Does AI find, read, and recommend your site?
Your buyers increasingly ask an assistant before they search. Enter a page and the email the report should go to; in about ten seconds you get the ten things ChatGPT, Claude, Perplexity, and Google check before they find, read, and trust a site, each judged Pass, Needs work, or Critical with the fix. Then how fast it loads on a phone and a desktop, and a free preview of whether the assistants actually recommend you. The $5 Pro audit asks every engine every question your buyers ask, tells you how to get recommended, and grades how ready you are for AI agents.
Public pages only. The check reads the page, robots.txt, sitemap, and llms.txt the way a crawler would. The full report is emailed to you and shown here; your address is used for that report and, at most, one follow-up from George. Protected by Cloudflare Turnstile and rate limits; results are cached for 15 minutes per URL. What it checks · Privacy.
-
Fix
Fetching the page, robots.txt, sitemap, llms.txt, and agent files in parallel… this can take up to ten seconds for slow sites.
Click any row for the evidence and the fix. Scores are per page, not per site.
: asking whether it recommends you. Watch it below; this page updates itself.
Sent The full report is on its way to . Check spam if it hasn’t arrived in a few minutes.
Not sent The email couldn’t be delivered right now; the full report is above. Copy the link to come back to it.
Lighthouse is loading the page twice, as a mid-range phone on slow 4G and as a desktop. This takes 30 to 60 seconds; the rest of the report doesn’t wait for it.
The 75th percentile of real visits (Chrome UX Report). Google ranks on these, not on the lab numbers.
Fix
No measurable savings left. This page is fast.
Measured by Google’s PageSpeed Insights (Lighthouse ). Scores vary a few points between runs; the metrics and the savings list are the signal.
asked of your buyers’ questions with web search on, the way a real user asks. “Recommended” means the answer named, linked, or cited . Category: .
Nobody else was linked in the answers where you were missing.
-
Example
Pro audit: all your buyers’ questions, every engine
The preview asks one engine a few questions. The Pro audit asks buyer questions across and shows every answer.
- buyer questions drafted for your category, editable
- Recommended-in-N-of-M for each engine, and who was picked instead
- Every answer shown, with its sources
- How to get recommended: 6–8 concrete actions drafted from your results, each with a copyable example (the page to publish, the listing to claim, the thread to answer)
- Agent readiness: 13 checks, a 0–5 level, and the step to the next level
- Edit the questions and competitors, re-run up to times in 30 days
- Full report emailed when it finishes
Secure payment by Stripe. Runs one to three minutes; results stay at this link for 90 days. No subscription.
Edit the questions and competitors, then re-run ( left)
All re-runs used. Book a call for ongoing tracking.
Answers change from run to run; the trend over several runs is the signal.
-
Fix
Fetching the well-known files, robots.txt, and DNS records…
Ten areas, weighted by how much they decide
The score is out of 100. Each area counts for its weight: a Pass earns all of it, Needs work half, Critical none. The checks come from the same audit I run at the start of every discoverability engagement, run against one page.
| Reachable · 15 | Does the URL answer 200 over HTTPS, quickly, without a noindex header? A page a crawler can’t fetch is invisible, however good it is. |
|---|---|
| Crawler access · 15 | robots.txt and the page’s robots meta tag, tested against sixteen named crawlers: Googlebot, Bingbot, OAI-SearchBot, ChatGPT-User, ClaudeBot, Claude-SearchBot, PerplexityBot, Applebot, and the training crawlers (GPTBot, Google-Extended, CCBot, and others). Blocking a search or answer crawler is Critical; blocking training crawlers is flagged as a choice. |
| Structured data · 15 | JSON-LD that says who is behind the site (Organization or Person) and what the page is (WebPage, Article, Service, FAQPage, BreadcrumbList). Invalid JSON is caught; microdata-only is Needs work. |
| Readable without JS · 12 | Is the text in the HTML, or does it arrive only after scripts run? Most AI crawlers and fetchers don’t execute JavaScript. An empty application shell is Critical. |
| Answer-ready content · 12 | One H1, section headings, question-shaped headings or an FAQ, and enough visible text to quote. The shape answer engines lift directly. |
| Page metadata · 10 | Title and meta description of a usable length, a self-referencing canonical, a lang attribute, viewport, and complete Open Graph tags. |
| Sitemap · 6 | An XML sitemap at /sitemap.xml or wherever robots.txt declares it, with lastmod dates. |
| llms.txt · 6 | A Markdown summary for AI agents at /llms.txt with the title, blockquote summary, and links the convention asks for. Optional, cheap, and still rare. |
| Entity consistency · 6 | A named entity with sameAs links to the profiles assistants already trust (LinkedIn, GitHub, Wikipedia, directories), contact details, a logo, and the same name in the title. |
| Agent readiness · 3 | Newer conventions: Markdown on Accept: text/markdown, llms-full.txt, and agent skills under /.well-known/agent-skills/. Optional extras, low weight. The Pro audit runs the full thirteen-check version. |
Load time, mobile vs desktop
Not part of the 100-point score, because it isn’t about being understood; it’s about being tolerated. A slow page gets crawled less, cited less, and abandoned by the people the assistant sent you. Google’s PageSpeed Insights loads the page twice through Lighthouse, as a mid-range phone on slow 4G and as a desktop, and adds what real Chrome users experienced over the last 28 days when the site has enough traffic.
