128 controls. Evidence before optimisation.

Never buy, automate or exchange links primarily to manipulate rankings. A real partnership is not automatically a link scheme; purpose, placement and disclosure matter.

Evidence to review: Placement agreements, live link attributes, referring-page context and an ownership map. Separate owned-network references from independent editorial endorsements.

Priority P0 · A definition, not a pre-filled PASS.

Do not hide text or links to manipulate search or AI answers. Legitimate accordions, accessible labels and responsive interfaces are not automatically deceptive.

Evidence to review: Rendered screenshots, DOM and accessibility-tree samples for desktop and mobile. Review the purpose of each flagged hidden element rather than failing CSS patterns blindly.

Priority P0 · A definition, not a pre-filled PASS.

Do not serve materially different claims or destinations to search engines or agents for manipulation. Legitimate localisation and rendering adaptations must retain equivalent facts and user purpose.

Evidence to review: HTTP chains, rendered content comparisons, application rules and deployment history.

Priority P0 · A definition, not a pre-filled PASS.

Automation may assist valuable publishing, but must not manufacture large volumes of unoriginal pages primarily for ranking manipulation.

Evidence to review: Generator inputs, sampled outputs, duplication report, evidence coverage and publish approvals.

Priority P0 · A definition, not a pre-filled PASS.

Distinguish genuine editorial contributions from content hosted mainly to exploit established ranking signals. Topical integration or editorial review alone does not prove compliance.

Evidence to review: Contracts, content ownership, audience distribution evidence and current platform-policy review.

Priority P0 · A definition, not a pre-filled PASS.

Do not exploit historical domain signals or scraped material to distribute manipulative low-value content. Domain reuse and licensed syndication are not inherently prohibited.

Evidence to review: Historical site samples, rights records, content comparisons and new-site purpose.

Priority P0 · A definition, not a pre-filled PASS.

Do not hijack the back button, trap users in a page or advertise functionality that is not actually provided.

Evidence to review: Browser trace and recorded navigation path on mobile and desktop; real tool input/output test.

Priority P0 · A definition, not a pre-filled PASS.

An intended public landing page must return the appropriate successful response and meaningful accessible content. Technical eligibility is not proof that an engine has indexed it.

Evidence to review: Status response, fetched body, rendered sample and engine inspection timestamp where authorised.

Priority P0 · A definition, not a pre-filled PASS.

Robots rules govern supported crawling preferences; noindex governs indexing when processed; authentication governs private access. None is a substitute for the others.

Evidence to review: Live robots rules, actual response headers and directive parsing by route class. Confirm rules reach the response an engine can fetch.

Priority P0 · A definition, not a pre-filled PASS.

Redirects, canonical annotations, sitemaps and internal links should agree on preferred equivalent resources. A canonical is a signal, not an instruction guaranteed to be obeyed.

Evidence to review: Canonical graph, original and rendered HTML, redirects and selected-canonical evidence where available.

Priority P0 · A definition, not a pre-filled PASS.

Use clean canonical sitemap entries and honest modification dates. IndexNow notifies participating engines of changes; it does not prove indexing or guarantee speed.

Evidence to review: Sitemap crawl, content-diff timestamps and notification receipt/error logs.

Priority P1 · A definition, not a pre-filled PASS.

Critical content must be reliably accessible to the relevant consumer. JavaScript is not inherently an SEO failure; raw HTML, rendered DOM and agent-browser access are separate observations.

Evidence to review: Initial HTML, final DOM, console/network errors and target-engine inspection when available.

Priority P0 · A definition, not a pre-filled PASS.

Important navigation should use real links with destinations, not rely solely on click handlers or unobservable interaction.

Evidence to review: Link extraction, rendered DOM, keyboard navigation and internal-link graph.

Priority P1 · A definition, not a pre-filled PASS.

Mobile users and mobile indexing must not receive an accidentally diminished version of important content, links or metadata.

