Methodology

Five layers. One score per layer. Never one composite.

HoldThat measures AI visibility the way real engineers measure systems: discrete layers, real evidence, re-runnable results. No vibes. No vanity score.

Layer 01

Technical

Crawl health, render integrity, structured data coverage.

Sample metric

Schema coverage / indexed pages

What we test
  • Crawl access for GPTBot, PerplexityBot, Google-Extended, ClaudeBot, CCBot
  • Server-rendered HTML vs JS-only rendering
  • Structured data coverage (Organization, Article, Product, FAQPage, BreadcrumbList)
  • Canonical hygiene across paginated and faceted URLs
Evidence captured
  • Per-bot fetch logs
  • Rendered vs raw DOM diff
  • Schema validator output per page type
Layer 02

Semantic

Topic clustering, intent mapping, answer-ready content.

Sample metric

Intent-mapped topics / total topics

What we test
  • Topic cluster coverage vs buyer intent map
  • Answer-readiness of priority pages (question framing, definition blocks, comparison tables)
  • Internal linking density from cluster pillars to supporting articles
Evidence captured
  • Intent-to-page mapping spreadsheet
  • Per-page answer-readiness score with example excerpts
Layer 03

Entity & Trust

Author bios, entity graph, third-party citations.

Sample metric

Verified entity nodes mapped

What we test
  • Founder, author, and operator entity nodes
  • sameAs links to LinkedIn, Crunchbase, GitHub, professional registries
  • Third-party citations and unstructured mentions across review surfaces
Evidence captured
  • Entity graph diagram
  • Citation source list with date stamps
Layer 04

AI Presence

Citation rate and position across 5 generative engines.

Sample metric

Cited prompts / total prompts × engine

What we test
  • Per-engine prompt test across 5 generative engines
  • Citation rate, position, and competitor displacement
  • Brand mention vs domain citation vs no mention
Evidence captured
  • Prompt × engine matrix
  • Screenshot per prompt run with timestamp
  • Raw engine response stored for re-verification
Layer 05

Business Impact

Assisted conversion attribution from AI-referred sessions.

Sample metric

AI-assisted conversions / period

What we test
  • Sessions referred from AI engines
  • Assisted conversions tied to AI-referred touches
  • Pipeline contribution from AI-attributed sources
Evidence captured
  • GA4 / Plausible export of AI-tagged sources
  • Assisted-conversion model output
Why re-runnable matters

A score you can't re-run is a slideshow.

Most AI audit reports are PDFs frozen in time. HoldThat stores every prompt, every engine response, and every screenshot as a versioned record. Next quarter you re-run the same prompts and compare to a fixed baseline — no methodology drift, no moving goalposts.