About Blockquote

What does Blockquote measure?

Blockquote measures whether a page is structured for citation by an AI answer engine. A scan scores one URL from 0 to 100 across three weighted categories: schema at 40 percent, structure at 30 percent, and citability at 30 percent.

Schema covers the machine-readable layer — JSON-LD blocks, canonical URLs, Open Graph tags, sitemap and llms.txt presence. The vocabulary behind most of it is schema.org, the shared type system that search engines and answer engines both read. Structure covers the shape of the page — a definition in the opening paragraph, question-phrased headings, subheading frequency, lists and tables. Citability covers the prose itself — word count, external citations, emphasis, an author byline.

A check that does not apply to a page's type is marked N/A and left out of the score rather than counted against it. A refund policy carries no Article schema and is not penalised for it; a product page carries no FAQ block and is not penalised for that. The scanner decides applicability from a deterministic page-type classifier, which sorts a page into one of five labels — content, legal, transactional, navigational, or other — from its URL, its structured data, and the shape of its prose.

Who runs Blockquote?

Arne Kellmann runs Blockquote as a sole operator from Ober-Ramstadt, Germany. One person writes the checks, the scanner, the pages and the copy; there is no team, no company and no register entry behind the service. The whole system runs on Cloudflare's edge network, and the language-model checks run on Cloudflare Workers AI rather than on a third-party API.

The imprint carries the full statutory disclosure — name, postal address, VAT identification number, and the person responsible for content. This page states what the service does; that one states who answers for it.

Why does Blockquote exist?

AI answer engines cite a handful of sources per answer. ChatGPT, Perplexity and Google AI Overviews read a page, synthesize an answer, and name the few sources that answer fed. A site owner has no ranked list to inspect and no click-through report that names the winner, so the question of why one page gets quoted and the next does not stays invisible.

Blockquote exists to make that question answerable: it reads the page the way a retrieval system reads it, states which structural signals are missing, and hands back the markup or the rewrite that supplies them. The audit is deliberately narrow. It does not measure traffic, rankings, backlinks or conversion — it measures one property of one page, and reports the evidence it counted.

That narrowness is also what makes the result checkable. A score built on a competitor's estimated traffic cannot be reproduced by the person reading it. A score built on counted properties of the page itself can: every verdict names the number it counted and the threshold it compared against, so a reader who disagrees can open the page source and settle it.

How does the 34-check method work?

A scan fetches the URL once, parses it without a browser, classifies the page type, and then runs 34 checks against the result. 29 of them are deterministic: they read markup, structure and metadata, and the same page always produces the same verdict. The remaining 5 run a language model against a fixed rubric at temperature 0, and the report labels them as model-judged.

Every check returns a pass, a fail, or N/A, plus the evidence behind the verdict — the counted number, not an adjective. "3 external links in the main content" is a verdict a reader can check; "weak citations" is not. Each category then contributes its weight to the final score:

Agents run the same 34 checks through a remote server that speaks the Model Context Protocol, so a scan started by a person and a scan started by an agent return the same report.

The model-judged checks run at temperature 0 against a written rubric, not against a free-form prompt, because a citability score that moves between two runs of the same page is not a measurement. The report marks which 5 verdicts came from the model, so a reader who trusts the deterministic half and doubts the other half can tell them apart instead of having to trust the number as a whole.

What is AI citability, and how does it differ from classic SEO?

AI citability is the property of being the source an answer engine quotes inside a synthesized answer. Classic SEO optimizes for position in a ranked list that a human scans and clicks. Citability optimizes for a different outcome entirely: there is no list, no ten blue links and no click. A model reads a set of retrieved pages and writes one answer, naming the sources it leaned on.

The two overlap — both reward clear headings and fast, crawlable pages — but they diverge where it matters. A page can rank third for a query and still never be quoted, because its answer is spread over six paragraphs instead of stated in one, or because a pronoun makes the key sentence meaningless once it is lifted out of context. That failure mode has a name in the research literature, which studies it as generative engine optimization, and several of the citability checks are built on it.

Who is Blockquote for?

Blockquote is for content teams, technical writers, and developers who publish pages meant to be read and cited by both humans and AI agents. A writer uses it to find the sentence that will not survive being quoted out of context. A developer uses it to find the JSON-LD block a page never shipped.

Agents can run the same audit themselves rather than waiting for a person to paste a URL into a form. The AI agent setup page documents that path: which tools the remote server exposes, how an API key raises the quota, and what the report looks like when an agent reads it instead of a browser.

Blockquote is not for everyone. A page whose problem is that nobody links to it, or that it answers a question nobody asks, will not be fixed by better markup, and the audit says nothing useful about either. It reads one page at a time and reports what that page's structure does and does not supply.

What happens after a scan?

A scan returns a score, a per-category breakdown, and a verdict for every check with the evidence behind it. Failed checks carry a fix: a JSON-LD block to paste into the head, or a stated content change with the threshold it has to clear.

The fixes are ranked, and the report enriches the top recommendations with page-specific wording rather than generic advice. The LLM-friendly content playbook writes out the full checklist behind the score, so a page can be built to pass before it is ever scanned. Blockquote replays its 29 deterministic checks against every page in its sitemap, and scans its homepage weekly with all 34. This page is written to pass them.