Most AI engines never see your site.
AI crawlers execute zero JavaScript, so any content missing from your raw HTML is invisible to them. geociter runs 37 evidence-tagged checks locally in about a minute, then writes the fixes for you.
$
Your results
Audit:
AEO/GEO score
–/100
Categories
Retrievability
What to fix first
Get the full report and the paste-ready fixes
All findings with evidence, plus generated llms.txt, robots.txt AI-bot policy, JSON-LD blocks and rewrite TODOs. We'll send one follow-up, no drip sequence.
Why this happens
The research behind AEO/GEO is public. Most sites just aren't built for what it says.
Keyword stuffing scored below baseline in the Princeton study. geociter doesn't optimise for keywords, it optimises for being quotable.
37 evidence-tagged checks
Every check is grouped into one of four categories. Each one cites the research behind it, not a made-up score.
| Category | What it verifies | Evidence |
|---|---|---|
| Crawler access | Whether GPTBot, PerplexityBot, ClaudeBot, and Google-Extended get an unblocked, JavaScript-free path to your content. | AI crawlers execute zero JavaScript, Vercel × MERJ, 500M+ fetches |
| Structured data | Whether there's parseable JSON-LD an answer engine can lift without inferring it from prose. | Machine-readable facts are cited more directly than paraphrased prose |
| Content shape | Whether claims sit in 120–180-word sections, backed by direct quotations, statistics, or a cited source. | Princeton GEO (KDD 2024): quotations +41%, stats and citations ~30% each |
| Retrievability | Whether a RAG system could answer your page's own likely questions using only text already on the page. | 44.2% of ChatGPT citations pull from a page's first 30% of text, Semrush |
The retrievability simulation
geociter treats your page as a small RAG corpus. For each question a person might ask an AI engine about what you do, it checks whether your page contains a single, self-contained passage that answers it, quotable, sourced, and short enough to lift whole.
The answer is buried after three paragraphs of scene-setting. No source is named. The section runs 400+ words, so a citation engine would have to summarise it, and summarising means it might not pick you.
The claim leads the section, a source is named inline, and the whole passage sits inside 120–180 words, the shape geociter's checks are built around.
Paste-ready fixes
Every failed check ships with the fix already written, not a to-do, a file.
- llms.txt
- robots.txt AI-bot policy
- JSON-LD blocks
- rewrite TODOs
Every check that fails carries the citation for why it's checked, and the exact block to paste to fix it.
{
"@context": "https://schema.org",
"@type": "Article",
"headline": "Replace with your page's H1",
"datePublished": "2026-01-01",
"author": {
"@type": "Organization",
"name": "Your organization"
},
"citation": "https://source-you-cited.example/study"
}
What geociter doesn't do (yet)
It audits your site and writes the fixes. It doesn't monitor whether ChatGPT, Perplexity, or Google's AI Overviews are actually citing you after you ship them, that's live citation tracking, and it isn't built yet.