What is agentic engine optimization?
Search engines send people to your site. Answer engines read your site and summarise it. AI agents do something new: they visit your site on a user's behalf, compare you with alternatives and then act — add to cart, request a quote, book a demo or call an API. Agentic engine optimization is the practice of making a website discoverable, parsable, token-efficient, capability-signalled and safely accessible for those agents. If an agent can't get in, can't read the page without a browser, or burns its whole context budget on your markup, it simply moves on to a competitor it can use.
This checker fetches your page the way a typical agent does — raw HTTP, no JavaScript — and probes the files agents look for. For the full strategy, read our complete guide to agentic engine optimization.
How each pillar is scored
Every check is pass, warn or fail and carries a weight; warnings earn half credit and optional “bonus” checks only ever add points. Each pillar gets its own 0–100 sub-score, and the overall score is the weighted total.
- Access control. We parse robots.txt for thirteen AI user-agents and separate search and user-triggered assistants (which should be allowed) from training crawlers (your choice — a warning at most). We also flag bot challenges: a 403, 429 or 503 response, or a Cloudflare
cf-mitigatedheader, means agents see a wall instead of your page. An optional/agent-permissions.jsonearns a bonus. - Discovery. Is there an
/llms.txtwritten in Markdown with an H1, links and a one-line description per link, small enough (under about 5,000 tokens) to read in one go? Is there anllms-full.txtand a reachable XML sitemap? - Parsability. At least 150 words of content in the server HTML, a single H1, no skipped heading levels, a clear statement of what the page is within the first 200 words, a page under 30,000 tokens, and a Markdown version via
.md,?format=md,Accept: text/markdownor arel="alternate"link. - Token efficiency. How many tokens the raw HTML costs compared with the readable text inside it, and whether you declare a token count in a header or meta tag so agents can budget before downloading.
- Capability signaling. A
skill.mdwithnameanddescriptionfrontmatter, plus bonus credit for MCP discovery (/.well-known/mcp.json) and an OpenAPI description. AGENTS.md is a convention for code repositories, so it isn't checked on websites. - Trust & action. Organization schema, FAQPage or HowTo markup, Product/Offer schema with a price on commercial pages, specific link labels instead of a wall of “Learn more”, and prices shown as text rather than inside images.
How it differs from SEO and answer engine optimization
Classic SEO earns a ranking and a click; the person still does the reading. Answer engine optimization earns a citation inside an AI answer — our AI visibility checker covers that. Agentic engine optimization assumes there may be no person on the page at all. The visitor is software working under a token budget, choosing links by their labels, reading prices from structured data and deciding whether it's allowed to act. Many fixes overlap — server-rendered content and good schema help all three — but agents add new needs: Markdown, small files, explicit permissions and machine-readable capabilities.
Quick wins for most Indian business sites
- Allow ChatGPT-User, Claude-User and Perplexity-User in robots.txt and in your CDN or firewall — our robots.txt generator has presets.
- Publish an llms.txt file with a description on every link.
- Show prices as text in rupees, including GST notes, and mark them up with Offer schema.
- Rename generic “Know more” buttons to say exactly where they go.
- Paste the skill.md and agent-permissions.json snippets from your results, then edit them to match what you really offer.
Credit: the idea of optimising sites for agents as discoverable, parsable, token-efficient, capability-signalled and safely accessible comes from Addy Osmani's article Agentic Engine Optimization. This checker turns that five-property framing into six measurable pillars.
Frequently asked questions
What is an agent readiness score?
It's a weighted 0–100 score that estimates how easily AI agents — ChatGPT, Claude, Perplexity and browsing assistants — can get into your site, find your key pages, read them without JavaScript, fit them into their context window, learn what they can do there, and trust the facts enough to act. Access and server-rendered content carry the most weight.
How is agentic engine optimization different from AEO?
Answer engine optimization is about being quoted in an AI answer. Agentic engine optimization goes one step further: it's about being usable by an agent that is completing a task for someone — comparing prices, filling a form, booking a slot or calling your API. An agent needs cheap, structured, permissioned access, not just a quotable paragraph.
Do I need llms.txt, skill.md and an MCP server?
Not all of them. llms.txt and a clean, server-rendered page help every site. A Markdown version and skill.md are cheap wins. An MCP server or OpenAPI file only matters if you have an API or tools an agent could call, so the checker counts those as bonuses rather than failures.
Should I block AI training crawlers?
That's your call. Training crawlers like GPTBot, ClaudeBot and CCBot are separate from the search and user-triggered agents (OAI-SearchBot, ChatGPT-User, Claude-User, Perplexity-User) that fetch your page when a customer asks. You can block training and still allow assistants; the checker only warns about training blocks.
How are tokens estimated?
We divide the number of characters by four, which is a close rule of thumb for English text in most modern tokenizers. We report both the raw HTML an agent would download and the readable text inside it, so you can see how much of each visit is markup overhead.