AI Readiness Audit · llms.txt, AI crawler access, structured data, agent files

Can AI search engines and agents actually use your website? This actor checks what they see: whether ChatGPT, Claude, Perplexity, Gemini and Copilot crawlers are allowed in, whether an llms.txt exists and follows the spec, whether the page carries structured data and a real title and description, whether the content is there without running JavaScript, whether a Markdown version is served, and whether agent discovery files (A2A agent card, MCP server card, OpenAPI) are published. Every site gets a 0–100 score, category scores and a prioritised fix list with the evidence behind each finding. Up to 1,000 sites per run, about ten small requests per site, no browser. The first audit each month is free for every account.

Built by agentsgearbox. Questions: support@agentsgearbox.com.

Why this matters now

AI answers send readers before Google does for new sites, and they can only cite what they can read. A page that renders only in JavaScript is invisible to most AI fetchers; a robots.txt that blocks OAI-SearchBot or PerplexityBot removes you from AI answers, not from training; llms.txt is now checked by Chrome's Lighthouse and published by OpenAI, Anthropic and Google for their own docs. Agencies sell "GEO readiness" work on exactly these points; this actor is the measurement.

What you get

One audit item per site:

{
  "site": "docs.apify.com",
  "status": "OK",
  "score": 83,
  "grade": "B",
  "accessScore": 100,
  "guidanceScore": 100,
  "structureScore": 17,
  "readabilityScore": 100,
  "performanceScore": 100,
  "agentScore": null,
  "llmsTxt": "valid",
  "aiSearchBotsAllowed": "11/11",
  "jsOnly": false,
  "topFix": "Add JSON-LD for Organization or WebSite on the homepage and Product, Article or FAQPage where they apply; it is the most machine-readable description of what the page is.",
  "fixes": [
    { "check": "structured-data", "severity": "high", "fix": "Add JSON-LD for Organization or WebSite on the homepage…" },
    { "check": "canonical", "severity": "medium", "fix": "Add <link rel=\"canonical\"> so agents and engines keep one URL per page." }
  ],
  "checks": [
    { "id": "ai-search-access", "category": "access", "label": "AI answer engines allowed to read the homepage", "status": "pass", "value": "11/11", "evidence": "All evaluated answer-engine crawlers may fetch the homepage.", "fix": null, "weight": 6 },
    { "id": "llms-txt", "category": "guidance", "label": "llms.txt present and valid", "status": "pass", "value": "valid, 12 section(s), 467 link(s)", "evidence": "\"Apify documentation\" with 467 links, 5/5 sampled links resolve", "fix": null, "weight": 6 },
    { "id": "content-without-js", "category": "readability", "label": "Main content present without JavaScript", "status": "pass", "value": "10726 words in raw HTML", "evidence": "Text-to-HTML ratio 31.4%", "fix": null, "weight": 8 },
    { "id": "markdown", "category": "readability", "label": "Markdown version available", "status": "pass", "value": "content negotiation (Accept: text/markdown)", "evidence": "text/markdown; charset=utf-8", "fix": null, "weight": 2 }
  ],
  "facts": {
    "homepage": { "finalUrl": "https://docs.apify.com/", "status": 200, "redirects": 0, "ttfbMs": 250, "bytes": 184233, "botProtected": false },
    "robots": { "present": true, "sitemaps": ["https://docs.apify.com/sitemap.xml"], "contentSignals": [], "bots": { "GPTBot": { "allowed": true, "rule": null, "matchedGroup": "*" } } },
    "llmsTxt": { "present": true, "valid": true, "title": "Apify documentation", "sections": 12, "links": 467, "linksChecked": 5, "linksOk": 5, "fullPresent": true, "problems": [] },
    "sitemap": { "present": true, "kind": "urlset", "urls": 710, "declaredInRobots": true },
    "structure": { "title": "Apify Documentation", "description": "…", "canonical": null, "lang": "en", "h1Count": 1, "jsonLdTypes": [], "metaRobots": null },
    "readability": { "words": 10726, "htmlBytes": 184233, "textRatio": 31.4, "jsOnly": false, "spaShell": false },
    "markdown": { "negotiated": true, "indexMd": false },
    "agentFiles": { "/.well-known/agent-card.json": { "present": false, "status": 404 }, "/.well-known/mcp/server-card.json": { "present": false, "status": 404 } }
  },
  "auditedAt": "2026-10-17T15:40:12.118Z"
}

Plus one free summary item per run (sites audited, average score, grade distribution, the five most common failures) and a free error item for any site that could not be reached at all.

