bencium-aeo

Improve content for answer engines and AI-assisted search through evidence-rich direct answers, clear entity/topic structure, crawlable pages, valid.

SKILL.md

GitHub

Evidence-First AEO

Answer-engine optimization cannot guarantee citation or ranking. Optimize for user value, factual support, technical accessibility, and extractable structure, then measure across current surfaces.

Audit

  1. Define audience, query intents, entities, geography/language, conversion goal, and success metric.
  2. Verify crawl/index controls, canonical URL, status/rendering, headings, internal links, page speed, mobile/accessibility, and structured-data validity.
  3. Map each important claim to a primary or authoritative source, date, method/sample, units, limitations, and owner.
  4. Identify vague claims, hidden definitions, stale statistics, conflicting pages, duplicated answers, and unsupported superlatives.

Content design

  • Lead each intent section with a direct self-contained answer, followed by evidence, nuance, examples, and next action.
  • Use descriptive headings, lists/tables only when they clarify relationships, and stable entity names/definitions.
  • State dates, jurisdictions, units, assumptions, and uncertainty explicitly.
  • Cite original sources near claims; do not fabricate studies, benchmarks, customers, reviews, or schema fields.
  • Create FAQs only for genuine recurring questions. Do not stuff paraphrased keywords or manufacture Q&A solely for markup.

There is no universal "18-token," paragraph-length, or keyword-density rule. Test readability and extraction on the actual content/query set.

Structured data

Choose only schema types that match visible page content and current search-engine guidance. Keep JSON-LD consistent with the page, use valid URLs/dates/entities, and validate syntax. Markup does not make an unsupported claim true and does not guarantee a rich result.

Measurement

Track indexation, impressions/clicks, cited/linked answer appearances, referral quality, conversions, branded/non-branded query coverage, and factual/citation errors over a dated query set. Separate correlation from causation and record surface, location, account state, and personalization limits. Compare against a baseline and roll back changes that harm users or organic performance.

Deliverable

Return prioritized findings, rewritten sections/schema, evidence map, technical fixes, validation, experiment plan, and claims that remain unverified. Do not select or target a fixed AI model; evaluate current answer surfaces as products that can change.