Scanner changes and score comparisons
A different score does not always mean your website changed. This history explains verified scanner changes and their effect on findings.
Current scanner version: 2026-10-16.3.
Before comparing scores
Saved reports are not recalculated when the scanner changes. Find the recorded scanner version under “What this report can (and can’t) tell you” in each report. Use the entries below to see whether a difference could come from the scanner itself.
Compare the exact page URL, individual findings, unavailable checks and Google’s performance measurement time as well as the score. Unavailable checks are excluded from scoring, so a change in coverage can move the total. Even matching scanner versions do not establish that a website edit caused a difference.
Compare two saved reports without running a new scan, or read the current scoring methodology.
Release history
These notes describe scanner changes, not every interface or hosting update. They are not a complete archive of past scoring formulas. Release identifiers use a date followed by a revision number.
2026-10-16.3 · Inspect a bounded sample of child sitemaps
New scans follow up to two distinct direct children of a valid sitemap index. Confirmed page membership can now reveal noindex or canonical conflicts, and reports explain how many child lists were inspected.
Effect on reports: Incomplete or unmatched samples do not establish sitewide absence. Further index levels and listed pages are not fetched. Child requests share a four-second deadline, the existing size limits and scan request budget. All 59 checks and scoring rules remain unchanged; older reports keep their original results.
2026-10-16.2 · Check where canonical URLs lead
New reports inspect the response status and redirects of an alternate canonical URL, distinguishing missing destinations and observed loops from results that could not be verified. Same-page declarations reuse the original response.
Effect on reports: Adds an unscored observation in Search visibility; all 59 checks and scoring rules remain. The optional probe reads headers only, follows at most two redirects within four seconds, and stops at the scan’s shared target/DNS request budget. No browser, VM, paid API or database migration is required. Older reports keep their original data.
2026-10-16.1 · Review conflicting search visibility signals
New reports connect sitemap membership, crawler-specific noindex instructions, robots.txt restrictions and canonical declarations into a plain-language search visibility summary with links to the supporting findings.
Effect on reports: These combinations are review notices: the settings may be intentional. The 59 checks and their scoring rules are unchanged. Only already-fetched responses are used; child sitemaps, canonical destinations and actual search indexing are not checked. Older reports retain their original data.
2026-10-15.15 · More useful structured-data findings
New reports identify selected malformed names, URLs and dates, missing organization/person names and article headlines, and article author names that cannot be resolved from the page’s JSON-LD.
Effect on reports: Adds one unscored diagnostic, bringing reports to 59 checks. Missing details are recommendations, not universal schema requirements. Inspection limits and unsupported markup are disclosed. No additional requests or services; existing scores and saved reports are unchanged.
2026-10-15.14 · See what the scanner could read
New reports include a collapsed readable-content preview with headings, paragraphs, lists, detected facts and their sources. It explains which content region was selected and links to related findings.
Effect on reports: The saved excerpt is limited to 6,000 characters and 60 text blocks, with up to 300 characters per detected fact. It is visible to anyone with the report link and included in JSON exports. No extra requests, browser or AI service is used. The 58 checks and scoring rules are unchanged. Older reports have no preview; this is not a reproduction of what an AI model sees.
2026-10-15.13 · Read bounded gzip sitemaps
Gzip-compressed sitemaps are now decompressed before inspection instead of being read as text. Incomplete, corrupt or unsupported gzip files are marked unavailable, with guidance to verify the sitemap separately.
Effect on reports: Affected new sitemap findings and scores may change; saved reports keep their original results. Both the downloaded file and its expanded content must fit the 64,000-byte limit. The existing deadline, 58 checks and request count are unchanged. Listed URLs are not fetched.
2026-10-15.12 · Recognize plain-text sitemap URL lists
A sitemap containing one absolute HTTP(S) URL per nonempty line is now recognized as a plain-text sitemap instead of being treated as malformed XML. An unusable line is identified by its line number without copying its contents into the report.
Effect on reports: Affected new sitemap findings and scores may improve; saved reports retain their original results. The 58 checks, request count and 64,000-byte sitemap limit are unchanged. Listed URLs are not fetched, and ownership or complete site coverage is not verified.
2026-10-15.11 · Follow sitemap declarations with flexible spacing
Sitemap discovery now reads robots.txt one line at a time, accepts spaces and tabs around the colon, and removes comments. Malformed declarations and URLs on separate lines no longer divert the scan from a later usable declaration.
