Generative Engine Optimization checker
ChatGPT, Perplexity, Claude and Google AI Overviews choose sources they can reach, parse and quote. Check any page in seconds and get the fixes.
How it works
Answer engines pick sources they can reach, read without JavaScript and quote with confidence. The audit checks each of those in turn.
GPTBot, ClaudeBot and PerplexityBot do not run JavaScript. We fetch the raw HTML and extract only the text, headings and structure they receive.
robots.txt is evaluated for 20 AI user-agents, grouped into search indexes, user-triggered fetchers and training crawlers. Only the first two affect your score.
We request the page again as OAI-SearchBot, ChatGPT-User, Claude-User, PerplexityBot and GPTBot to catch firewall and CDN blocks that robots.txt never shows.
35+ signals: answer-first sections, statistics, sources and quotations, structured data, llms.txt, freshness and authorship. Each comes with the evidence and a fix.
“Through rigorous evaluation, we demonstrate that GEO can boost visibility by up to 40% in generative engine responses.”
Scoring
Access and readability come first: if a crawler can't reach or read the page, nothing else counts. Checks that don't apply are marked N/A, not zero, and hard blockers cap the score.
AI crawler access | 25% | AI search & assistant bots in robots.txt · Live requests with bot user-agents · noindex / nosnippet · HTTPS, status, redirects |
Content & answerability | 25% | Text present without JavaScript · Heading hierarchy & question headings · Answer-first paragraphs · Lists, tables, passage-sized sections |
Citation-worthiness | 15% | Statistics density · External sources cited · Quotations · Keyword stuffing |
Discovery files | 15% | robots.txt & sitemap.xml · llms.txt (llmstxt.org spec) · llms-full.txt / Markdown versions · Descriptive internal links |
Structured data | 12% | JSON-LD validity · Organization & page types · sameAs brand identity · Schema matches visible content |
Metadata & trust | 8% | Title & description · Author and dates (freshness) · About / contact pages · Hidden prompt-injection |
FAQ
Last updated
GEO is the practice of making content easy for AI answer engines — ChatGPT, Perplexity, Claude, Gemini and Google AI Overviews — to find, understand and cite. Where SEO aims for a ranked link, GEO aims for your page to be quoted or referenced inside the generated answer.
Source: Aggarwal et al., GEO (KDD 2024)
The foundations overlap: crawlability, indexability and clear content still matter. GEO adds signals specific to language models: content readable without JavaScript, passages that answer a question on their own, statistics, sources and quotations to cite, structured data that identifies your brand as an entity, and access for AI-specific crawlers such as OAI-SearchBot and PerplexityBot.
We run 35+ checks in six weighted categories: AI crawler access (25%), content and answerability (25%), citation-worthiness (15%), discovery files (15%), structured data (12%) and metadata and trust (8%). Checks that do not apply are marked N/A instead of zero. Hard blockers — noindex, AI search bots blocked in robots.txt, or content that only exists after JavaScript runs — cap the score at 30–40.
Distinguish training from search. GPTBot, ClaudeBot, Google-Extended and CCBot collect training data; blocking them is a legitimate choice and does not affect your score. OAI-SearchBot, ChatGPT-User, Claude-SearchBot, Claude-User, PerplexityBot and Perplexity-User fetch pages for AI search and answers; blocking them removes you from those answers.
Source: OpenAI crawler documentation
llms.txt is an emerging convention (llmstxt.org) that gives language models a curated Markdown map of your site. Adoption is growing, but no major AI vendor has confirmed relying on it and Google says AI Overviews need no special files, so it carries a small weight here. It is cheap to add — we generate a starter for you.
Source: The llms.txt specification
No. Each check runs live against the public URL you enter and the result is returned straight to your browser. There are no accounts and nothing is saved; a shared link simply runs the check again.