What is AISO?

AISO is a free AI-search visibility scanner that grades any URL from 0 to 100 based on whether ChatGPT, Claude, Perplexity and Gemini can discover, understand and cite the page. It runs eight deterministic static-analysis checks plus a live prompt panel across four real LLM engines, then surfaces a prioritized punch-list of fixes you can apply yourself or hand off to AGNT Auto-fix.

AISO is built by the AGNT team and runs as a public service at aiso.tools. The same scan logic is available via REST API, a Model Context Protocol server, and an npm CLI, so you can integrate it into CI, Claude Code, or any terminal workflow.

Why AI search is different from Google search

Google has spent 25 years training webmasters to optimize for keywords, backlinks, and crawl budget. The new class of AI search — ChatGPT Search, Google AI Overviews, Claude's web search, Perplexity's citation engine — ranks differently. It prefers pages that read like definitions, cite authoritative sources inline, ship structured data, and expose themselves cleanly to crawlers with robots.txt allow-rules and llms.txt files.

Princeton's 2024 paper “GEO: Generative Engine Optimization” showed that adding inline statistics and external citations to a page can lift the probability of being cited by a generative search engine by up to 40%. That result is the single strongest optimization lever in the current literature, and it's the reason AISO weights citation density at 20 out of 100 points.

How AISO scores a URL

A scan runs in two phases. The first is deterministic: eight static-analysis modules each check one dimension of the page, producing a weighted score out of 100 and a set of issues. Full breakdown at /docs/dimensions, but the short version:

• Crawler Access (15 pts) — checks robots.txt for 14 AI bot user-agents and probes for Cloudflare WAF false-positives.
• Content Structure (20 pts) — definition-lead opener, 2–4 sentence above-fold summary, FAQ presence, heading hierarchy.
• Citation Density (20 pts) — external links to an authoritative allowlist (.edu, .gov, Wikipedia, W3C, MDN, Nature, arXiv, major publications) plus inline statistics. This is the Princeton GEO lever.
• Schema Markup (15 pts) — valid Organization, Article, FAQPage, WebSite, Person, BreadcrumbList JSON-LD.
• llms.txt (10 pts) — presence, valid H1, link resolvability, optional llms-full.txt.
• Freshness (5 pts) — visible Updated timestamp, <time datetime>, or Article dateModified.
• Entity Clarity (10 pts) — explicit self-identification, named author, about/team page, Organization founder.
• Off-site Authority (5 pts) — linked presence on Wikipedia, LinkedIn, YouTube, GitHub, G2, Reddit.

Each dimension produces a pass/warn/fail status and a list of specific issues with plain-English fix text. The final score is the weighted sum, and it maps to a tier — Invisible (0–39), Partial (40–64), Visible (65–81), Cited (82–100).

What “runtime visibility” means

Static analysis predicts how crawlable and citable a page should be. Runtime visibility measures what actually happens when a real LLM is asked a real question. AISO synthesizes five industry-specific prompts per domain (using Claude Haiku for consistency), runs each one against ChatGPT, Claude, Perplexity and Gemini with live web search enabled, and parses the returned citations.

The panel produces a separate 0–100 index. A prompt where your domain is the first citation earns 3 points; top-three earns 2; any citation earns 1; none earns 0. Brand-name mentions in the response text contribute a second signal. Both are normalized against the number of engines actually configured, so a two-engine run still produces a meaningful number.

Who AISO is for

Three audiences use AISO in production:

• Developers integrating via the MCP server or CLI to audit their site from Claude Code or a CI pipeline.
• Content teams shipping launches who want to know whether LLMs will cite them before the launch article goes live.
• SEO agencies adding AI-search audits to client reports. The free PDF export makes this one click.

Free and open

The scanner, CLI, and MCP server are MIT-licensed and live at github.com/aiso-tools. No email gate. No signup. Every scan has a public URL so you can share, link, or embed your score. The AGNT Auto-fix add-on (paid) ships PRs that patch the issues we find — that's how the project sustains itself. Everything else is free and stays free.