Effect on reports: Affected sitemap findings and new scores may change; saved reports keep their original results. The scanner checks only the first usable HTTP(S) declaration, or /sitemap.xml when none is found. Public-URL safety checks, the 58 diagnostics and the one-sitemap request limit remain unchanged.
2026-10-15.10 · Decode page text using its declared character encoding
HTML and XHTML pages now use byte-order marks, HTTP charset declarations and supported declarations in the first 1,024 bytes before their text is analyzed. This prevents corrupted titles and content on pages using encodings such as Windows-1252, Windows-1251 or Shift_JIS. UTF-8 remains the fallback when no supported declaration is found.
Effect on reports: Affected new text findings and scores may change; saved reports retain their original results. Encodings unsupported by the runtime return an explanation instead of a report. The 58 checks, byte limits and request count are unchanged. This does not run a browser, infer encoding from language, or change supporting-file decoding.
2026-10-15.9 · Read canonical link relationships using HTML token rules
Canonical detection now uses HTML whitespace boundaries inside the rel attribute. Nonbreaking spaces and other lookalike separators no longer create a canonical declaration or a false conflict with a valid declaration. Valid mixed-case and space-separated relationships remain supported.
Effect on reports: Affected canonical findings and new scores may change; saved reports keep their original results. This checks declarations in the received HTML, not the destination or a search engine’s selected canonical. The 58 checks and request count are unchanged.
2026-10-15.8 · Avoid scoring partial or bodyless page responses
Main-page responses with HTTP 204 or 205 now explain that no page content was returned. HTTP 206 explains that a full-page response is needed. These responses stop the scan before PageSpeed or supporting-file requests instead of producing findings from missing or incomplete HTML.
Effect on reports: Affected new scans return an explanation instead of a potentially misleading report. Complete supported pages keep the same 58 checks and scoring rules, including valid HTTP 201 pages. Saved reports remain unchanged. The scanner does not request or combine byte ranges.
2026-10-15.7 · Preserve website errors without reading unusable responses
Error response bodies are no longer downloaded for analysis, and the main page must declare HTML or XHTML before its body is read. A large or stalled error page no longer hides an observed HTTP status such as 403 behind a size-limit or timeout error.
Effect on reports: Affected supporting-file findings and scores may change when a previously unavailable response can now be classified by its HTTP status. Successful HTML and supporting files keep their existing limits. The 58 checks and saved reports are unchanged; no additional requests are made.
2026-10-15.6 · Recognize author bylines with multiple attribute values
Authorship detection now recognizes author within space-separated rel and itemprop values, including case variations in rel. Bylines such as rel="nofollow AUTHOR" are no longer missed. Lookalike words and hidden bylines remain excluded.
Effect on reports: Affected new GEO and Content findings and scores may improve; saved reports keep their original results. This remains a source-HTML attribution signal, not verification of identity, expertise or complete microdata semantics. The 58 checks and request count are unchanged.
2026-10-15.5 · Check language tags more accurately
The HTML language check now inspects tag structure, repeated variants and extension identifiers, and registered core subtags using an offline IANA snapshot. It recognizes grandfathered and private-use tags and rejects malformed declarations such as en-US-US.
Effect on reports: Affected new scores may change; saved reports retain their original results. This does not detect the actual content language, enforce recommended variant prefixes or validate extension-specific meanings. Declarations over 1,024 characters are unavailable. The 58 checks and scan request count are unchanged.
2026-10-15.4 · Recognize sitemap feeds
Sitemap inspection now recognizes absolute HTTP(S) entry links in RSS 2.0 and Atom 1.0 feeds. Unsupported URL forms and empty feeds remain unavailable rather than being treated as missing sitemaps.
Effect on reports: Affected new scores may change. This inspects URL discovery, not full feed metadata or indexing; feeds may contain only recent pages. The 58 checks, request limits and saved reports are unchanged. No additional requests are made.
2026-10-15.3 · Avoid false HTTPS failures
An observed HTTPS response still passes. An HTTP response now leaves the secure-connection check unavailable, because Cloudflare can bypass automatic HTTPS redirects for scanner requests. Guidance explains how to verify visitor behavior and scan HTTPS directly.
Effect on reports: Unavailable HTTPS findings are excluded from scoring, so affected new scores may change. Requested and analyzed addresses remain as observed. The 58 checks and request limits are unchanged; saved reports remain unchanged.
