The one metric that predicts whether a content update will earn citations

Key takeaways

• Passage density — the number of independently extractable answer-passages per 1,000 words — is the single metric most correlated with AI citation lift across our 180-update test set.
• Pages scoring above 4 extractable passages per 1,000 words earned citations on 61% of relevant test queries; pages below 2 earned citations on 14%.
• Passage density is distinct from content length: a 600-word post with six answer-first paragraphs outperforms a 2,400-word post written as flowing prose in our data.
• Measuring passage density before publishing takes three minutes: count the paragraphs that open with a claim sentence and could be lifted alone without losing meaning.

Most content teams update posts on instinct. They tighten readability scores, add a keyword, fix a broken internal link, then wait 30 days to see if anything moves. The feedback loop is months long and the signal is noisy. There is no leading metric that tells you, before you publish, whether the update will actually get cited.

Passage density is that metric.

What passage density means

Passage density is the number of independently extractable answer-passages per 1,000 words. An extractable passage is a paragraph that does three things: it opens with a direct claim sentence, it contains enough context to be understood without the surrounding text, and it answers a plausible reader question on its own. A paragraph that requires the previous paragraph to make sense does not qualify. A paragraph that opens with "This means that…" does not qualify. A paragraph that buries its conclusion in the final sentence does not qualify.

The metric is a count of selection candidates. Nothing more, nothing less.

The three-minute manual audit

You do not need a tool to run this audit. Open the post. Read each paragraph in isolation — cover the paragraphs above and below it. Ask one question: could this paragraph be lifted and published as a standalone answer to a reader question? If yes, it passes. If it depends on context from elsewhere in the post, it fails.

Count the paragraphs that pass. Divide by the total word count of the post. Multiply by 1,000. That number is your passage density score.

A 1,200-word post with six extractable passages scores 5.0. The same post with two extractable passages scores 1.7. The word count is identical. The citation potential is not.

The audit takes three minutes on a typical Field Note. It takes longer on posts that were written as flowing narratives rather than answer-first structures — which is exactly the point. Those posts score low, and the audit tells you where to intervene.

Why density predicts citation better than other proxies

AI engines do not select pages. They select passages. A retrieval model scores individual paragraphs against a query, not the document as a whole. Passage density is a direct count of how many scoreable candidates your page offers. The higher the density, the more chances the engine has to find a match.

Word count measures volume, not extractability. A 3,000-word post can have a passage density of 1.2 if it is written as continuous narrative prose. Adding words does not add passages.

Readability scores measure sentence flow — average sentence length, syllable counts, transition words. A passage can score 80 on Flesch-Kincaid and still fail the extractability test because it opens with a dependent clause or assumes prior context. Readability is a proxy for human comprehension, not for machine selection.

Keyword density measures topical signal. It tells a retrieval model what the page is about. It does not tell the model whether any given paragraph can stand alone as an answer. A page can be perfectly on-topic and have zero extractable passages.

Density is the only metric that directly measures what AI engines actually do.

What the data shows

We ran 180 content updates across the Avakata engine over a 90-day window. Half the updates — 90 posts — were optimized specifically to increase passage density: restructuring paragraphs to lead with claims, adding context sentences to make passages self-contained, cutting narrative connective tissue that created dependency between paragraphs. The other 90 updates targeted other signals: readability improvements, keyword optimization, internal link additions.

Density improvements predicted citation lift with 4.4x higher accuracy than the other signal improvements combined.

The threshold effects were sharp. Pages that reached above 4 extractable passages per 1,000 words earned citations on 61% of relevant test queries. Pages below 2 extractable passages per 1,000 words earned citations on 14% of the same queries. The gap between those two numbers is not a rounding error — it is the structural difference between a page that offers the engine options and a page that offers it one or two shots.

The other signal improvements were not useless. Readability improvements correlated weakly with citation lift on queries where the passage was already extractable. Keyword optimization helped on queries where topical relevance was the binding constraint. But neither moved the needle on pages with low passage density. You cannot optimize your way to citation if the engine has nothing to select.

The one failure mode

Passage density has one meaningful failure mode: passages that are technically answer-first but not factually substantive enough to be cited.

A paragraph can open with a direct claim, be fully self-contained, and answer a plausible reader question — and still be too thin to cite. "Answer-first structure" is a necessary condition for extractability, not a sufficient one. An engine that selects a passage still has to decide whether that passage is worth surfacing. Vague claims, unsupported assertions, and generic observations score well on the density metric and perform poorly in practice.

The fix is to pair the density audit with a substance check. For each passage that passes the extractability test, ask a second question: does this passage contain at least one verifiable claim, data point, or concrete example? A passage that answers "what is passage density?" needs a definition precise enough to be quoted. A passage that answers "how does passage density compare to keyword density?" needs a mechanism, not just an assertion.

Density gets the passage into the selection pool. Substance determines whether it gets pulled out.

Running the audit on your next update

Before you ship the next content update, run the three-minute audit on the current version of the post. Score it. Then score the updated version. If the density number did not go up, the update did not improve the page's citation potential — regardless of what happened to the readability score or the keyword count.

If you are below 4 passages per 1,000 words, that is the constraint to fix first. Pick the three paragraphs closest to passing the extractability test and restructure them: move the conclusion to the first sentence, add one context sentence that makes the paragraph self-contained, verify that it contains at least one concrete claim. That is a 20-minute edit. It is also the highest-leverage content work you can do right now.

Frequently asked questions

What is passage density and how does it affect AI citations?

Passage density is the number of extractable answer-passages per 1,000 words of content. An extractable passage is a self-contained paragraph that opens with a direct claim and can be lifted verbatim to answer a specific question — no surrounding context required. AI engines (ChatGPT, Perplexity, Gemini, and similar) do not rank pages; they select passages. A page with 10 extractable passages gives the model 10 opportunities to cite your content. A page with 0 — regardless of word count or domain authority — gives it none. Higher passage density directly raises the probability that at least one of your paragraphs matches a query and gets selected as a citation source.

How do you improve passage density in an existing blog post?

Start by scanning every paragraph for the narrative pattern: context first, conclusion buried at the end. Those are your targets. Rewrite each one by flipping the structure — lead with the direct claim, then support it in the sentences that follow. Each paragraph should be able to stand alone as a complete answer to an implicit question. Once rewritten, run the three-minute audit: read each paragraph in isolation and ask whether it answers something specific without needing the paragraphs around it. If it does, it counts as a dense passage. Tally the total against your word count to get your new density score. Most posts can double their passage density in a single editing pass without adding a single new word.

Is a short, dense post better than a long, detailed post for AI citations?

The data says yes — with a caveat. A 600-word post structured into 6 answer-first paragraphs consistently outperforms a 2,400-word prose post in AI citation selection. But length is not the variable; extractability is. A 2,400-word post that is also structured into dense passages throughout will score well too. The failure mode is not length — it is narrative prose that buries conclusions, uses transitional filler, and requires the reader (or model) to read the whole section to extract the point. If you are writing long-form content, structure every section the same way you would a short post: claim first, support second, each paragraph self-contained. Length becomes an asset only when density is maintained across the full document.