Back
Methodology changes

Changelog

Every revision to the scoring rubric or AI prompt. What changed, when, and why. Past scans are never re-scored, so a scan’s date implicitly fixes the methodology version it ran under.

v1.4.0Sentence rhythm, search/social previews, and indexability

  • Reports now show a sentence-rhythm chart: one bar per sentence of the crawled text, in reading order, with average sentence length, rhythm variation, and vocabulary diversity alongside. Even cadence is one of the strongest machine-writing tells; this makes it visible instead of only describing it. The chart is a heuristic lens computed from the crawl — it is not an input to the AI score.
  • A new search & social preview section rebuilds how the page presents in a Google result and as a shared link, from the tags we crawled — and flags missing or truncation-prone titles and descriptions, and a missing share image.
  • Indexability is now surfaced: the page identity card lists the robots directive and canonical URL, and a page that opts out of search indexing gets a prominent alert at the top of the report, since that overrides everything else ranking-wise.
  • LLM-marker phrase counts are now contextualised as a density per 1,000 words with a plain-language density band, and a page with zero matches now says so instead of hiding the section — absence of stock AI vocabulary is itself a signal.
  • The by-the-numbers strip gains a text-to-HTML content ratio.
  • Backwards-compatible: everything above is computed at render time from the crawl snapshot already stored with each scan, so existing reports gain the new sections without being re-scored. Scores and verdicts are unchanged.

v1.3.2Reasoning summary moved up; report previews use it

  • The model's one-paragraph reasoning summary now renders directly beneath the verdict, so the plain-English explanation of the score is the first thing a reader sees.
  • Link previews and search snippets for a report now use that same per-scan summary instead of a generic score string, so a shared report describes the page it analysed rather than just the numbers.
  • Scoring is unchanged, and no report content was added or removed — only the section order and the preview text moved.

v1.3.1Pro plan required for the full governance breakdown

  • The governance section now gates on plan. Free viewers see the alert summary, the score band, and the first risk as a teaser; a Pro CTA stands in for the remaining risks.
  • The top-of-report alert card stays visible to everyone — the count and severity badges still answer how many risks were found and how severe they are on a free scan.
  • Pro and Business viewers see every risk's plain-English explanation, regulatory framework, official guidance links, and copy-paste fix snippets.
  • Scoring is unchanged. Only the rendering branches on the viewer's plan; a cached scan reads the same numbers as before, what changes is what the page chooses to show.

v1.3.0Content-domain detection and governance scoring

  • Reports now classify the page's content domain — what kind of site it is, in plain-English sector labels — and use that classification as one input to the new governance score below.
  • A new governance score sits alongside the SEO and AI scores. It summarises how the page's content domain and its AI-automation signal combine into regulatory and trust exposure for that kind of site.
  • Findings now include a governance risk section: each risk renders as its own card with a severity, a plain-English explanation of the exposure, and a pointer to the relevant regulatory framework when one applies.
  • Mitigations the page is already doing — visible AI disclosures, expert-author bylines, regulatory boilerplate — are listed alongside the risks rather than silently folded into the score, so readers can see what is working.
  • Backwards-compatible: scans cached under earlier versions render without the new section. Past reports are not re-scored.

v1.2.0Evidence, recommendations, and a richer page snapshot

  • Findings now ship with verbatim evidence quotes pulled from the page, tagged by where they appeared, so readers can see exactly what triggered the call.
  • Findings now carry a one-sentence recommendation: the single most actionable next step for that issue.
  • Reports now name what the page is doing well alongside the findings, not only what to fix.
  • Reports surface a page identity card (title, meta description, final URL, language, OG card), by-the-numbers stats (words, images, alt coverage, link counts, schema blocks, HTML size), the full H1–H3 outline, and a sample of internal links checked for broken paths.
  • Backwards-compatible: scans cached under earlier versions render without the new fields. Past reports are not re-scored.

v1.1.0Public release

  • Initial public methodology: 7 SEO checks, 4 AI signal categories (content / structure / imagery / tone), 3 verdict bands.
  • Two-model pipeline: a fast first pass, with escalation to a stronger model when the first pass returns low confidence.
  • Findings capped at 3 per scan to keep reports scannable.
  • Single-page scans only. Multi-page passes deferred to a later release.