The checks

Category (weight)	Checks
Access (25)	robots.txt reachable and plain text; which AI answer-engine crawlers may read the homepage (OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot, Perplexity-User, Google-Extended, Googlebot, Bingbot, Applebot-Extended, DuckAssistBot); homepage not noindex; Content-Signal declarations; noai tags (reported, not scored). Training crawlers (GPTBot, ClaudeBot, CCBot, Bytespider, Amazonbot, meta-externalagent) are reported as the owner's policy and never scored.
Guidance (20)	llms.txt present and valid to the v2 spec (H1, blockquote summary, H2 sections with - [name](url): note lists), up to five of its links verified; llms-full.txt; XML sitemap (index or urlset, URL count) and whether robots.txt declares it.
Structure (20)	title, meta description, canonical, html lang, exactly one H1, Open Graph, JSON-LD structured data (types listed, invalid blocks counted).
Readability (20)	main content present in the raw HTML without JavaScript (word count, text ratio, empty app-shell detection, "enable JavaScript" messages); Markdown version via Accept: text/markdown or /index.md.
Performance (10)	homepage reachable without a bot challenge, HTTPS, time to first byte, HTML weight.
Agent files (5)	A2A agent card (/.well-known/agent-card.json), MCP server card, agent-skills manifest, OpenAPI description, security.txt, TDM reservation file, ai.txt. Their absence is never scored against a site (fewer than 0.01% have them); their presence earns the bonus.

Scoring: pass = full weight, warn = half, fail = none; categories with no scorable result are left out instead of counted as zero. Grades: A ≥ 90, B ≥ 75, C ≥ 60, D ≥ 40, F below.

Input

Field	Type	Default	Description
urls	string[]	required	Homepages or bare domains, up to 1,000. A URL with a path (e.g. /docs/) also looks for a path-level llms.txt first, as the spec allows.
aiBots	string[]	17 crawlers	User agents evaluated against robots.txt.
checkLlmsLinks	boolean	true	Fetch up to five links from llms.txt and count how many resolve.
includeFixes	boolean	true	Add the prioritised fix list.
concurrency	integer	6	Sites audited in parallel.
timeoutSecs	integer	20	Per-request timeout.

Pricing

Try it free. The first 1 audit each calendar month are free for every Apify account, with nothing to enter: the actor keeps count in a small key-value store in your account (free-slice-ai-readiness-audit) and charges only from the next one on. The free results are real results you keep; the counter resets on the first of each month.

Pay per event: $0.02 per site audited (site-audit). Summary and error records are free; a site that cannot be reached costs nothing. Apify plans above Free get an automatic discount on every event (10% to 40%).

Run	Cost
Audit one site	$0.02
Audit a 50-site client portfolio	$1.00
Weekly audit of 1,000 domains for a prospect list	$20.00 a week

A single site takes 2 to 5 seconds; 1,000 sites finish in a few minutes.

Example tasks

Ready-made configurations you can open, copy and run as they are:

• Audit a portfolio of websites for AI search and agent readiness: six well-known sites scored side by side, from a JavaScript-only homepage graded D to fully readable sites graded A, each with its fix list.

Run it from your code

Node.js

import { ApifyClient } from 'apify-client';
const client = new ApifyClient({ token: process.env.APIFY_TOKEN });
const run = await client.actor('agentsgearbox/ai-readiness-audit').call({ urls: ['https://example.com', 'competitor.com'] });
const { items } = await client.dataset(run.defaultDatasetId).listItems();
for (const a of items.filter((i) => i.type === 'audit')) console.log(a.site, a.score, a.grade, a.topFix);

Python

from apify_client import ApifyClient
client = ApifyClient(token=os.environ["APIFY_TOKEN"])
run = client.actor("agentsgearbox/ai-readiness-audit").call(run_input={"urls": ["https://example.com"]})
for item in client.dataset(run["defaultDatasetId"]).iterate_items():
    if item.get("type") == "audit":
        print(item["site"], item["score"], [f["fix"] for f in item["fixes"][:3]])

cURL

curl -s -X POST "https://api.apify.com/v2/acts/agentsgearbox~ai-readiness-audit/run-sync-get-dataset-items?token=$APIFY_TOKEN" \
  -H 'Content-Type: application/json' -d '{"urls":["https://example.com"]}'

For AI agents

Give your agent hands. Instead of fetching a dozen files and reasoning about robots syntax, an agent calls this actor and gets a verified audit back as JSON, priced per site. Auditing one site by hand means reading robots.txt rules for seventeen crawlers, validating llms.txt against the spec, parsing JSON-LD and measuring the page without JavaScript: twenty minutes for a specialist. This actor returns it in about three seconds for $0.02.

