What the AI Citation Readiness Grader does
The AI Citation Readiness Grader scores an article on the things that make a passage safe to quote: specific claims, sources placed next to those claims, visible dates, a named author, stable facts about the publisher, paragraphs that make sense on their own, and links to outside evidence. Every point it awards or withholds comes with the sentence, date or link it found.
It is a published rubric, not a model of any AI system. Nobody outside Google, OpenAI, Anthropic or Perplexity can see how their products choose sources, so this tool does not guess. It checks what a careful editor would check before expecting anyone - a journalist, a researcher or a machine - to cite the page.
How to use it
- Paste the article's full HTML (so the byline, dates and JSON-LD in the head are included), or Markdown or plain text, or switch to Fetch a URL.
- Optionally give the page address when pasting. It is not fetched; it only tells the grader which links are your own and which point to outside sources.
- Select Grade content. Each of the eight criteria shows its points, its basis (documented guidance or A2Z heuristic) and the evidence behind the score.
- Work through the fix checklist, largest gap first, and grade the new version.
- Copy or download the checklist as Markdown, or the criteria as CSV, to hand to the writer or editor.
Reading the results
The score is out of 15: seven criteria worth 2 points and one (no anonymous "studies show") worth 1. 80% or more reads as strong on this rubric, 55-79% as mixed, and below that as weak. These are labels for the rubric, not predictions.
A specific claim is a sentence of at least six words that contains a number, date, price or quantity. Bylines and "Updated..." lines are excluded. A claim counts as sourced when the same paragraph has a link or an attribution such as "according to", "source:", a bracketed reference number or a year in parentheses.
Two criteria are marked Documented: sourcing and named authorship both appear among the self-assessment questions in Google's helpful-content guidance. That guidance describes what people-first content looks like; it does not say these are ranking factors, and neither does this tool. The other criteria are A2Z heuristics.
Worked example: a robots.txt explainer with two gaps
The article example is a short HTML page about robots.txt with a byline, a <time> element and Article JSON-LD carrying datePublished, dateModified, the author and the publisher. The grader finds nine sentences and three specific claims: the September 2022 publication of RFC 9309 (linked), the 500-kibibyte parsing limit (attributed "according to RFC 9309") and Google's 24-hour cache (no source). Two of three sourced is 67%, which earns full marks for sourcing.
Dates, authorship and publisher facts each score 2 of 2. "Studies show that most sites get their robots.txt wrong" costs the attribution point. Two of seven paragraphs are self-contained - the closing "This is why testing matters..." depends on what came before - and only two outside domains are linked. The total is 12 of 15, strong on the rubric, with the checklist asking for a named source or no claim, one more outside source, and paragraphs that stand alone.
Why sources belong next to the claim
A passage is often read, or quoted, without the rest of the page. If the evidence for a figure sits in a references list at the bottom, the excerpt carries the figure and loses the evidence. Putting the link or the attribution in the same paragraph keeps them together wherever the paragraph goes.
The same logic drives the self-contained check. A paragraph that starts "This means..." or "However..." only makes sense after the one before it; quoted alone, it can be misread. Naming the subject again costs a few words and removes the ambiguity.
Limitations: what the result does not prove
- It checks that a source is present, not that it supports the claim. A link to a home page next to a statistic scores the same as a link to the table the figure came from.
- Claim detection is a pattern (digits, currency, percentages), so qualitative claims without numbers - "the most popular framework" - are not counted as claims and not checked for sources.
- Bylines are recognised in English ("By", "Written by", "Reviewed by", "Author:"). Other formats may be missed unless they are in meta tags or JSON-LD.
- A high score does not mean any AI product will cite the page; a low score does not mean it will not. It measures the page against this rubric and nothing else.
Privacy: where your data goes
The analysis runs in your browser. If you choose to load a public URL instead of pasting, a2z.tools fetches that address read-only and returns the text to your tab; private and internal addresses are refused and nothing is stored. Session recording and tag-manager scripts are switched off on this page.
Standards and sources
- Google Search Central - Creating helpful, reliable, people-first content - checked 19 Sep 2026
- Google Search Central - AI features and your website
- Google Search Central - Article structured data (author and dates)
Frequently asked questions
What makes content more likely to be quoted by AI assistants?
No operator publishes a formula, so no one can answer that with certainty. What is documented is general: Google's helpful-content guidance asks whether content has clear sourcing, evidence of expertise and information about the author. This grader checks those visible qualities and a few editorial ones, and says which is which.
Does an author byline really matter?
Google's helpful-content guidance lists "Is it self-evident to your visitors who authored your content?" among its self-assessment questions and suggests bylines where readers would expect them. It is advice about trustworthy content, not a documented ranking signal, and the grader labels it that way.
Why is my references section at the bottom not enough?
The grader looks for a source in the same paragraph as each specific claim, because a quoted paragraph travels without the rest of the page. A numbered reference such as [3] in the sentence does count, since it ties the claim to the list.
Which dates does the grader look for?
Visible dates in the text (ISO dates and forms such as 14 August 2026), <time datetime> elements, article:published_time and article:modified_time meta tags, and datePublished or dateModified in JSON-LD. Full marks need a visible date and a machine-readable one.
Why does "studies show" lose a point?
An unnamed study cannot be checked. When a paragraph says studies show, experts agree or research suggests without a link, the grader flags it and asks you to name and link the source, or remove the claim.
Does a perfect 15 out of 15 mean the page will be quoted?
No. It means the page meets every criterion in this rubric. Whether a particular search engine or assistant uses the page depends on things no outside tool can observe, including the question asked and every other page available.
Last reviewed by the A2Z.Tools team against the sources listed above.
