The 40-Point AEO Content Geometry Standard

The 40-Point AEO Content Geometry Standard is the per-article scoring rubric Data for AI Search uses to grade editorial content against the structural patterns AI assistants — ChatGPT, Perplexity, Claude, Gemini, Grok — preferentially extract when synthesizing citations. The standard scores ten dimensions across 100 points total, with 80/100 set as the publishing gate. It's the article-level analog to Check 5 of the 10-Point AI Citation Framework and the per-article rubric for the recursive-aeo publishing skill. Every article published on this site scores against this standard before publishing. Articles below 80 trigger a refactor loop until passing. The standard exists because article-level citation prediction is empirically distinct from domain-level citation prediction — a 3,800-word pillar with poor citation geometry scores worse on AI citation than a 1,500-word supporting article with strong geometry, regardless of domain authority. This guide is the canonical reference: every dimension scored, every sub-criterion, worked examples from real audits, and the practical refactor patterns for moving an article from failing to passing.

What does the standard measure?

The 40-Point AEO Content Geometry Standard measures one article against the structural patterns AI assistants extract preferentially during synthesis. The standard is intentionally article-scope. Site-scope signals — NAP consistency, Knowledge Graph entity, directory presence, crawler accessibility — belong in the 10-Point AI Citation Framework. The 40-Point Standard tests one piece of editorial content for structural extractability + citability.

The ten dimensions, in order of point allocation:

Total: 40 sub-criteria across 10 dimensions, summing to 100 points maximum. The 80/100 gate is the publishing threshold in recursive-aeo Phase 4.

Why are these the right dimensions?

Each dimension reflects an observed regularity in AI citation behavior across 20+ client audits, calibrated against the SERanking November 2025 study of 300,000 domains and against observational research on per-platform extraction patterns.

The point allocation reflects evidence weight. Dimensions with the strongest correlation to AI citation get the highest point values:

Extractable opening passage (5 pts). AI assistants preferentially extract opening passages because the structure mirrors how the assistant produces its own response. The 134-167 word range emerged empirically from audit observation — below 100 words the passage lacks context for synthesis; above 200 words AI assistants summarize rather than quote.

Question-formatted H2s (5 pts). Per internal audit data, question-formatted H2s outperform statement-formatted H2s by roughly 3× in citation rate. The structural pattern matches buyer queries directly and mirrors the assistant's own generative process.

Named entity density (5 pts). Factual specificity discriminates citable content from rhetorical content. AI assistants prefer sources that demonstrate concrete reference patterns over sources that argue abstractly.

Sourced statistics with dates (5 pts) and inline source URLs (5 pts). Together these account for 10 of the 100 points because verifiable claims are the strongest content-level credibility signal. AI assistants increasingly weight claim verifiability during retrieval ranking.

Definitive first sentences per H2 section (4 pts). Sections opening with a definitional answer in sentences 1-2 get extracted at materially higher rates than sections opening with throat-clearing ("The question is empirical," "This is a complex topic").

Comparison tables (3 pts), FAQPage schema (3 pts), declared author entity (3 pts). Each contributes meaningfully to specific citation surfaces (buying-decision queries, FAQ-shaped queries, credibility filtering) without dominating overall structure.

Cross-links to sibling cluster articles (2 pts). The smallest single-dimension weight, but cluster cross-linking density contributes to topical authority signal (Check 9 in the 10-Point Framework). Articles linked into a coherent cluster outperform isolated articles.

How do you score an article against the standard?

The scoring procedure for a drafted article:

Step 1: Count opening passage words. Use a word counter on the first paragraph before any H2. Award points based on the 134-167 range bracket. Document whether the opening contains a sourced statistic.

Step 2: Inspect H2 structure. List every H2 in the article. Count those phrased as questions ("What is X?", "How does Y work?", "Why does Z matter?"). Compute the question-formatted percentage. Award points based on the percentage bracket.

Step 3: Count named entities. Extract every named entity in the article — people, brands, places, dated statistics, specific products. Document the count. For pillars, target ≥15; for supporting articles, target ≥10.

Step 4: Audit numerical claims. Identify every numerical claim in the article. For each, check whether a date is present. Then check whether an inline hyperlinked source URL is present. Compute percentages for each criterion.

Step 5: Audit section openers. For each H2 section, read the first 1-2 sentences. Award points based on the percentage that lead with a definitional answer versus throat-clearing.

Step 6: Check for comparison tables. Identify whether the article topic is comparison-worthy (X vs Y, choosing between options). If yes, check for at least one comparison table. If no, the dimension defaults to full points.

Step 7: Verify schema declarations. Check whether the article has FAQPage JSON-LD schema (FAQ section + registered faqs array). Check BlogPosting schema (article registered + ArticleSchema component renders). Check declared Person author entity with sameAs array.

Step 8: Count cross-links. Tally internal links to other articles in the same cluster + to foundation pillars. Target: 3+ for the dimension to earn full points.