2026-10-15.2 · Recognize document formats consistently
Page responses are classified by their actual HTML or XHTML media type. JSON-LD blocks with different capitalization or media-type parameters are recognized, and JSON API submissions accept case variations while rejecting lookalike types.
Effect on reports: Prevents unsupported responses from being scanned as HTML and valid structured data from being missed. The 58 checks, request limits and scoring rules are unchanged. New findings may change affected scores; saved reports remain unchanged.
2026-10-15.1 · Observe optional Markdown responses
A bounded request asks the final page for text/markdown. The new informational finding distinguishes observed Markdown, unsupported negotiation and unavailable inspection, and reports cache declarations without testing cache behavior.
Effect on reports: New reports contain 58 checks. This optional format has zero scoring weight. Content equivalence and linked destinations are not verified; unsupported Markdown does not lower scores. Existing saved reports are unchanged.
2026-10-14.25 · Explain the condition of robots.txt
A new informational diagnostic summarizes the robots.txt response, counts crawler and sitemap declarations, and points to a bounded set of basic syntax observations.
Effect on reports: New reports contain 57 checks. This diagnostic uses the existing response and does not affect scores or change interpreted access rules. It is not full validation or proof of crawler behavior.
2026-10-14.24 · Validate performance measurements before saving
Invalid PageSpeed scores and measurements are treated as unavailable while valid measurements remain visible. Timing values include milliseconds when Google omits formatted text.
Effect on reports: Prevents malformed provider values from making a report unreadable or producing misleading findings. Valid scores, thresholds and the number of PageSpeed requests are unchanged.
2026-10-14.23 · Explain the phases behind LCP
Largest Contentful Paint findings include observed server, resource-loading and render timings when Google supplies a supported breakdown. They use the existing PageSpeed response without another request.
Effect on reports: Helps choose what to investigate. Trace timings can differ from the simulated LCP measurement; they do not change scores or prove a root cause. Missing breakdowns leave the existing measurement intact.
2026-10-14.22 · Explain which crawler policy applies
Crawler findings identify a named bot policy, the wildcard policy, or the absence of either. They cite User-agent declarations and the winning Allow or Disallow rule when one matches.
Effect on reports: Clarifies why a path is allowed or restricted, including Allow exceptions. Matching rules, weights and scores are unchanged; actual bot access is still not tested.
2026-10-14.21 · Canonical addresses in HTTP headers
Canonical checks now read HTTP Link headers as well as the HTML head. Matching declarations pass; conflicts fail. Headers the scanner cannot fully interpret leave the result unavailable.
Effect on reports: Can change canonical findings and scores without a website change. The scanner does not verify the destination or the canonical selected by a search engine.
2026-10-14.20 · Clearer unavailable performance results
PageSpeed findings distinguish timeouts, response-size limits, unreadable responses and other request failures. Guidance explains when to retry and when to use PageSpeed Insights directly.
Effect on reports: Improves diagnosis. It does not change performance thresholds or make unavailable measurements count as failures.
2026-10-14.19 · Author references within structured data
An author identifier can now resolve to a name defined elsewhere in the same page’s structured data, including another script block. Remote profiles are not fetched.
Effect on reports: Previously unresolved author references can satisfy attribution checks. A detected name does not verify identity or expertise.
2026-10-14.18 · Authors associated with the scanned page
Structured-data authors must belong to an article or page associated with the scanned URL. Names on unrelated articles, reviews or collection entries no longer satisfy author checks.
Effect on reports: Can remove misleading passes and change attribution scores. Author metadata and visible author links remain supported.
2026-10-14.17 · Dates associated with the scanned page
Date checks prefer article metadata, then dates on associated structured-data entities, then visible dates in the selected content area. Evidence names the source.
Effect on reports: Can change date findings and scores. A declared date is not proof of publication history, and a visible date can describe an event.
2026-10-14.16 · More useful publisher-link evidence
Publisher and contact findings identify the matched page link and why it matched. Displayed destinations omit credentials, query values and fragments; email addresses and phone numbers are not copied into the new evidence.
Effect on reports: Evidence change only; matching rules, weights and scores are unchanged.
2026-10-14.15 · Names on publisher and image links
Link checks recognize names supplied by explicit labels, references to named elements, image alternatives and title fallbacks, alongside visible text.
Effect on reports: Can correct publisher/contact findings and warnings about unnamed image links. This uses source HTML, not a browser accessibility audit.
2026-10-14.14 · Declared content-use preferences
A new informational finding lists supported Content-Signal declarations in robots.txt. Unsupported or contradictory declarations remain unavailable.