| Performance score | Lighthouse’s 0–100 for each device. 90 and up is Pass, 50–89 Needs work, under 50 Critical. The mobile score decides the verdict, as it does for Google. |
|---|---|
| Core Web Vitals | Largest Contentful Paint (under 2.5 s), Cumulative Layout Shift (under 0.1), and Interaction to Next Paint (under 200 ms, real users only). These are the three Google ranks on. |
| Supporting metrics | First Contentful Paint, Speed Index, Total Blocking Time, and Time to First Byte, each rated Good, Needs work, or Poor. |
| Biggest savings | Lighthouse’s opportunity audits (render-blocking resources, unused JavaScript, image sizing, cache lifetimes, and so on) with the estimated time saved, biggest first. |
Agent readiness: thirteen checks, five levels
The same conventions Cloudflare’s agent-readiness scanner grades, minus the commerce and API-authentication protocols a content or service site doesn’t need. Each check is Pass or Missing with the evidence and the fix; the level tells you where you stand and the one step that moves you up.
| Discoverability | robots.txt (RFC 9309), an XML sitemap declared in it, HTTP Link headers on the homepage pointing at machine-readable resources (RFC 8288), and DNS-AID SVCB/HTTPS records under _agents.yourdomain. |
|---|---|
| Content for agents | Markdown content negotiation (Accept: text/markdown), /llms.txt, and /llms-full.txt. |
| Bot access control | Explicit User-agent groups for the AI crawlers, a Content-Signal line, and Web Bot Auth (a key directory at /.well-known/http-message-signatures-directory; informational, for sites that run their own agent). |
| Agent integration | An MCP server card (/.well-known/mcp/server-card.json), an A2A agent card (/.well-known/agent-card.json), and agent skills (/.well-known/agent-skills/index.json). |
| Levels | 0 Not ready → 1 Basic web presence (two of robots, sitemap, Link headers) → 2 Bot-aware (plus AI bot rules and Content Signals) → 3 Agent-readable (plus Markdown negotiation) → 4 Agent-integrated (plus one of MCP card, A2A card, agent skills) → 5 Agent-native (plus two of Web Bot Auth, all three integrations, DNS-AID). |
What this checker can and can’t tell you
It measures the half you control: whether machines can reach, parse, and trust your site. It doesn’t query the models, so it can’t tell you whether ChatGPT recommends you today. That’s why every report ends with the prompts to ask them yourself, and why a full engagement measures those answers before and after.
| Per page | One URL per run. Check your homepage, then your most important service page; they often score differently. |
|---|---|
| No JavaScript | Like most AI crawlers, the checker reads the raw HTML and runs no scripts. If your page scores badly here, that is the point. |
| Model queries | The ten-area check sends nothing to any AI provider. The preview and the Pro audit do: they send your buyers’ questions (never your email) to OpenAI, Anthropic, Perplexity, and Google and record what comes back. |
| Load time | Measured by Google’s PageSpeed Insights, which receives the URL you entered and nothing else. Lighthouse scores move a few points between runs; the metrics and the savings list are the signal. Sites that block Google’s Lighthouse user agent can’t be measured. |
| No rankings | A high score means nothing is stopping engines from finding and citing you. It is not a promise that they will. Nobody can promise that. |
| Cached | Results are cached for 15 minutes per URL. Fix something, wait, and re-run. |
| The report goes to the address you enter, and the same report appears on screen. The address is used for that email and, at most, one follow-up from George; it is never sold or added to a list. Privacy policy. | |
| Protected | Cloudflare Turnstile keeps bots out, and each address and connection gets a handful of checks per minute, so the tool stays fast for everyone. |
Questions, answered
- What is an AI visibility checker?
