How the GEO Audit Works

AI Visibility — visibility in AI search

The first question AI systems ask when forming a response is: should this site be cited at all? Is there a better source for the same fact — one they trust more? The AI Visibility agent tries to answer that question before ChatGPT does.

The highest-weighted signal is Citability (35%). This is not about whether the text is well-written. It is about whether specific passages can be extracted from context and inserted into an AI response without losing meaning. Short, self-contained paragraphs with clear statements are cited far more often than long discursive prose.

Brand Mentions (30%) measure the presence of the brand on external platforms that AI treats as primary sources: Wikipedia, Reddit, YouTube, LinkedIn. If a company exists nowhere beyond its own website, AI treats it as an unknown entity.

Crawler Access (25%) checks whether robots.txt allows GPTBot, ClaudeBot, PerplexityBot, and others. A blocked crawler cannot see the site. No visibility, no citation. This sounds obvious, but one in three sites we audit blocks at least one AI agent through inherited SEO rules.

Content Quality — content quality

E-E-A-T stands for Experience, Expertise, Authoritativeness, and Trustworthiness — four criteria Google introduced in its Quality Rater Guidelines. AI systems inherited this logic. Content that does not demonstrate real first-hand experience competes with thousands of similar pages and rarely makes it into responses.

Each of the four criteria is scored from 0 to 25 points. Experience means concrete detail: case studies, numbers, inside descriptions of a process. Expertise is reflected in how the material is structured and how deeply it covers a topic. Authoritativeness looks at external signals — mentions and links. Trustworthiness checks for transparency: author information, sources, absence of contradictions.

The agent also audits technical text parameters: volume, Flesch readability score, and heading hierarchy. Not because volume matters in itself — but because thin content pages consistently receive lower trust scores across AI platforms.

Platform Analysis — per-platform scoring

Five AI platforms operate on different logic. Factors that boost a site in Perplexity answers may be irrelevant for Bing Copilot. The agent evaluates each platform independently and shows exactly where a site is losing to competitors.

Google AI Overviews looks at source authority through a Search Console lens: how many pages are indexed, whether rich snippets exist, how well the content answers user questions. ChatGPT Web Search focuses on entities — named objects that can be connected to real-world subjects. If a company has a Wikipedia article or a Google Knowledge Graph entry, its chances of appearing in a response increase significantly.

Perplexity builds a large portion of its answers around Reddit and forums — real user opinions that Perplexity prefers over corporate pages. Google Gemini gravitates toward its own ecosystem: Google Business Profile, Knowledge Graph, YouTube. Bing Copilot uses Microsoft-ecosystem signals: LinkedIn, IndexNow for fast indexing, Bing Webmaster Tools.

Schema — structured data

Structured data is a direct way of telling AI: "here is who we are, here is what we do, here is where you can verify it." The agent checks for the presence, validity, and priority of Schema.org markup.

The primary GEO signal is Organization markup with a populated sameAs field — a list of links to all official company profiles: Wikipedia, Wikidata, LinkedIn, Google Business. This is how AI systems connect a website to a real-world entity. Without it, you are just an anonymous domain.

speakable is a lesser-known but important Schema property. It tells AI assistants which specific page sections are meant to be read aloud. Google Assistant and Siri use it for voice responses. Most sites do not implement it.

The agent also flags deprecated schemas. HowTo markup was removed from Google's supported features in 2023. FAQPage now appears only for authoritative sources on medical and government topics. These schemas do not break a site, but they produce no effect either.

Technical — technical accessibility

The technical portion of a GEO audit overlaps with SEO but shifts the emphasis. The core question here is: can an AI crawler actually read the page? Not whether a user can find it — whether a machine can read it.

Server-Side Rendering accounts for 25% of the technical score — not arbitrarily. GPTBot, ClaudeBot, and PerplexityBot do not execute JavaScript during crawling. Sites built on React, Vue, or Angular without SSR or SSG serve crawlers an empty HTML shell. Content that appears only after JS loads does not exist from an AI's perspective.

robots.txt and sitemap.xml are checked not just for existence but for correctness: whether key sections are accidentally blocked, whether the sitemap is up to date, whether canonical tags are properly set on duplicate pages.

Security headers — HTTPS, HSTS, Content Security Policy — affect how trustworthy a source appears. A site without HTTPS or with outdated security headers can receive a "low trust" flag in certain AI systems.