01
Hypothesis
Entity Clarity
LLMs plausibly prefer brands with an unambiguous entity presence across Wikipedia, Wikidata, and knowledge graphs — a stub-level or ambiguous entity is a reasonable candidate for being harder to retrieve confidently, though this specific causal link hasn’t been isolated in either AIVI™ benchmark. Treat entity clarity as a signal worth testing, not a guaranteed lever. See entity reconciliation and Google, Knowledge Graph Search API for how a recognized entity is defined.
02
Platform documentation
Structured Data Density
Schema.org markup — FAQPage, Article, and DefinedTermSet types among them — makes a page’s entities and relationships more explicit to machines. Google’s own documentation states structured data isn’t required for a page to appear in its generative AI features, and that no special schema.org markup is needed (see Google Search Central, Structured data markup). Treat schema as an accessibility layer, not a ranking mechanism — consistent with FAQ Q12.
03
Brainpan.AI benchmark observation
Corroboration Across Sources
A claim repeated only on owned channels is a claim; the same claim corroborated across independent, high-authority sources behaves differently in retrieval. Brainpan.AI’s AIVI™ studies track Citation Authority and Citation Independence separately from a brand’s own content performance for exactly this reason. Build distributed, independent placements — not just owned-channel content. See FAQ Q14 on why LLMs weight independent corroboration over self-description.
04
Hypothesis
Declarative Sentence Structure
LLMs plausibly extract answers more easily from declarative, subject-verb-object sentences than from hedged or heavily qualified marketing prose — a reasonable inference from how extractive summarization works, though it hasn’t been isolated as its own variable in either AIVI™ benchmark. Write the load-bearing sentence plainly; save nuance for the paragraph around it. See answer accuracy for how Brainpan.AI scores extracted claims.
05
Brainpan.AI benchmark observation
Domain Authority & Trust Signals
High-authority domains aren’t weighted because of a generic “Domain Authority” score — that’s a third-party SEO metric, not a factor in any LLM’s published retrieval architecture. What the AIVI™ benchmarks show instead is that independent, editorially-earned corroboration — being cited by sources a brand doesn’t control, rather than simply holding a high link-authority score — correlates with stronger citation behavior. See FAQ Q14: LLMs weight independent corroboration more heavily than self-description.
06
Hypothesis
Named Author Authority
Content attributed to a verifiable, named expert — a real byline, credentials, a consistent public record — is a plausible trust signal for readers and retrieval systems alike, though Brainpan.AI hasn’t isolated byline presence as its own measured variable across either benchmark. Named-author authority is a reasonable bet, not a verified lever on its own. See Kevin Walsh’s author page for the pattern this site follows.
07
Hypothesis
Topical Depth & Completeness
LLMs plausibly favor sources that comprehensively answer a topic cluster over thin, fragmented pages — consistent with how retrieval-augmented systems rank passages by relevance and coverage, though this remains an inference from retrieval mechanics rather than a figure measured in either AIVI™ study. Own the full topic, not a single keyword. See RAG for how retrieval-based engines select passages.
08
Hypothesis
Recency Signals
For RAG-enabled engines — Perplexity, Copilot, Gemini with Search — freshness plausibly affects retrieval eligibility as these systems browse live before answering, though Brainpan.AI hasn’t isolated a specific recency weighting in either benchmark. Publish on a consistent, dated cadence rather than letting cornerstone content go stale. See FAQ Q11 for the fuller picture of what makes content retrievable.
09
Platform documentation
robots.txt & llms.txt Basics
Incorrect crawler blocking in robots.txt can exclude a page from AI retrieval and training pipelines entirely — audit this before anything else on this list. llms.txt is a separate, voluntary convention for publishing a concise, LLM-readable site overview; it is a discovery aid, not a crawler-control or training-authorization mechanism — those are governed independently by each provider’s own crawlers (OpenAI, for example, separates OAI-SearchBot from GPTBot; see OpenAI, Crawlers). See FAQ Q13 for the full retrieval-vs-training distinction.
10
Hypothesis
Semantic Proximity to Query Intent
Content that mirrors how buyers actually phrase prompts — rather than how marketers phrase headlines — is plausibly retrieved with higher fidelity, consistent with how embedding-based retrieval matches semantic similarity, though Brainpan.AI hasn’t isolated prompt-pattern mirroring as its own measured variable. Prompt-pattern research is an ongoing GEO discipline, not a one-time audit. See AI search measurement.
