Every GEO check findable runs, and how to fix each one

By findable team. Published 2026-10-14. Updated 2026-10-14.

AI tools now write answers and cite pages. GEO is the work of making your page worth citing. findable runs 27 GEO checks, and a short set of unscored Site checks for AI crawlers. This guide covers both. Remember that these are heuristics. They test signals, not quality.

What does the GEO family test in findable?

GEO is 25% of your findable score. It has 27 checks on citability: numbers, tables, quotes, outbound sources, clear definitions, author and organization markup, trust pages, fresh dates and content length. Each check is a pattern match, so a pass shows the signal exists, not that it is good.

How do I fix the evidence checks?

Five checks reward evidence: geo_statistics_density, geo_data_viz, geo_comparison, geo_expert_quotes and geo_citations. Add real numbers with units, a table, a named quote and links to sources. Never invent a figure to pass. A fake number hurts trust. The stat matcher wants a unit it knows, such as %, ms, MB, px, hours or a $ amount; "300 words" does not count.

<blockquote>
  <p>Quote text from a named person.</p>
  <footer>Name, role, <a href="https://example.org/source">source</a></footer>
</blockquote>
<p>Source: <a href="https://arxiv.org/abs/2311.09735">GEO: Generative Engine Optimization</a>.</p>

The GEO paper by Aggarwal et al. tested ideas like adding citations, quotations and statistics. In their tests, these helped visibility. Results will vary by engine and topic. geo_expert_quotes is satisfied by any <blockquote> with text in it; nothing checks that a named person said it.

How do I fix the passage and definition checks?

Three checks cover passages: geo_definitions, geo_passage_citability and geo_content_length. State what a thing is using "is a" or "is the". Write at least one paragraph of 40 to 120 words that stands alone. Reach 300 words in total. Only <p> text counts toward the passage check, so an answer written as a list or inside a <div> is never seen.

A passage is citable when you can lift it out and it still makes sense. Name the subject in the first sentence. Avoid "this" and "it" with no clear referent. The geo_definitions test only looks for the phrases " is a ", " is the " or " defined as ", so it is easy to pass and easy to fool.

How do I fix the author and organization checks?

Six checks cover who is behind the page: geo_author_schema, geo_author_credentials, geo_org_schema, geo_knowledge_graph, geo_trust and geo_entity_markup. Show a byline, add Person and Organization JSON-LD with sameAs links, link to your about and contact pages, and use at least 2 schema types.

<p>By Alex Morgan</p>
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@graph": [
    {"@type": "Organization", "name": "Example", "url": "https://example.com",
     "sameAs": ["https://github.com/example"]},
    {"@type": "Person", "name": "Alex Morgan"}
  ]
}
</script>

The geo_trust check looks for the strings "about" and "contact" in your internal link URLs, so /about-us and /contact-sales both count. The schema author only counts when author is an object with a name; a bare string is ignored. See schema.org for types, and Google's structured data intro.

How do I fix the freshness, intent and format checks?

Five checks cover shape and dates: geo_freshness, geo_time_element, geo_intent_clarity, geo_format_variety and geo_snippet_control. Show a date in a <time> element, share words between title and first heading, mix 3 formats, and avoid nosnippet in the robots meta. The five formats counted are lists, tables, blockquotes, code and definition lists.

<time datetime="2026-10-14">October 14, 2026</time>

geo_freshness passes on a <time> element, a year from 2024 to 2029 in the text, or a dateModified in schema. A year in the text is a weak signal, and a 2023 date does not count. geo_intent_clarity counts shared words longer than 3 letters between the title and the first heading. geo_snippet_control fails if nosnippet limits what can be shown.

What are the Site checks and why do AI crawlers care?

The Site checks run only with the -site flag and are unscored. They fetch /robots.txt and /llms.txt and test whether named AI bots may read your page. They matter because a blocked crawler cannot read even a perfect page. -site needs an http or https URL; it refuses a local HTML file.

findable -site https://example.com

How do I let AI bots in, or keep them out?

Edit /robots.txt. A bot you name with Disallow: / is blocked. A bot you do not mention follows the * rules. The choice is yours. A bot that cannot fetch the page cannot cite it. Blocking one keeps your text out of that vendor's crawl. findable only reports the current state.

User-agent: GPTBot
Allow: /

User-agent: ClaudeBot
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Google-Extended
Allow: /

Official docs: OpenAI crawlers, Anthropic ClaudeBot, Perplexity bots, Google common crawlers and the robots.txt intro. Note that robots.txt is a request, not a lock.

For llms-txt, add a plain Markdown file at /llms.txt that lists your key pages. See the llms.txt proposal. It is a proposal, and no one promises that every AI tool reads it.

Key takeaways

• GEO is 25% of the findable score and has 27 checks.
• Real sources, quotes, tables and numbers with units make a page easier to cite.
• Never invent a statistic to pass geo_statistics_density.
• Show a byline, add Person and Organization JSON-LD, and link about and contact pages.
• Site checks are unscored. Run them with -site.
• Check robots-gptbot, robots-claudebot, robots-perplexitybot and robots-google-extended, then decide who you want in.
• Add /llms.txt if you want to, knowing it is only a proposal.