Skip to content
AuditMe
Measurement transparency

How AuditMe Measures a Website

Every number in an AuditMe report comes from one of four sources, and every finding says which one. This page explains what each source measures, what it cannot measure, and how those measurements become a score.

Category checks, severity & version history: Audit MethodologySee a sample report walkthrough

What data sources does an audit use?

A full audit combines our own crawl with lab metrics, adds real-user field data only when it exists, and applies rules to everything it parsed. Nothing is carried over from another site, another tool or another customer's audit.

Server-side crawl

Always runs

AuditMe fetches the URL over HTTP from our servers and parses the raw HTML, response headers, robots.txt and sitemap. No browser is launched and no third-party index is consulted.

What it measures

  • Title, meta description, headings, canonical, robots meta
  • Internal and external links, redirect chains, image alt coverage
  • JSON-LD blocks and whether they parse
  • Response headers used for security, caching and compression checks

What it cannot do

  • It does not execute JavaScript the way a browser or Googlebot does - JS-dependent content is flagged as a risk, not assumed present
  • It does not see rankings, clicks, conversions or backlinks from other sites
  • It measures the URL you submitted; other pages are only covered if you run a crawl

Lab metrics (PageSpeed Insights)

Runs when a PageSpeed key is configured

A single Lighthouse mobile run on Google's infrastructure produces lab metrics: LCP, FCP, TTFB, CLS, TBT, Speed Index, a performance score and resource evidence such as render-blocking files and third-party weight.

What it measures

  • Lab LCP / FCP / TTFB / CLS / TBT / Speed Index for one simulated visit
  • Named culprits: the hero image, the blocking stylesheet, the heavy script
  • A performance score on a 0-100 scale for that one run

What it cannot do

  • It is one synthetic visit, so it cannot describe what real users experience
  • Consent walls, bot checks or a slow test moment can skew the run
  • Without a key configured the audit still completes, with performance findings marked not captured

CrUX field data

Only when a key is set and the origin has coverage

Real-user p75 Core Web Vitals from the Chrome User Experience Report, queried per origin for the PHONE form factor. Field values take precedence over lab values in every Core Web Vital check, and each finding prints its source tag.

What it measures

  • 28-day rolling p75 for LCP, CLS, INP, FID and FCP from real Chrome users
  • Good / Needs improvement / Poor categories straight from Google's thresholds
  • A source disagreement warning when field looks fine but the lab run failed hard

What it cannot do

  • It is not available for every URL - low-traffic origins have no CrUX record, and we never promise field data for all URLs
  • It needs a key configured on the deployment; without one the field tier stays empty
  • It is origin-level and 28 days behind, so it cannot show today's fix

Heuristics and rules

Always runs

Deterministic rules and heuristic checks score the parsed signals themselves - length ranges, hierarchy, coverage ratios, consistency between fields, entity and answer structure.

What it measures

  • Pass / warning / fail per check with the value it judged
  • Classification per finding: deterministic rule, heuristic or model-assisted
  • Directional guidance such as 'title too long' or 'missing H1 on this URL'

What it cannot do

  • A rule that fires is a signal, not a measurement of user behaviour
  • Heuristics approximate judgement - they cannot prove ranking impact
  • Model-assisted findings are labelled as such and ranked below measured evidence

How do I know which numbers were actually measured?

Each finding is classified into an evidence tier before it reaches the report. The short answer: a number only appears as a number when the underlying source was captured for this audit.

Measured

Direct observation: HTML parse, response headers, a lab run, CrUX field data or search-console rows.

Inferred

Derived by a rule, heuristic or model from page signals. Directionally useful, never quoted as a measurement.

Not captured

The source was not captured - no key configured or no field coverage - so the report states that instead of inventing a value.

How is the audit score calculated?

Every registered check returns pass, warning, fail or info. Checks that do not apply to your page type leave the denominator entirely instead of counting as failures, category pass rates are folded into weighted groups, and the result is rounded to a whole number on a 0-100 scale. Here is the shape of it, without pretending to more precision than the model has.

Weighted groups

16 audited dimensions roll up into these groups
Technical SEO
25%
Content Quality
20%
Meta Tags
15%
Performance
15%
Links & Authority
10%
Schema & Structured Data
10%
AI / GEO Readiness
5%

When a group has no applicable checks it is excluded and the remaining weights are renormalized, so a score never drops because a dimension does not apply to your page type. Grades use fixed published thresholds, not a curve.