Effect on reports: Adds the 56th check without adding scoring weight. Declared preferences do not prove enforcement or establish effective legal permission.
2026-10-14.13 · More precise mobile viewport checks
Viewport checks read complete properties from all matching meta tags. Lookalike values no longer pass; conflicting width values leave the result unavailable.
Effect on reports: Can change viewport findings and scores. This checks declarations, not the rendered layout or zoom accessibility.
2026-10-14.12 · Resources advertised in response headers
A new informational finding lists selected API, documentation and alternate-format links advertised in HTTP Link headers. Advertised destinations are not fetched.
Effect on reports: Adds the 55th check without adding scoring weight. Presence does not verify that an API or format works.
2026-10-14.11 · Main content declared with an accessibility role
Content selection recognizes a visible element with a leading main role, alongside native main and article elements.
Effect on reports: Can change content counts and related findings. It is still a source-HTML heuristic rather than a computed accessibility tree.
2026-10-14.10 · Google Search crawler access
Googlebot joins the scored search-crawler checks. Google-Extended remains an informational model-use control, separate from Google Search access.
Effect on reports: Adds the 54th check and changes the crawler score to include five search crawlers. Declared access does not guarantee indexing or AI citations.
2026-10-14.9 · Page titles must come from the document head
An SVG logo label or misplaced body title no longer supplies the page title. Missing titles are reported as missing from the HTML head.
Effect on reports: Can remove false title-check passes and change the displayed report title and score.
2026-10-14.8 · llms.txt becomes informational
Both llms.txt observations are excluded from scoring whether the optional file is present or absent. Fetch failures remain unavailable.
Effect on reports: Can change scores by removing an optional convention from the calculation. A missing llms.txt file is not a scored failure.
2026-10-14.7 · Correct robots.txt agent grouping
Extension records no longer split consecutive agent names. An empty final matching group no longer borrows restrictions from the wildcard group.
Effect on reports: Can correct crawler-access findings and scores. It does not test whether a bot can actually fetch the page.
2026-10-14.6 · Scanner version saved with reports
New reports retain and display the scanner release that produced them. Older reports without a recorded version say so explicitly.
Effect on reports: Provenance only; check weights, statuses and score formulas are unchanged.
2026-10-14.5 · Separate search access from training policy
Ten training, model-use and user-requested access policies become informational, whether allowed or restricted. At this release, four search crawlers contributed to the crawler score.
Effect on reports: Can change scores by removing non-search policies from the calculation. Keep intentional training restrictions; they are not a reason to lower readiness scores.
2026-10-14.4 · Notice when content may depend on JavaScript
Reports can flag an empty app container with scripts and very little source content as a possible JavaScript-content limitation.
Effect on reports: Adds context without changing scoring. Absence of this notice does not prove the scanner saw everything a browser would render.
2026-10-14.3 · Performance measurement time
Performance findings show Google’s measurement time when available and explain that Google may reuse recent results. Missing timestamps are not replaced with the report creation time.
Effect on reports: Evidence change only; performance measurements and scoring thresholds are unchanged.
2026-10-14.2 · Stronger sitemap validation
Sitemap checks parse XML structure and usable locations instead of recognizing tag names alone. Unsupported or over-limit responses are unavailable.
Effect on reports: Can change sitemap findings and scores. The scanner still reads only the first declared sitemap or fallback, without following its entries.
2026-10-14.1 · Indexing, canonical and structured-data corrections
Checks combine repeated and scoped indexing directives, distinguish image-preview restrictions from noindex, reject conflicting or misplaced HTML canonicals, respect document base URLs and recognize full Schema.org type URLs.
Effect on reports: Can change findings across existing checks without a website change. This release did not add checks.
2026-10-02.2 · Page context and more reliable content detection
Contextual or inapplicable expectations become informational. Content parsing better handles hidden elements and word boundaries, and recognizes common publisher links and questions in six languages.
Effect on reports: Can change scores and content counts. Tool homepages and other non-article pages are no longer expected to satisfy every article-specific suggestion.
Earlier or unrecorded versions
A report may predate these notes or have no recorded scanner version. Its creation date alone does not establish which rules produced it. Read the evidence saved in that report rather than assuming it used today’s rules. We do not infer or replace missing version information.
If you need a fresh assessment, choose to run a new scan; it uses your normal scan allowance. Keep an export of an older report before its shared link expires. See saved-report help for retention and downloads.