Open-source · 32 real checks · free
Your customers ask ChatGPT, Perplexity and Gemini before they ask Google. If AI can't read your site, you're invisible in the answers they trust. See exactly what every AI sees — and the precise fixes to get cited.
A real browser, not a basic botPro
AI agents now browse, click and buy on behalf of real people — on WordPress, Shopify and custom stacks alike. Smart agent readability drives a real browser-based agent through your pages and scores, 0–100, whether it can truly see your content, follow the layout and complete the actions that matter, like search and checkout. And it's only getting smarter: soon the same agent runs every isready.ai check for you, then fixes what it can and hands you a ready-to-paste prompt for ChatGPT, Claude Code or any LLM to finish the job. Pro turns every failure into a concrete, copy-paste fix.
Powered byagent-browser— an open-source Vercel Labs project.
CLI · open source · no install
The entire scanner ships as an open-source CLI — every check, no install, no signup. One command audits your whole site, `--json` feeds your tooling, `--llm` writes the fix plan, and the exit code fails your CI the moment your score drops. Your full AI-readiness gate, in a single line.
$ npx isreadyai yourdomain.comGitHub Action · README badge · CI gate
Add one step to your workflow: the full report lands in the job summary, the run fails when your score drops below the threshold, and Pro or Team unlocks a live badge for your domain.
IsReadyAI — readiness audit
isreadyai/audit-action
permissions:
id-token: write
steps:
- uses: isreadyai/audit-action@v1
with:
url: ${{ env.DEPLOY_URL }}
# define DEPLOY_URL yourself, e.g. env: DEPLOY_URL: https://yoursite.com
threshold: 80
api-key: ${{ secrets.ISREADYAI_API_KEY }}
# optional — report upload + repo badge (Pro/Team)isreadyai/audit-action@v1 scores every deploy against a threshold and writes the full report to the job summary. Add api-key to also upload the report and unlock a repo-scoped badge (Pro or Team).
Live README badge
%22%3E%0A%20%20%20%20%3Crect%20width%3D%2280%22%20height%3D%2220%22%20fill%3D%22%231f1f1c%22%2F%3E%0A%20%20%20%20%3Crect%20x%3D%2280%22%20width%3D%2227%22%20height%3D%2220%22%20fill%3D%22%2344cc11%22%2F%3E%0A%20%20%20%20%3Crect%20width%3D%22107%22%20height%3D%2220%22%20fill%3D%22url(%23s)%22%2F%3E%0A%20%20%3C%2Fg%3E%0A%20%20%3Cg%20transform%3D%22translate(7%2C%206)%22%3E%0A%20%20%20%20%3Cpath%20d%3D%22M4%200%20L8%204%20L4%208%20L0%204%20Z%22%20fill%3D%22%23b8f53d%22%2F%3E%0A%20%20%3C%2Fg%3E%0A%20%20%3Cg%20fill%3D%22%23fff%22%20text-anchor%3D%22middle%22%20font-family%3D%22Verdana%2CGeneva%2CDejaVu%20Sans%2Csans-serif%22%20font-size%3D%2211%22%3E%0A%20%20%20%20%3Ctext%20x%3D%2246.5%22%20y%3D%2215%22%20fill%3D%22%23010101%22%20fill-opacity%3D%22.3%22%3EAI%20ready%3C%2Ftext%3E%0A%20%20%20%20%3Ctext%20x%3D%2246.5%22%20y%3D%2214%22%3EAI%20ready%3C%2Ftext%3E%0A%20%20%20%20%3Ctext%20x%3D%2293.5%22%20y%3D%2215%22%20fill%3D%22%23010101%22%20fill-opacity%3D%22.3%22%3E92%3C%2Ftext%3E%0A%20%20%20%20%3Ctext%20x%3D%2293.5%22%20y%3D%2214%22%3E92%3C%2Ftext%3E%0A%20%20%3C%2Fg%3E%0A%3C%2Fsvg%3E){kind=small}
Served live by isready.ai for your verified domain (Pro & Team) and re-scanned on a schedule — it stays current on its own, no CI step. Copy the Markdown into your README. Prefer a badge scoped to this repo and branch instead? Add api-key to the workflow above to get one from CI.