Step 9: Sum the points. Total across all 10 dimensions. The result is the article's score out of 100.

Step 10: Apply the publishing gate. Articles scoring ≥80 publish. Articles below 80 enter the refactor loop until they cross the gate.

The complete procedure takes 5-10 minutes per article once you've done a few. Most experienced authors can mentally score draft articles during writing and produce passing-quality content on first draft.

What are the most common gap patterns?

Cross-article patterns we surface most frequently during Phase 4 audits:

Inline source links missing on dated statistics. Statistics cited by named source ("per the SERanking November 2025 study") but without a clickable hyperlink to the actual study. The most common gap. Costs 3-4 points; usually a 15-minute fix that involves adding [study name](https://source-url) markdown around existing references.

FAQPage schema not registered in LEARN_ARTICLES. Visible FAQ section in the MDX but no faqs array in the article registry entry. Schema generation requires the registry entry. Costs 1-2 points; trivial fix — copy the visible Q&A into the registry's faqs array.

Section openers softened to throat-clearing. "The question is empirical" or "This is a multifaceted topic" instead of a definitional first sentence. Costs 1 point per occurrence; fix by tightening the H2 section's first sentence to lead with the answer.

Insufficient cross-links to sibling cluster articles. Authors forget to link back to the foundation pillar or to sibling supporting articles in the same cluster. Costs 1-2 points; usually fixed during the publish-and-reinforce step by adding cross-links to recently-shipped sibling articles.

Weak named entity density. Articles below 15 entities for pillars or below 10 for supporting articles. Often happens when authors default to generic rhetorical phrasing. Fix by replacing generic phrases ("a study showed") with specific entity references ("the SERanking November 2025 study of 300,000 domains showed").

Worked example: scoring "What is AEO?"

The pillar article /learn/what-is-aeo on dataforaisearch.com scored as follows on first audit (commit 034d22f):

Total: 39/40 = 98/100. Above the 80 gate by 18 points. Published after the inline-source-links fix.

The one point lost was on inline source URLs — initial draft had named the SERanking study and Position Digital research by name but without hyperlinks. The Phase 5 refactor added 9 inline citations. Subsequent re-audit moved the article from 39/40 to 40/40 = 100 on that dimension, though we record the published score as 98/100 reflecting the original publish-readiness.

How does this standard relate to broader AEO/GEO scoring?

The 40-Point Standard sits inside the broader 10-Point AI Citation Framework. The relationship:

• 10-Point Framework = site-level scoring (Check 1 crawler, Check 2 brand mention frequency, Check 3 directories, ..., Check 10 per-platform). Used by /ai-audit <domain> for site-wide assessment.
• 40-Point Standard = article-level scoring (this document). Used by recursive-aeo Phase 4 for per-article publishing gate.

Both methodologies share the underlying content-geometry signals. The 40-Point Standard is the expanded version of Check 5 in the 10-Point Framework, with point allocations refined for article-scope evaluation.

A site can score 75/100 on the 10-Point Framework (Grade C) while having every published article score 85+/100 on the 40-Point Standard. The two metrics measure related but distinct things — domain authority signal vs editorial content quality.

Frequently asked questions

Why 80/100 as the gate?

Eighty reflects the threshold at which structural quality compounds across the cluster. Articles scoring 80-89 contribute positively to cluster topical authority signal. Articles scoring below 80 dilute the cluster's average and reduce overall AI citation lift. The gate is empirical, not arbitrary — calibrated against observation that clusters with universally high-scoring articles outperform clusters mixing high and low scorers.

Can articles publish below 80 in exceptional circumstances?

In our recursive-aeo skill, no — the gate is hard-block. For brands using the standard internally without enforcement infrastructure, the recommendation is: don't publish below 80 even when tempted. The recursion that makes methodology-loop publishing defensible depends on the standard being applied uniformly.

Does AI-generated content score differently against the standard?

The standard is content-agnostic to authorship. AI-assisted, AI-generated, and human-written content all score against the same dimensions. The patterns the standard tests (extractable passages, sourced statistics, question H2s, declared authorship) are achievable in any authoring mode. The relevant question is whether the content meets the structural quality — not how it was produced.

How often does the standard get updated?

Updates ship with explicit version bumps. The current version is v1.0 (locked 2026-06-23). Future versions will document changes inline with rationale, the same way the 10-Point Framework changelog works. The standard is versioned because retroactive audits should be runnable against the standard version in effect when an article published.

Does the standard apply to short-form content?

Partially. The standard is calibrated for editorial content of 1,500+ words. Short-form content (newsletters, social posts, micro-content) doesn't fit the structure — the extractable passage rule alone fails for sub-300-word content. For short-form, use a subset of the standard: focus on question-format headlines, sourced statistics with dates+URLs, and entity density.

---

Companion guides: How to write content that ChatGPT will cite · Question-format H2s deep-dive · Named entity density · The 134-167 word extractable passage rule · Source links and date stamps · The 10-Point AI Citation Framework.