- A tool that tests whether AI assistants and search engines can find, read, and trust a website. This one fetches a page the way a crawler does and judges ten areas: reachability, crawler access in robots.txt, sitemap, llms.txt, structured data, page metadata, answer-ready content, rendering without JavaScript, entity consistency, and agent readiness. It returns a score out of 100 with a fix for each gap.
- Is it free, and do I need to give an email?
- It’s free. You enter the page to check and the email address the report should go to; the full report is emailed to you and shown on screen at the same time. The address is used for that report and at most one follow-up from George, never sold or added to a newsletter. Results are cached for 15 minutes per URL.
- Why does my site score badly when it looks fine in a browser?
- Usually because the content is rendered by JavaScript. Browsers run scripts; most AI crawlers and fetchers don’t, so they see an empty shell. The other common causes are robots.txt or CDN bot rules that block AI crawlers, and missing structured data.
- Should I block AI training crawlers like GPTBot?
- It’s a legitimate choice, and the checker flags it as one rather than an error. Blocking training crawlers keeps your content out of future model training; it also means those models know less about you. Blocking search and answer crawlers (OAI-SearchBot, ClaudeBot, PerplexityBot, Googlebot, Bingbot) is different: that removes you from the answers people see today.
- Does this tell me whether ChatGPT recommends my company?
- No. It tells you whether anything on your side is stopping that. Each report ends with three prompts about your company and one-click links to ask ChatGPT, Claude, Perplexity, and Google AI Mode, so you can see the other half in a minute.
- Does it measure page speed?
- Yes, free, after the ten-area report. Google’s PageSpeed Insights loads the page as a mid-range phone on slow 4G and as a desktop and returns a Lighthouse performance score for each, the Core Web Vitals (Largest Contentful Paint, Cumulative Layout Shift, Total Blocking Time) and supporting metrics, the biggest savings Lighthouse found, and, when the site has enough traffic, what real Chrome users experienced over the last 28 days. It takes 30 to 60 seconds and appears on the report page below the ten areas.
- What is the Pro audit, and what does it cost?
- The free check tells you whether machines can read you, plus load times and a short preview on one engine. The Pro audit answers the question buyers care about: does AI recommend you? It drafts the questions a buyer in your category would ask an assistant, asks ChatGPT, Claude, Perplexity, and Gemini every one of them with web search on, and reports how many answers named, linked, or cited you, who was picked instead, and every answer word for word. Then it tells you what to do about it: six to eight concrete actions drafted from your results, each with a copyable example (the page to publish and its opening line, the listing to claim, the thread to answer, the comparison to write). It also grades agent readiness: thirteen checks on the conventions AI agents use to find, read, and connect to a site, on a 0–5 level scale with the step to the next level. You can edit the questions and competitors and re-run it for 30 days. It is a one-time payment, shown on the page before you pay, and the full report is emailed when it finishes.
- What does “agent readiness” mean?
- Whether a site speaks the conventions AI agents (not just chat assistants) use to find, read, and connect to it: robots.txt with explicit AI bot rules and Content Signals, a sitemap, Link headers, DNS-AID records, Markdown content negotiation, llms.txt and llms-full.txt, Web Bot Auth, and an MCP server card, A2A agent card, or agent skills so an agent can act, not only read. The Pro audit grades those thirteen checks on the same 0–5 scale Cloudflare’s agent-readiness scanner uses, leaving out the commerce and API-authentication protocols a content site doesn’t need.
- How do you decide an answer “recommended” me?
- Deterministically: an answer counts when it names your brand, links your domain, or cites your site as a source. No model is asked to judge another model. The engines are asked with web search on and no memory of previous questions, so answers vary from run to run; the trend over several runs is the signal, which is why re-runs are included.
- How is this different from an SEO audit?
- It overlaps on the technical foundation (crawlability, canonicals, metadata, sitemaps) and adds the checks specific to AI assistants: named AI crawlers in robots.txt, llms.txt, rendering without JavaScript, entity consistency through sameAs links, and agent conventions. It doesn’t cover backlinks, keywords, or rankings.
- Can you fix what it finds?
- Yes. That’s the AI Discoverability service: I make the technical changes myself or alongside your team, write answer-first content, and measure what the assistants say before and after. Book a free 30-minute call and bring your report.
Found gaps? George Rios closes them.
This checker is the first step of his AI Discoverability service: the technical fixes done hands-on, answer-first content, and what the assistants say measured before and after. Book a free 30-minute call and bring your report.