Grounded in how AI crawlers actually behave — not SEO folklore.
We request your pages exactly like GPTBot and ClaudeBot do: raw HTTP, no JavaScript, short timeouts, honest user-agent.
We parse the raw HTML exactly as GPTBot and ClaudeBot do — no JavaScript, because they don't run it either. Empty app shells and content that only appears after hydration get flagged.
32 evidence-based checks across 5 dimensions produce a 0–100 score, and every finding ships with a concrete fix, impact and effort.
Every check is sourced from official crawler docs and peer-reviewed GEO research.
robots.txt per AI bot (GPTBot, ClaudeBot, PerplexityBot…), Cloudflare anti-bot challenges, redirects, TTFB, noindex.
Empty app shells, missing semantic landmarks, scripts with no server-rendered text, noscript fallbacks.
JSON-LD presence and types, meta basics, Open Graph, author & E-E-A-T signals, lang attributes.
HTTPS enforcement, TLS validity, HSTS, mixed content.
Content depth, heading structure, statistics & citations — the signals that raise visibility in generative answers.
llms.txt and Content Signals, reported honestly as informational: server logs show AI crawlers fetch llms.txt in ~0.1% of visits — we never score it.
AI assistants are becoming a primary discovery channel. Cloudflare measured AI crawlers at roughly 50 billion requests per day in 2025 — nearly 1% of all web traffic — and GPTBot traffic grew over 300% year-over-year. Yet the rules are different from classic SEO: GPTBot, ClaudeBot, PerplexityBot and OAI-SearchBot read raw HTML only and never execute JavaScript, while Googlebot renders everything. A React or Vue app with client-side rendering can rank on page one of Google and still be a blank page to every AI assistant.
The research is equally clear about content. The peer-reviewed GEO study (Aggarwal et al., KDD 2024) tested 10,000 queries across multiple generative engines and found that adding quotations improved visibility in AI answers by 41%, statistics by 32%, and citing credible sources by up to 115% for lower-ranked sites. As the authors put it:
Generative engines synthesize answers from a handful of sources — the competition is no longer for a ranking position, but for being part of the answer itself.— GEO: Generative Engine Optimization, KDD 2024
Meanwhile, one misconfigured header can erase you: Cloudflare's AI-bot blocking now protects millions of sites, and many owners don't realize it also removes them from ChatGPT and Perplexity answers. Independent server-log studies show 2.5 million+ websites already disallow AI training — a legitimate choice, but one that should be made deliberately, not by accident. That is exactly what this scanner makes visible: 32 evidence-based checks, each with the observed value, the consequence, and the precise fix.
Primary sources: GEO research paper (arXiv) · OpenAI crawler documentation · Anthropic crawler documentation · Cloudflare AI Crawl Control
Plans
Run the full audit and the real-browser agent check for free. The CLI adds the resolution plan and CI gate. Pro and Team unlock web fixes, Ask-your-site chat, live badges, monitoring and automated fix PRs.
Free forever
€19/mo
€49/mo
Your customers ask ChatGPT, Perplexity and Gemini before they ask Google. If AI can't read your site, you're invisible in the answers they trust — no matter how well you rank on classic search. isready.ai shows you exactly what every AI sees and the precise fixes to get cited.
No. GPTBot, ClaudeBot, PerplexityBot and OAI-SearchBot generally read raw HTML without rendering your interface. Browser-capable agents can execute JavaScript and operate controls. We measure these separately as Agent Readability and Smart agent readability.
The standard score is a weighted 0–100 across five crawler and content dimensions. Smart agent readability is a separate 0–100 across visible content, understandable structure, content quality, accessible controls, navigability and browser barriers. Both are versioned and evidence-based.
Yes — the scanner engine and CLI are MIT-licensed open source, the web scan is free, and running npx isreadyai gives you the full audit plus a resolution plan (--llm) at no cost. The hosted dashboard itself is proprietary (source-available under PolyForm Shield, © Smart Squad S.r.l.), not MIT. Pro and Team add the web resolution, live domain badges, Ask your site, monitoring, scheduled re-scans, report history and automated AI fix pull requests.
Public pages only for now. Scanning login-gated SaaS via GitHub-connected source is on the roadmap — vote on it in the repo.
