Full Audit Pipeline

How Paceghost Works

Paceghost treats every website as an execution environment — not just a document index. Our methodology combines real browser execution, AI-powered analysis, and triple-pillar benchmark measurement.

How repeatable is the audit?

The Paceghost method is a documented, step-by-step procedure: fixed rules in fixed code, with every input and every result recorded. Where the table says “always the same”, that is a property of the procedure — checked line by line against the code that runs it.

Step	What happens	Do you get the same result every time?
Account checks	Fixed rules check your account before anything runs: your plan, your Credits, and whether an audit is already running for your domain.	Yes — the same account records always lead to the same decision.
Crawl permissions	We read your site’s robots.txt, llms.txt and ai.txt files, plus your sitemap, using fixed parsing rules.	The same files always produce the same verdict. The files themselves can change — that’s your call as the site owner, and we record exactly what we found.
Page snapshot	A real browser loads your page and captures the moment: text, structure, metadata and a screenshot.	The capture rules are always the same. What gets captured is your page exactly as it renders at that moment — cookie banners, slow-loading content or a fresh redesign included. That’s why every assessment refers to the snapshot, and we say so.
Website analysis	Fixed E-E-A-T rules score the snapshot in five pillars (0–10 scale). An AI model writes the plain-language analysis and tidies the example quotes — it can never change the score. E-E-A-T is our own assessment framework, not a Google ranking factor.	Rule scores: deterministic — the same snapshot, the same numbers. Written analysis: a documented model with fixed instructions; we record which model produced each analysis, and its wording can vary slightly.
Search benchmark	We ask search engines (SEO) and AI answer engines (GEO, AEO) a fixed set of questions — the same set for your domain at every audit — and record every answer word for word. Fixed matching rules then check the recorded answers for your domain and brand. For AEO, one extra step sorts each recorded answer into a fixed set of categories that decides how it counts toward your visibility. That sorting is done by a machine-learning model built for classification — not an LLM — and we use a fixed version of it.	Scoring a recorded answer is always the same. The live answers themselves change over time — that change is exactly what we measure. Which AI models answered is recorded per scan; the exact model mix can change between scans.
Report	Your report is assembled entirely from the recorded results of steps 1–4.	The same recorded results always build the same report — apart from the generation date. Deterministic from the records.

If an answer is missing (for example, because Google shows no AI Overview), we report it as “not measured” — it is never counted as a zero.

The 5 Pipeline Steps

Step 1 — Crawlability Check

Paceghost fetches your site’s robots.txt, llms.txt, and ai.txt files and checks which AI crawlers (GPTBot, ClaudeBot, PaceghostBot, Googlebot, PerplexityBot, and others) are allowed or blocked. The result is a crawlability status: fully open, partially blocked, or fully blocked.

If robots.txt blocks PaceghostBot, the audit halts here — we respect your site’s directives. llms.txt and ai.txt are displayed as informational signals.

Step 2 — Website Snapshot

Paceghost fetches your page through a ladder of fetch approaches, probing before it commits:

• Disclosed bot — a headless browser that identifies itself openly with the Paceghost/1.0 user agent. This rung is available to every audit.
• Genuine browser — a real, headed Chromium with no user-agent override, so it appears to your site exactly as a regular visitor’s browser would. This rung is available when you have verified ownership of the domain.

The navigation itself is the probe: if the page loads, the capture continues on the same page; if the site’s security layer refuses the request, the next allowed rung is tried. The rung that succeeded is remembered per domain, so later audits start with the approach that worked — and a rung that was refused is re-probed after a cool-down, because sites heal.

When every allowed rung is refused by the site (HTTP 401, 403, or 429), that block is reported as a finding — the snapshot is never taken from an error page, because auditing a block page would misrepresent your site. The PaceghostBot page explains how to let our bot through a CDN or firewall, and how to verify that you own the domain. Failures for any other reason — for example our fetch infrastructure fails, or your site returns an error page so there is nothing to audit — end the step with a full refund of its credits. Which fetch client served a given audit is disclosed per report.

The snapshot itself captures a screenshot, the full DOM structure, page metadata (title, description, headings, links, schema markup), and raw HTML. This is the factual record of what the page looks like at the time of the scan.

The snapshot is the input to Step 3. It also lets you verify that the crawl is seeing the correct version of your page — not a cookie-walled or JavaScript-broken version.

Step 3 — Website Analysis (Synthetic Domain Profile)

The snapshot content is sent to an AI model to produce the Synthetic Domain Profile (summary, branded/unbranded keywords, user-intent queries). Separately, Paceghost computes a deterministic E-E-A-T score, built from five on-page measurement pillars — directly from the page’s structure (schema markup, outbound citations, original media, content density, and first-person language), not from a model’s opinion. See the E-E-A-T Score page for how each pillar is measured.

The output is a Synthetic Domain Profile — a structured result derived from your real DOM content, not from model training data. It includes:

• AI-generated one-sentence summary of your site
• Branded keywords — your brand, product names, etc.
• Unbranded keywords — topics relevant to your domain
• User intent queries — what users might search to find you
• The E-E-A-T score and its five pillars (Entity & Author, Outbound Citations, Experience Media, Content Density, Linguistic POV), computed by fixed rules, not by the model

Important: The Synthetic Domain Profile is NOT what an AI model “knows” about you from its training data. It is what the model extracts from your actual page content at scan time.

Step 4 — Visibility Benchmark

The keywords from your Synthetic Domain Profile are run through three search channels simultaneously:

• SEO (Search Engine Optimisation) — Traditional search results from several search engines.
• GEO (Generative Engine Optimisation) — AI-augmented search results and AI overviews from several providers.
• AEO (Answer Engine Optimisation) — Direct LLM queries testing what several AI models say about your domain when asked directly.

For each query and each engine/model, we record:

• Whether your domain appears in results (domain found / domain cited)
• What position or rank
• What snippet, summary, or citation is shown

Both linked and unlinked citations are detected via triple-pass brand matching. Results are measured at a single point in time per audit.

The number of endpoints and models included scales with your tier — see the Endpoints chapter.

Step 5 — Create Report

All data from Steps 1–4 is compiled into a downloadable report package:

• PDF — Executive summary with scores, charts, and key findings
• JSON — Full data bundle for programmatic use
• CSV — Zipped export for spreadsheet and BI tool import
• Share link — A link that anyone who has it can open for 7 days. It gives access to the report PDF.

Report files are retained for 30 days. Per domain, we keep your most recent reports during that time: the Free plan keeps the latest report, paid plans keep the latest 4. The underlying audit data (crawlability results, snapshot, analysis, benchmark responses) is retained for up to 24 months in accordance with our privacy policy.

Full Audit

Running all 5 steps in sequence is called a Full Audit. Only the Crawlability Check is free; the Snapshot, Analysis, Benchmark, and Report steps each consume credits.

AI Models & Search Endpoints

Higher subscription tiers access more AI models and search endpoints, producing broader and more accurate benchmark results. See the AI Models page for how we use models, and the Endpoints chapter for what each tier includes.