How points work
The number on this page is points earned divided by points of measured checks. Unmeasured items (for example a phone match when JSON-LD has no telephone) are N/A and drop out of the denominator. Training crawlers are listed and never scored. There is no “average site,” no traffic-light for AI mentions, and no claim that a score predicts citations.
| Check | Points | What is counted |
|---|---|---|
| AI crawler access | 24 | OAI-SearchBot, ChatGPT-User, and PerplexityBot 6 each if this URL’s path is allowed or not mentioned. Googlebot 4; Bingbot 2. GPTBot, ClaudeBot, and Google-Extended are shown, 0 points. |
| llms.txt | 8 | File found as HTTP 200 text 4; H1 title 1; summary 1; at least one Markdown link 1; optional llms-full.txt 1 (N/A if missing). |
| Structured data | 28 | Organization or LocalBusiness 8; name 4; url 3; logo 3; sameAs 3; telephone 3; address 2; WebSite 2. Parse errors are flagged. |
| Entity and NAP | 20 | JSON-LD name vs title or og:site_name 6; name in visible text 4; phone match 4 when a number exists; address match 4 when an address exists; sameAs present 2. |
| Crawlable content | 20 | HTTP 200 5; not noindex 5; title 3; meta description 2; H1 2; visible letters in the first HTML 3. |
What each check means and how to fix it
AI crawler access. We parse robots.txt into User-agent groups, then ask whether this URL’s path is allowed, blocked, or not mentioned for each bot. A named group beats User-agent: *. Inside a group, the longest matching path wins; equal length prefers Allow, following Google’s robots.txt documentation. Blocking GPTBot or Google-Extended does not lose points — OpenAI and Google treat those as training / Gemini-product controls, separate from search. If OAI-SearchBot, ChatGPT-User, or PerplexityBot is blocked for this path, allow that token (or stop Disallowing the path) so answer products that honour robots.txt can fetch it.
llms.txt. We actually GET /llms.txt on the origin, following public redirects. A 200 text file with a Markdown H1, a short summary, and links matches the llms.txt proposal. HTML served as 200 is treated as absent. Add the file at the site root if you want agents to have a curated map; skip it if you do not. Absence is a small deduction, not a ranking event.
Structured data. Every application/ld+json block is parsed. Arrays and @graph are walked. Organization or LocalBusiness (or a subtype) should carry name, url, logo, sameAs, telephone or contactPoint, and address when you publish those facts. A WebSite node should sit alongside. Broken JSON is listed so you can repair the script, not guess at types.
Entity and NAP. We only compare strings that are on this page. The JSON-LD name should appear in the title or og:site_name, and again in visible HTML. If JSON-LD has a phone or address, that same number or street should show in the body. sameAs domains are listed; we do not visit those profiles.
Crawlable content. The same public HTML the website grader already fetches: the page should return 200, omit noindex, include a title, a meta description, an H1, and enough letters in the first response that a crawler which does not run JavaScript still has copy to read.
AI crawler user-agent reference
Tokens, owners, and purposes below are copied from each vendor’s own crawler documentation, last checked 18 November 2026. We re-read those pages monthly. A robots.txt rule is a request: firewalls and CDNs can still block a crawler the file allows. This table is not a ranking forecast, and it is not a list of bots this checker scores.
| User-agent | Owner | Purpose | robots.txt | Docs | Last checked |
|---|---|---|---|---|---|
GPTBot | OpenAI | Training | Honours robots.txt | Overview of OpenAI Crawlers | |
OAI-SearchBot | OpenAI | Search | Honours robots.txt | Overview of OpenAI Crawlers | |
ChatGPT-User | OpenAI | User fetch | May ignore (user-initiated) | Overview of OpenAI Crawlers | |
ClaudeBot | Anthropic | Training | Honours robots.txt | Anthropic crawler help article | |
Claude-SearchBot | Anthropic | Search | Honours robots.txt | Anthropic crawler help article | |
PerplexityBot | Perplexity | Search | Honours robots.txt | Perplexity Crawlers | |
Google-Extended | Training and grounding | Control token (does not crawl) | Google common crawlers | ||
Applebot-Extended | Apple | Training | Control token (does not crawl) | About Applebot | |
Bingbot | Microsoft Bing | Search | Honours robots.txt | Which crawlers does Bing use? | |
CCBot | Common Crawl | Open web archive | Honours robots.txt | Common Crawl FAQ |
Limits
One public URL, not a site crawl. JavaScript is not executed. Login walls, challenges, and private addresses are refused. ChatGPT-User is scored from robots.txt even though OpenAI says user-initiated fetches may ignore those rules. We do not call any AI API and we do not store the address you enter. For a wider pass that includes speed and HTTPS, use the website grader. To talk about a build, use the contact form.
Fixing these is part of building a site made to be found by Google and AI search. Nuvora builds custom websites for $500/month, first month paid upfront, with full ownership after 24 months. AI search work beyond the build is an add-on we scope and quote before it starts.