AI Citation Preview

How it works

1

Paste a URL

Any public page — a blog post, a docs page, a landing page. A bare domain works too; we default to https.

2

We fetch it like a crawler

We request the HTML the way a search or AI crawler does — no JavaScript — and read the title, meta tags, Open Graph, JSON-LD, headings, dates and author.

3

See how it renders, and what to fix

A Google result, a social card and an AI answer citation side by side, plus a scored checklist of what is missing or malformed.

What AI answer engines look at

Nobody publishes the algorithm. What we do have are citation studies — which pages ChatGPT, Perplexity and Claude cite, and what those pages have in common — and the crawler behaviour we see in Kitbase's own bot analytics. These are correlations, not rules, but they are consistent enough to act on. The tool checks each of them.

1. 1

   A title and first paragraph that answer the question

   Answer engines lift the opening of a page far more often than a paragraph buried halfway down. If your first paragraph states the answer plainly, that is the sentence they quote — and it is what this tool shows in the citation card.

2. 2

   A visible, machine-readable date and an author

   Recency and accountability are two of the strongest correlates of being cited. Put datePublished / dateModified in JSON-LD or article:published_time / article:modified_time meta tags, and name an author. A page with no date reads as stale.

3. 3

   Structured data that names what the page is

   Article or BlogPosting for editorial content, Organization for who publishes it, BreadcrumbList for where it sits, FAQPage when the page genuinely holds question-and-answer pairs. Valid JSON-LD is what turns a wall of HTML into something an engine can classify.

4. 4

   Canonical and indexable

   A self-referencing canonical and no noindex. Most AI engines lean on a search index somewhere in the pipeline; if the page cannot be indexed, it is very unlikely to be cited.

5. 5

   Fast, server-rendered HTML

   The content, meta tags and schema should be in the HTML the server sends. Tags injected by JavaScript are invisible to almost every crawler — including this tool, on purpose.

For the engine-by-engine detail behind these, read how Perplexity chooses its citations and how Claude picks its web sources.

How the score works

Every page starts at 100. Each distinct issue we find deducts a fixed amount by severity, and the score never goes below zero. It is a checklist score — a measure of how complete and well-formed the page's markup is — not a prediction of where the page will rank or how often it will be cited.

Fix the errors first: they are the things that stop a page being indexed or parsed at all. Warnings are the tags an engine expects to find and will hold against you when they are missing. Infos are polish.