The 8 dimensions of AI-search visibility

Crawler Access (15 pts)

The first thing any AI search engine does is fetch your site. Crawler Access measures whether 14 known AI user-agents — GPTBot, ClaudeBot, PerplexityBot, Google-Extended, ChatGPT-User, Meta-ExternalAgent, Applebot-Extended, CCBot, Bytespider, DuckAssistBot, Amazonbot, DiffBot, YouBot, Timpibot — can access your site without being blocked by robots.txt, WAF rules, or Cloudflare bot-fight mode.

How it scores: round((allowed / 14) × 15). One auto-fail override: if a HEAD request with User-Agent: GPTBot returns 403, the whole dimension drops to 0.

Fix fast: publish a /robots.txt with explicit Allow: / rules for each AI bot, and make sure any WAF (Cloudflare, Fastly, Akamai) does not ship a bot-blocker preset that catches them.

Content Structure (20 pts)

Four sub-tests, 5 points each. (1) Definition-lead opener — the first 150 tokens match the pattern “X is a Y that Z”. (2) Answer-first above fold — a 2–4 sentence summary paragraph after the H1. (3) FAQ present — either a <details> block with ≥3 entries or FAQPage JSON-LD. (4) Heading hierarchy — exactly one H1, no H3 before the first H2, and every H2 has ≥3-word topic text.

Fix fast: rewrite the homepage H1 as a definition-lead sentence, add a 3-sentence summary directly underneath, and include a <details>-based FAQ block with at least five Q/A pairs.

Citation Density (20 pts)

This is the Princeton GEO lever. The 2024 paper “GEO: Generative Engine Optimization” found that adding inline statistics and external citations to an authoritative allowlist lifts the probability of being cited by a generative search engine by up to 40%. AISO weights this dimension at 20 pts for that reason.

Authoritative allowlist: .edu, .gov, .mil, .int, all Wikipedia, W3C, IETF, MDN, arXiv, Nature, IEEE, ACM, Reuters, AP, BBC, Bloomberg, WSJ, FT, Economist, and a curated set of standards bodies and tier-1 publications.

How it scores: citations density ≥ 3 per 500 words earns 10 pts, ≥ 1 earns 5. Stats density ≥ 2 per 500 words earns 10, ≥ 1 earns 5. Max 20.

Fix fast: for every statistical claim (“40% faster”, “$12M saved”, “2M users”) on your homepage, add an inline link to the origin study or data page. One arxiv.org or nature.com link is worth more than ten medium.com links.

Schema Markup (15 pts)

Scanner parses every <script type="application/ld+json"> block on the page and awards points per valid schema type:

• Organization with name, url, logo, sameAs ≥ 3 — 3 pts
• Article / BlogPosting / Product with author + datePublished + headline — 4 pts
• FAQPage with ≥ 3 Question/acceptedAnswer pairs — 3 pts
• BreadcrumbList — 2 pts
• Person — 1 pt
• WebSite with potentialAction: SearchAction — 2 pts

Fix fast: if your site has visible FAQs but no FAQPage JSON-LD, that is a high-severity flag with a one-shot fix — mirror the Q/A pairs into a JSON-LD block.

llms.txt (10 pts)

The llmstxt.org spec defines a markdown index at /llms.txt that AI crawlers use to navigate your documentation. Full guide at /docs/llms-txt-guide.

How it scores: valid /llms.txt with an H1 earns 6 pts; /llms-full.txt exists earns 2; all sampled linked URLs resolve to 200 earns 2. Max 10.

Fix fast: publish one. Takes ten minutes.

Freshness (5 pts)

AI engines explicitly weight recency. AISO looks for a last-updated signal in this order: <time datetime>, <meta property="article:modified_time">, dateModified in Article JSON-LD, visible “Updated: …” text, or the Last-Modified HTTP header.

How it scores: < 14 days since update = 5 pts; 14–30 days = 3; 30–90 = 1; older or no signal = 0.

Fix fast: add <time datetime="YYYY-MM-DD">Updated …</time> near the top of your homepage and bump it when you touch the copy.

Entity Clarity (10 pts)

Four sub-tests. (1) First paragraph contains an explicit self-identification (“we are”, “we're”, “this is”, “our mission”). (2) Named author or byline visible — either as a Person JSON-LD or a “By <Name>” pattern. (3) /about or /team page exists and mentions founders, staff, or credentials. (4) Organization JSON-LD includes a founder field.

Why it matters: AI engines heavily weight provenance. A site that does not tell you who runs it gets ranked below a site that does.

Off-site Authority (5 pts)

Lightweight proxy for off-site visibility. AISO scans your header and footer for outbound links to: Wikipedia, LinkedIn company page, YouTube channel, G2 profile, GitHub org, Reddit community. 1 point per platform detected, max 5.

Fix fast: link from your footer to your LinkedIn page and your GitHub org. That's 2 free points.