It can't hurt, but don't expect miracles: no major AI provider has confirmed consuming it, and Google has said it doesn't use it. We report it as informational and never let it affect your score.
Yes. Run npx isreadyai yourdomain.com for the full audit in your terminal, with --json for machines and --llm for a fix plan you can paste into an AI coding agent. The isreadyai/audit-action@v1 GitHub Action gates deployments on a minimum score and writes the full report to the job summary. Add api-key (Pro/Team) and grant the job permissions: id-token: write to also upload the report and unlock a branch-stable repo badge for that repo and branch.
Yes, on Pro and Team plans. Add isreadyai/fix-action@v1 to a workflow with an ISREADYAI_API_KEY secret and permissions: contents: write, pull-requests: write plus id-token: write (the run also uploads the CI report and unlocks the repo badge); it scans the URL, runs an AI agent inside your own runner to apply the minimal fixes, and opens a pull request for you to review.
What it measures: Whether AI crawlers can reach, fetch and index the site without being denied, challenged, redirected excessively or served a degraded response.
How the checks work: The scanner fetches robots.txt and the page with standard and AI-specific user agents, then checks bot directives, anti-bot responses, HTTP status, redirects, noindex and snippet restrictions, sitemap discovery, host consistency and response time.
Source code: browse this category's checks · review the scoring model
What it measures: Whether the useful page content exists in the raw server response read by non-JavaScript AI crawlers, rather than appearing only after client-side rendering.
How the checks work: The scanner parses the returned HTML and measures visible text, semantic main content, empty application shells, noscript fallbacks, image alternatives and optional Markdown content negotiation.
Source code: browse this category's checks · review the scoring model
What it measures: Whether machines can identify the page, its subject, language, canonical URL, publisher and content type without inferring them from presentation alone.
How the checks work: The scanner parses JSON-LD and validates useful identity and content types, then checks title, description, canonical URL, Open Graph metadata, author and E-E-A-T signals, lang and hreflang declarations.
Source code: browse this category's checks · review the scoring model
What it measures: Whether an AI system can fetch and cite the page over a valid, consistently secure connection.
How the checks work: The scanner verifies HTTPS and TLS validity, checks the Strict-Transport-Security header and detects insecure HTTP sub-resources embedded in an HTTPS page.
Source code: browse this category's checks · review the scoring model
What it measures: Whether the page contains substantial, clear and evidence-backed material that a generative engine can extract and cite in an answer.
How the checks work: The scanner evaluates content depth, heading hierarchy, statistics, quotations and external citations, content-to-navigation ratio, machine-readable freshness and answer-friendly structures such as lists, tables and question-shaped sections.
Source code: browse this category's checks · review the scoring model
What it measures: Whether a browser-capable agent actually perceives real content in the rendered accessibility tree, rather than an empty shell, a loading state or a challenge page.
How the checks work: After the page renders, the agent reads the accessibility snapshot and measures perceivable text nodes, the amount of content exposed and whether images carry a text alternative.
Source code: browse this category's checks · review the scoring model
What it measures: Whether the rendered page exposes a clear semantic structure an agent can reason about — landmarks and a sensible heading hierarchy.
How the checks work: The agent inspects the snapshot for ARIA landmarks (banner, navigation, main, contentinfo) and a well-formed heading outline with a single h1 and ordered subheadings.
Source code: browse this category's checks · review the scoring model
What it measures: Whether the rendered content is substantive and self-describing — a meaningful title, enough depth and elements that carry accessible names.
How the checks work: The agent evaluates the page title, the depth of the rendered content and the share of elements exposed with an accessible name rather than anonymous nodes.
Source code: browse this category's checks · review the scoring model
What it measures: Whether the interactive controls an agent must operate — links, buttons, inputs — expose accessible names it can target.
How the checks work: The agent collects the interactive elements from the snapshot and measures how many carry a non-empty accessible name versus unlabelled controls.
Source code: browse this category's checks · review the scoring model
What it measures: Whether anything blocks a browser-capable agent from reaching the content — bot challenges, cookie walls, 'just a moment' interstitials or login gates.
How the checks work: The agent scans the rendered title and snapshot for known interstitial and challenge patterns that would stop it before the real content loads.
Source code: browse this category's checks · review the scoring model
