Skip to main content
All Services →AI Ops AuditAI Foundation BuildFractional Head of AIExecutive AI WorkshopClaude for TeamsAI VisibilityVoice DNA
IndustriesCase StudiesMethodologyResourcesAboutContactEspañolSend me one workflow
Free tool · Open source

citable: a free audit of what AI search can read on your site

citable is a free, open-source tool that audits a website for AI search. It runs 15 checks on what the crawlers behind ChatGPT, Claude, Perplexity and Google AI Overviews can read. Then it scores the site and writes a fix plan an owner can follow. It runs on your computer.

Built and maintained by Ignacio Lopez

MIT license · Source on GitHub

Why I built it

Most AI visibility problems I see start as plumbing, not content. A robots.txt file that blocks the crawler an assistant uses. A sitemap nobody updated. A homepage that tells a machine nothing about who the company is. Nobody sees these from the browser, and they decide whether an AI engine can read the site at all.

I kept running the same checks by hand at the start of every engagement. So I wrote them down as code and published it. Running them is mechanical, and nobody should have to pay for that part.

What citable checks

Fifteen checks in six groups. Any check that fails or raises a warning comes with the technical fix.

Crawlability (18% of the score)

  1. robots.txt is present and can be parsed.
  2. The main AI crawlers are allowed: GPTBot, ClaudeBot, PerplexityBot, OAI-SearchBot, Google-Extended and Applebot-Extended.
  3. sitemap.xml exists and lists URLs.
  4. The page responds in time when fetched as Googlebot.

Entity clarity (18%)

  1. Canonical URLs are present and point at the page itself.
  2. The Open Graph tags are set, including the image and the URL.

Content structure (18%)

  1. Meta descriptions run 80 to 160 characters.
  2. Titles run 30 to 65 characters.
  3. Each page has one H1, and it is unique.

Schema markup (18%)

  1. FAQPage schema sits on pages that answer questions.
  2. The homepage carries Organization schema, with a logo and sameAs links.

Authority signals (10%)

  1. The site sends an HSTS header.

Internal linkage (18%)

  1. No broken internal links. It probes up to 50 targets.
  2. Anchor text says where it goes. Generic anchors like "click here" get flagged.
  3. No orphan pages: every URL in the sitemap gets at least one link from a crawled page.

How to run it

In a terminal. You need Python 3.10 or newer.

pipx install git+https://github.com/WorkSmartAI-alt/citable.git
citable audit example.com

Add --pages 50 to crawl more pages, or --output to choose where the report goes. citable doctor checks the install.

Inside Claude. citable also ships as a Claude plugin. In Claude Code, run:

/plugin marketplace add WorkSmartAI-alt/citable
/plugin install citable@citable

Then ask Claude to audit a site. It runs the same engine and explains the result in plain language.

What you get

One Excel file with five sheets, each written for a different reader.

  • Action Plan, for owners and marketing. The fixes ranked by priority, with who should do each one and how much work it is.
  • Summary, for executives. The overall score, the score per group and the top five actions.
  • Action List, for developers. Every check with its verdict and the technical fix, sorted by severity.
  • Per-Page Detail, for whoever debugs it. Titles, render time, Open Graph tags and schema for every page crawled.
  • Score Breakdown, for auditors. The math, line by line.

How the score works

A pass counts 1, a warning 0.5 and a failure 0. Each group score is the average of its checks, and the overall score weighs the six groups as listed above. Checks that do not apply, or that citable could not verify, are left out of the math instead of counted as failures.

Grades: A is 85 to 100, B is 70 to 84, C is 55 to 69, D is 40 to 54, F is below 40.

Every number in the report traces to a line in the Score Breakdown sheet. If you disagree with a score, you can see exactly where it came from.

What citable does not do

It does not run JavaScript. It reads the HTML your server sends. If your site builds its pages in the browser, citable sees what a crawler that does not render sees, which is often very little. Several major AI crawlers do not run JavaScript either, so treat that result as a finding, not a bug.

It does not measure citations. A site can pass every check and still never get cited, because citations depend on what you publish and who links to it. To track whether AI assistants actually cite your domain for your buyers' questions, I publish a second tool, ai-visibility-monitor.

It does not fix anything. It is read-only.

It audits one host per run, skips pages behind a login, reads anchor text in English only, and stops probing links after 50.

Who built it

Ignacio Lopez, founder of Work-Smart.ai. I work as a fractional Head of AI for mid-market firms in Miami and Latin America, and AI search visibility is one of the services I run. citable is the first step of that work, published so anyone can take it.

The checks tell you what is broken. The work is deciding which fixes matter for the questions your buyers ask, and writing the pages AI engines end up quoting. That is what the AI Visibility service does.

Frequently asked questions

Yes. It is open source under the MIT license. You can use it on client sites, change it and ship it inside your own tools.

No. It runs on your computer and fetches your pages the way a crawler would. There is no account, no API key and no server of mine in the middle.

No. citable checks whether AI crawlers can read and understand your site. Whether they cite it is a separate measurement. ai-visibility-monitor runs your buyers' questions through ChatGPT, Claude and Perplexity and records who gets cited.

It audits them, but it does not render JavaScript. If your content only appears after scripts run, citable will show thin pages. That is useful to know, because some AI crawlers see the same thing. Serving pre-rendered HTML to bots fixes it.

Part of it overlaps: canonicals, titles, sitemaps, broken links. The rest is specific to AI search: which AI crawlers your robots.txt allows, the schema answer engines read, and a fix plan written for the owner, not only for the developer.

Yes, if by GEO you mean generative engine optimization, getting a site read and cited by AI answer engines. citable covers the technical foundation of that work. It does not write content.

A clean crawl is where AI visibility starts

Getting quoted takes pages written for the questions your buyers ask. That is what the AI Visibility service builds.

See the AI Visibility serviceGet citable on GitHub