• MCP, this actor only: add https://mcp.apify.com?tools=agentsgearbox/ai-readiness-audit to Claude Desktop, Claude Code, Cursor or any MCP client (with your Apify token), then ask for "an AI readiness audit of example.com with the top three fixes".

• MCP, all agentsgearbox actors: https://mcp.apify.com?tools=agentsgearbox/telegram-channel-scraper,agentsgearbox/google-play-reviews-scraper,agentsgearbox/substack-scraper,agentsgearbox/google-news-scraper,agentsgearbox/brand-name-generator,agentsgearbox/ai-readiness-audit,agentsgearbox/company-lookup,agentsgearbox/app-intelligence,agentsgearbox/app-store-reviews-scraper

• No Apify account? This actor is eligible for Apify's agentic payments: an agent can buy a spend-capped token through agi.apify.com (x402, MPP or Skyfire) and call the API directly.
• Prices are per result and known in advance; unreachable sites come back as free error records instead of failed runs.

Reading the results

• jsOnly: true is the single most important finding: the homepage HTML carries almost no text, so AI fetchers that do not run JavaScript see an empty page. Server-render or pre-render the pages you want cited.
• aiSearchBotsAllowed counts the answer-engine crawlers allowed to fetch the homepage. Blocking them keeps you out of AI answers; blocking training crawlers is a separate decision and does not affect the score.
• llmsTxt is valid, present, with issues (the problems are listed in facts.llmsTxt.problems) or missing.
• status is OK, BOT_PROTECTED (the homepage answered with a bot challenge; everything a plain client cannot read is marked unknown) or HTTP_<code>. Sites that cannot be audited are free error rows instead: UNREACHABLE (DNS, connection or timeout), NOT_HTML (the URL serves a PDF, text or JSON file, not a page) or ERROR (invalid URL, private or local host). Different spellings of the same page (http://, trailing slash, upper case, www.) are audited and charged once.
• Each check carries its evidence so you can verify a finding before acting on it.

Limitations, stated plainly

• Measurements are taken from a European datacenter with a plain HTTPS client. Sites that serve different content to bots, or that pass bot challenges only for browsers, are reported as BOT_PROTECTED rather than guessed at.
• "Content without JavaScript" is measured on the raw HTML; pages that load text through JavaScript are treated as invisible, because that is how most AI fetchers treat them.
• Only the homepage (or the URL you give) is analysed for structure and readability; the audit is a site-level readiness check, not a crawl of every page.
• Discovery-file conventions for agents (A2A, MCP cards) are young and still changing; the paths checked are the current ones and are listed in the output so you can see exactly what was probed.

Troubleshooting

Symptom	Cause	Fix
BOT_PROTECTED	Cloudflare or similar challenged the request	Allow verified AI crawlers through, or audit a staging URL without protection
robots.txt reachable: warn, HTML returned	The server returns a web page instead of a text file at /robots.txt	Serve a plain-text robots.txt with HTTP 200
llms.txt present, with issues	File exists but misses the H1, the link lists or has broken links	Follow the problems listed in facts.llmsTxt.problems
agentScore: null	No agent discovery files found	Expected for almost every site today; nothing to fix unless you expose an API or agent

More from agentsgearbox

• Telegram Channel Scraper: public Telegram channels as JSON, no API key.
• Google Play Reviews Scraper: Google Play reviews with replies and versions, any market.
• Substack Scraper: Substack posts with full text and comments, custom domains too.
• Google News Scraper: Google News with real publisher URLs and full text.
• Brand & Business Name Generator: brand names verified live on domains, handles and app stores.
• Company Lookup: one row per domain with legal identity and VAT validation, contacts, socials, tech stack, open jobs, apps, news and Wikidata facts.
• App Intelligence: App Store and Google Play in one run: search, top charts, full app details and each app's twin on the other store.
• App Store Reviews Scraper · iOS + Google Play, Any Country: Reviews of any iOS or Android app from both stores in one run: stars, text, version, votes, developer replies, since-last-run checkpoint, rating trend per app..

All of them are pay-per-event, run without a browser, and are compared with their alternatives on agentsgearbox.com.

Support

Email support@agentsgearbox.com or use the issues tab. Compare this actor with alternatives on agentsgearbox.com.