Neither. AI engines respond to claim density how many specific, extractable statements a page contains relative to its length not to word count.
A 700-word page with twelve checkable facts is more citable than a 3,000-word page with two.
This matters because a decade of SEO advice pushed toward comprehensiveness, and a great deal of B2B content is long specifically because someone was told long content ranks. That advice was always a proxy for something else, and the proxy has now diverged badly from the thing it was standing in for.
What is claim density?
Claim density is the proportion of sentences on a page that state something specific and checkable.
A claim is checkable if it contains a number, a boundary, a condition, a named entity, a price, a timeframe, or a comparison. Something a reader could in principle verify or dispute
Examples that count:
- “Implementation takes four to six weeks for teams under 50 users.”
- “The integration syncs invoices and payments but not purchase orders.”
- “Competitor X handles offline access better than we do.”
Examples that do not:
- “We pride ourselves on rapid deployment.”
- “The integration is seamless and powerful.”
- “We believe every business is unique.”
How to measure it? Count the checkable sentences. Divide by total sentences. Under one in ten and the page is mostly connective tissue.
Most B2B marketing content scores very low. Product documentation scores high, which is a substantial part of why documentation gets cited more often than blog content.
Why does long-form advice mislead?
Long-form advice was a proxy for thoroughness, and padding to a word count achieves length without thoroughness.
The original observation was reasonable: pages that comprehensively answered a query tended to outrank thin pages. Longer pages were correlated with more complete answers.
Then the correlation got optimised against. Writers were given word count targets. Content got padded with restated context, unnecessary preamble and expanded explanations of things nobody asked about.
The result: Long content that is not more thorough, only longer. The correlation between length and quality weakened considerably, and for extraction specifically it inverted padding dilutes claim density and buries the statements that matter.
What was actually being rewarded was answering the question completely. That has not changed. Length was never the mechanism.
Does depth still matter?
Depth matters. Length is not depth.
Worth separating two things that get conflated:
Depth m means covering the question thoroughly the edge cases, the conditions, the exceptions, the things a knowledgeable person would want to know. Depth adds claims.
Length Is word count. It can come from depth, or from padding.
A genuinely deep page is often long, because covering a complex subject properly takes space. That is fine. The page is long because it contains many claims, not in order to reach a target.
The test: If you cut a page by 30%, would you lose any checkable claims? If not, the 30% was padding.
What should you cut?
Six things that consume length without adding claims:
Preamble: Paragraphs establishing why the topic matters before addressing it. Readers who arrived at the page already believe it matters.
Restated context: “As we discussed above” and every variant. If a section needs the previous section restated, restructure rather than repeat.
Definitions nobody needs: Explaining what a CRM is on a page comparing CRMs. Your reader knows.
Hedging that removes content: “Results may vary depending on your specific circumstances” says nothing. If results vary, say what they vary with that is a claim.
Sections written to hit a structural target: The section that exists because the outline had five points and only four were real.
Adjective stacking: “Powerful, intuitive, flexible platform” contains no information. One specific capability beats three adjectives
What about comprehensive pillar content?
Pillar pages should be long because they cover many sub-questions, each answered densely not because they restate one point at length.
Pillar and cluster architecture remains sound. A pillar page covering a topic broadly, linked to supporting pages covering each sub-topic in depth, works for both search and citation.
What makes a pillar page work:
- Each section answers a distinct question completely
- Each answer is self-contained and extractable
- The page is long because there are many questions, not because each answer is padded
- It links down to supporting pages for depth rather than attempting all of it
What makes a pillar page fail: Treating “comprehensive” as a word count target. A 5,000-word pillar with twelve real claims is a 1,500-word page with 3,500 words of connective tissue.
How do you increase claim density?
Add specifics you already know but have not written down.
This is the practical move, and most companies find the facts already exist internally.
Five sources of claims you probably have:
Support and success teams: How long does onboarding actually take? What breaks most often? What do customers ask before buying?
Sales calls: What objections come up? What does the competitor actually do better? What do prospects say when they choose someone else?
Product usage data: What proportion of customers use each feature? What does typical usage look like?
Your own documentation: Facts already written down, sitting somewhere marketing content does not reference.
Client and customer patterns: What you have observed across many engagements, anonymised.
The process: Take a page with low claim density, list the sections, and for each ask “what specific fact do I know about this that is not written here?” A fifteen-minute conversation with support or sales usually produces five.
Then check the substitution test. Could a competitor publish this page with their logo on it? If yes, the claims you added were not specific enough.
Frequently Asked Questions:
How long should a blog post be for AI citation?
Length is not the variable. Aim for at least one checkable claim per section, and cut anything that does not carry one.
Does word count affect Google rankings?
Not directly. Longer pages historically correlated with more complete answers, but padding to a length target does not produce the completeness that was actually being rewarded.
Should I cut down my existing long content?
Where it is padded, yes trimming preamble and restated context raises claim density without losing anything. Where it is long because it covers many sub-questions, leave it.
What is a good claim density?
There is no formal benchmark. As a working heuristic, at least one checkable claim per section and above one in ten sentences overall.
Do short pages rank?
Yes, when they answer the query completely. A 500-word page fully answering a specific question outperforms a 2,500-word page that answers it partially among other things.
How do I find claims to add?
Ask support and sales. The specifics almost always exist internally and have simply never been written into marketing content.