What the score is not

  • Not a ranking forecast: we measure on-page factors, not the hundreds of signals that decide positions.
  • Not a benchmark against other companies: the median and top-decile references shown in reports come from AuditMe's own internal dataset ("Internal benchmark reference — as configured September 2026"), not a published industry study.
  • Not fine-grained telemetry: a one- or two-point move is inside the noise of a single crawl. Compare evidence tiers and fixed issue counts instead of the decimal.

How does AuditMe measure AI and GEO visibility?

Short answer: we check whether a site is structurally reachable and quotable by AI systems, and we run one simulated answer test - we do not query the consumer chat engines to see if you appear.

What we check

  • llms.txt presence and completeness at the site root
  • AI crawler access in robots.txt (GPTBot, ClaudeBot, PerplexityBot and friends)
  • Semantic HTML, schema markup and content structure built for citation
  • Entity clarity: Organization, author and brand signals that resolve to one identity
  • Answerability and citation potential: can a snippet answer a specific question on its own
  • Machine and agent readability: extractable content, stable URLs, no JS-only walls
  • One simulated answer test: a single model is asked 12 questions about your brand and we record how often it mentions you. It is labelled a simulation everywhere it appears.
  • Optional web-search visibility via the Tavily and You.com search APIs when their keys are configured - these are web-search APIs, not AI engines.

What we deliberately do not do

  • We do not send queries to ChatGPT, Perplexity or Gemini search products to ask whether you were cited.
  • We do not scrape chat interfaces or buy third-party citation counts.
  • We do not report a "share of AI answers" for your market - we have no data source for it, so we do not publish one.
  • We do not treat a simulated mention as a guarantee: a high score means the structural checks passed and one model mentioned you in a simulation, not that you will be cited tomorrow.
  • Where an AI-written summary appears in the report, it is marked model-generated and separated from measured findings.

How often are results refreshed and verified?

An audit measures the moment you run it. Refreshes are explicit: you re-run, you schedule, or a monitor notices a change - scores are not silently rewritten in the background.

On-demand audit

Run an audit whenever you like; it fetches the page fresh and re-queries lab and field sources on that run.

Scheduled re-audits

A scheduled audit can run daily, weekly or monthly. The daily cron picks up what is due, stores a new audit and advances the next run date.

Verify Fix

After you change the page, re-scan the same URL from a saved audit: the engine diffs the two runs and reports what is fixed and what still fails, issue by issue.

Monitors

Change monitoring, crawl monitoring and incremental refresh jobs run daily, so alerts come from a fresh fetch rather than an old cached score.

One caveat we do not control: CrUX and PageSpeed values are Google's data on a 28-day rolling window, so a fix made today may take days to show up in field metrics even though the crawl already sees it.

Frequently asked questions about how we measure

Does every audit include real-user Core Web Vitals?

No. Field data is used only when a CrUX key is configured on the deployment and the audited origin has Chrome UX coverage. When either is missing, the audit runs on lab (PageSpeed) data plus heuristics and labels the missing source as not captured instead of substituting a guess.

How many checks does an audit run?

The full audit runs 159 registered checks across 16 dimensions (Meta Tags, Content Quality, Technical SEO, Links and the rest). 85 standalone tools cover single questions such as a robots.txt or canonical check.

Does AuditMe ask ChatGPT, Perplexity or Gemini whether my brand is cited?

No. We never query those services to check citations. AI visibility is measured with crawler-access, llms.txt, schema and entity checks, plus one simulated answer test: a single model is asked 12 questions about your brand and we record whether it mentions you - a simulation, not a direct query of ChatGPT, Gemini or Perplexity.

Is the audit score a promise of rankings?

No. The score is a weighted summary of on-page checks you can act on. Rankings depend on hundreds of signals we do not measure, including backlinks and competitor moves, so treat the score as a work list, not a forecast.

How do I know which numbers were actually measured?

Every finding carries an evidence tier: Measured (direct observation of HTML, headers, a lab run, field data or search-console rows), Inferred (derived by a rule or model) or Not captured (the source was not captured, so no number is claimed).

Can I verify a fix instead of trusting a new score?

Yes. Re-run the audit or use Verify Fix on a saved audit: the same URL is re-scanned and the report diffs what is fixed against what still fails, issue by issue, so progress is a comparison rather than a new number in isolation.

Where to go next

This page covers measurement sources and what they can prove. The category-by-category check lists, severity ladder, crawl limits and methodology version history live on the Audit Methodology page; a worked example of the report itself lives on the Sample Report page.

Read the audit methodologySee a sample reportRun a free audit

Related tools & guides

Proof Pass
Expert review and a signed proof of an audit
AuditMe vs Screaming Frog
How a scored audit differs from a crawler