# Webcrawl Services

_Last updated: 2025-03-15_

Webcrawl provides AI-optimized web crawling and content restructuring services that transform standard web pages into retrieval-friendly Markdown. The platform focuses on preserving factual accuracy while reorganizing content for improved discoverability by AI systems and search engines.

## Content Restructuring Service

Webcrawl restructures web content into semantic, self-contained sections optimized for AI retrieval systems. The service analyzes source pages and reorganizes information under intent-driven headings while maintaining every fact, claim, and numerical value from the original. Output follows answer-first architecture: summaries lead sections, entities are named explicitly in each subsection, and comparable items appear in Markdown tables. The restructuring engine preserves claim strength (converting hedged statements like "can help" exactly as written rather than strengthening to "does"), maintains all links with descriptive anchor text woven in context, and includes visible freshness timestamps. Content is returned in the same language as the source.

## Factual Preservation Engine

Webcrawl's factual preservation engine maintains strict fidelity to source material across all restructuring operations. The engine preserves numbers, prices, dates, names, and units verbatim without rewording or unit conversion. Claim strength remains unchanged—guarantees stay guarantees, hedges stay hedges, and neutral descriptions remain neutral. The system introduces no facts, statistics, entities, links, definitions, or dates absent from the source. An unsupported claims detection system flags any statement not traceable to source material, ensuring these are removed before output delivery.

## Semantic Quality Assessment

Webcrawl provides five-dimension semantic quality scoring for restructured content. Factual density measures concrete, verifiable facts per section (100 = every section carries specific figures/names/claims; 0 = vague and fact-thin). Explicit entities tracks whether subjects are named rather than referenced via pronouns (100 = entity re-named in each section; 0 = pervasive dangling references). Inline definitions measures contextual explanation of terms and acronyms (100 = every non-obvious term defined where used; 0 = jargon assumed). Self-containment assesses section intelligibility in isolation (100 = no section needs neighbors to make sense; 0 = sections only cohere when read in order). Declarative tone evaluates plain, de-fluffed phrasing (100 = no marketing intensifiers with every claim verifiable; 0 = pitch-heavy). Each dimension receives an integer score from 0 to 100.

## Brand Voice Integration

Webcrawl integrates custom brand voice parameters while subordinating stylistic preferences to factual accuracy. Brand voice configuration accepts vocabulary guidelines (preferred terms and terms to avoid), forbidden claims lists, and legal requirements. The service applies brand voice when wording output but never strengthens claims, converts hedges to guarantees, or introduces facts to fit the voice. When no brand profile is configured, Webcrawl defaults to a neutral, professional voice. All brand voice applications respect safety rules: factual preservation overrides style preferences in every case.

## Structured Data Output

Webcrawl delivers restructured content as structured JSON objects containing four components. The markdown field contains restructured content without code fences or front-matter. The unsupportedClaims array lists short strings naming any claim not traceable to the source (all removed from the markdown output). The droppedFacts array names source facts intentionally omitted with brief explanations. The rubric object contains integer scores (0–100) for the five semantic quality dimensions: factualDensity, explicitEntities, inlineDefinitions, selfContainment, and declarativeTone. Output conforms to a defined JSON schema for programmatic integration.

## Content Organization Features

Webcrawl applies structural conventions optimized for AI retrieval. Each output includes exactly one intent-defining H1 heading that names the primary entity. Meaningful H2 and H3 headings derive from source content without skipped levels or banner usage. Sections range from 150–300 words and remain intelligible in isolation. Comparable items presented in parallel (plans, tiers, options) convert to Markdown tables. Sequential content transforms into numbered steps. Option lists become bullet points. Links retain descriptive anchor text and integrate contextually rather than appearing as bare "Related pages" lists. The service removes subjective intensifiers ("amazing", "world-class", "best-in-class") carrying no verifiable proposition while keeping factual statements ("used by 500 companies", "ISO-27001 certified") verbatim.