# How SitePulsar Works — Audit Methodology

> A look inside the audit pipeline: what we crawl, which AI engines we query, how the Penta-Core engine turns raw signals into a score, and what shows up in your report.

## What SitePulsar audits
Every audit produces one **AEO (Agent Experience Optimization) score**, broken into three pillars:
- **FIND** — how agents discover you (engine visibility across ChatGPT, Claude, Gemini, Perplexity, Grok + on-site discoverability signals like llms.txt, sitemap, agent.json).
- **READ** — how agents comprehend your content (schema depth, structured data, trust signals, content quality).
- **USE** — how agents can act on your site (action readiness, APIs, MCP, agent task surfaces).

All three pillars roll into one 0–100 AEO score backed by 11 underlying sub-signals.

The audit is AI-generated and labelled as such (EU AI Act, Article 50). Scores are informational, not a guarantee of any future ranking or recommendation.

## The audit pipeline — 4 steps
1. **Crawl + parse** — fetch the target homepage (and configurable inner pages), extract JSON-LD, microdata, headings, sitemaps, robots.txt, llms.txt, ai-plugin.json, agent.json, and /.well-known/ AI manifests.
2. **Query 5 AI engines in parallel** — ChatGPT (OpenAI gpt-4o), Claude (Anthropic Sonnet), Gemini (Google 2.5 Flash), Grok (xAI), and Perplexity each receive category-specific reputation, comparison, and recommendation prompts. Responses are parsed for brand mentions, sentiment, recommendation context, and competitor set.
3. **Cross-reference Google APIs** — Knowledge Graph, Places, and Custom Search for entity validation, brand recognition, and external citations. Separates self-asserted facts from third-party signals.
4. **Aggregate + prioritise** — the aggregator combines deterministic USE-pillar signals (technical machine-readability) with sampled FIND-pillar signals (engine visibility) and READ-pillar signals (comprehension), produces pillar scores and a weighted overall, and generates a prioritised action plan with an estimated 0–100 impact score per recommendation.

## How scores are calculated
The AEO score uses a deterministic weighted model across three pillars (FIND, READ, USE), each backed by underlying sub-signals — 11 in total. Full pillar weights, per-check scoring rules, and priority-to-impact mapping are on /about. Two reruns of the same URL within the same window produce scores that match within AI-engine sampling noise.

## What you get in the report
- AEO score (0–100) with per-pillar breakdown (FIND / READ / USE)
- Per-engine reputation card (sentiment + authority per engine)
- Competitor share-of-voice map
- Google Knowledge Graph & Places status
- Prioritised action plan with estimated impact per fix
- Copy-ready implementation snippets (JSON-LD, robots.txt, llms.txt)
- PDF export (Pro+) and score history (Business)

## Audit duration
Most audits complete in 3 to 4 minutes. The crawl, 5 parallel AI-engine queries, and 20+ USE-pillar checks run concurrently; the aggregator pass adds 30–45 seconds.

## Citations
- OpenAI — 400M weekly ChatGPT users (early 2025): https://openai.com/index/how-people-are-using-chatgpt/
- Pew Research — AI summaries reduce click-through from ~15% to ~8% (2025): https://www.pewresearch.org/short-reads/2025/07/22/google-users-are-less-likely-to-click-on-links-when-an-ai-summary-appears-in-the-results/
- Aggarwal et al., "GEO: Generative Engine Optimization" (KDD 2024): https://arxiv.org/abs/2311.09735
- Noventa — Generative AI vs Web Search User Behaviour (JIEM, 2026): https://doi.org/10.61722/jiem.v4i2.9040

## Links
- How it works (HTML): https://sitepulsar.ai/how-it-works
- Methodology + scoring details: https://sitepulsar.ai/about#penta-core
- Run a free audit: https://sitepulsar.ai