Methodology

How OpenCrawl measures AI visibility

OpenCrawl turns crawl signals and answer evidence into a transparent five-pillar score. Every weight below matches the product scoring engine.

Design principles

  • Evidence over guesses. Scores come from fetch results and prompt outcomes, not vibes.
  • Neutral prompts. Buyer-intent queries avoid branded stuffing so mention rates stay meaningful.
  • Separate crawl from answers. Being crawlable is necessary but not sufficient for citations.
  • No outcome guarantees. Measurement helps you improve; it does not promise rankings.

Five pillars and weights

PillarWeightWhat it measures
Crawl access25%GPTBot / OAI-SearchBot allow signals and GPTBot homepage fetch success
Content readiness20%Readable HTML, title, meta description, Open Graph, JSON-LD, headings, llms.txt, sitemap
ChatGPT search30%Mention and recommendation rates across all live ChatGPT web-search prompts (20 generated + up to 10 optional manual)
Open-web coverage20%Mention and recommendation rates across an Exa sample (~10 searches) for open-web signal
Training presence5%Whether the domain appears in Common Crawl snapshots when checked

Prompt routing

A full scan always uses 20 GPT-generated buyer-intent prompts, optionally plus up to 10 manual prompts you save on the project. Every prompt is answered with ChatGPT web search. Exa runs a supplementary sample across intents to feed the open-web pillar and competitor discovery. Recommendation rate is weighted more heavily than raw mention rate inside each answer pillar. Methodology version: hybrid-v4.

Grades

Composite scores map to letter grades: A (90+), B (80–89), C (70–79), D (60–69), F+ (50–59), and F (below 50). Score deltas compare against the previous completed scan for the same project when available.

FAQ

Why weight ChatGPT search highest?

Live ChatGPT web-search answers are a direct signal of how often your brand appears in AI-mediated buyer journeys, so that pillar is weighted at 30% of the composite score.

Are prompts written to favor my brand?

No. OpenCrawl uses neutral buyer-intent prompts so results reflect organic mention and recommendation rates rather than branded query stuffing.