1-3% of B2B leads arrive via AI search (ChatGPT, Perplexity, AI Mode). Twelve months out, analysts forecast 10-15%.

The brands that show up first in AI results have:

- Clean schema markup (Google scrapes it for AI Overviews) - An up-to-date `llms.txt` - A response to `Accept: text/markdown` - Descriptions and meta tags optimised to 150-160 chars

This is not 2010-era SEO. This is AI SEO for 2026.

What we do for clients

A full audit via isitagentready.com plus per-platform implementation (WordPress, Webflow, Next.js, Shopify):

1. `llms.txt` + `robots.txt` with an AI-bot allowlist
2. Schema.org Organization + LocalBusiness + per-page schemas
3. Markdown content negotiation (Cloudflare, Vercel, Netlify edge middleware)
4. Agent Skills index if the site exposes public actions (leads, booking, search)
5. WebMCP provider where it earns its keep (Chrome-heavy brands)
6. Security headers + `.well-known/security.txt`

smmart.md is our own proof — pushed from ~40% to 90%+ on isitagentready in three days.

[Send a brief](/contact) if you want the same audit + setup on your site.

Key takeaways

- AI agents (ChatGPT, Perplexity, Claude, Gemini-in-Chrome) already browse the web on users' behalf. - Agent-ready checklist: `llms.txt`, robots allowlist, JSON-LD schemas, `agent-skills`, markdown negotiation, WebMCP, `security.txt`. - Today ~1-3% of B2B leads come via AI search; analysts forecast 10-15% within 12 months. - Markdown content negotiation (`Accept: text/markdown`) is the cheapest agent-ready win. - smmart.md moved from ~40% to 90%+ on isitagentready in three days — same process for clients.