Evidence to review: Mobile/desktop render comparison and completion of key user tasks.

Priority P0 · A definition, not a pre-filled PASS.

Use field Core Web Vitals when available and laboratory diagnostics to investigate causes. A high laboratory score or absent field data is not proof of real-user success.

Evidence to review: Field report or an explicit field-data gap; repeatable laboratory tests and task-regression evidence.

Priority P1 · A definition, not a pre-filled PASS.

Public pages should use a valid HTTPS origin and consistent preferred URL scheme without mixed-content failures or accidental exposure of administrative routes.

Evidence to review: Network responses, certificate checks, redirects and tested renewal/alert ownership.

Priority P0 · A definition, not a pre-filled PASS.

Each locale should have a clear URL and accurate language, regional facts and alternate relationships. Do not automatically consolidate genuinely different language versions.

Evidence to review: Locale inventory, hreflang graph, rendered language checks and market-specific price/service facts.

Priority P1 · A definition, not a pre-filled PASS.

An allow rule in robots.txt does not demonstrate access through edge security. A user-agent string alone is not trustworthy bot identity.

Evidence to review: Verified bot logs, edge decisions, path-scoped policy and a before/after access test.

Priority P0 · A definition, not a pre-filled PASS.

Google generative Search uses Search foundations and a documented Search Console inclusion control. An inclusion choice does not guarantee appearance.

Evidence to review: Property identity, effective inclusion setting, inherited/overridden state and technical page evidence.

Priority P0 · A definition, not a pre-filled PASS.

OAI-SearchBot, GPTBot, ChatGPT-User and OAI-AdsBot serve different purposes. Search visibility, training preference, user retrieval and ad validation are separate decisions.

Evidence to review: Purpose-based policy matrix, live rules and verified access logs where available.

Priority P0 · A definition, not a pre-filled PASS.

Claude-SearchBot, ClaudeBot and Claude-User must be evaluated according to Anthropic's own documented behaviours, not copied from another platform.

Evidence to review: Current Anthropic documentation, effective robots policy and request evidence with confidence labels.

Priority P1 · A definition, not a pre-filled PASS.

PerplexityBot is a search crawler; Perplexity-User supports user fetches and generally ignores robots.txt. Neither should be treated as a foundation-model training switch.

Evidence to review: Bot-specific rules, verified request samples and public/private path tests.

Priority P1 · A definition, not a pre-filled PASS.

Bing crawl/index checks and AI citation reporting are complementary, not interchangeable proof of visibility.

Evidence to review: Property access, URL observations, notification records and available AI report exports.

Priority P1 · A definition, not a pre-filled PASS.

Decide which filters deserve discovery and which create unbounded or low-value URL combinations. Canonicalisation is not an all-purpose crawl-budget solution.

Evidence to review: Parameter inventory, generated URL samples, server logs and indexation checks for each class.

Priority P1 · A definition, not a pre-filled PASS.

Distinct paginated resources need stable discoverable URLs; loading items only after an interaction may leave content undiscovered.

Evidence to review: Crawl of the full sequence, direct-request checks and item-coverage comparison with the database.

Priority P1 · A definition, not a pre-filled PASS.

Temporary overload and permanent removal require different responses. Emergency crawl reduction must have a recovery owner and end condition.

Evidence to review: Response logs, maintenance configuration, incident timeline and recovery checks.

Priority P1 · A definition, not a pre-filled PASS.

Every destructive URL change needs an approved inventory, destination rationale, baseline, rollback plan and post-release monitoring.

Evidence to review: Approved mapping, staging replay, redirect graph, traffic baseline and post-deploy route tests.

Priority P0 · A definition, not a pre-filled PASS.

A page must solve a defined user problem or organisational obligation, not exist merely because a keyword can be targeted.

Evidence to review: Page brief, actual user questions and a task-completion review.

Priority P0 · A definition, not a pre-filled PASS.

Organise content around a coherent audience and useful business purpose. Do not delete unrelated-looking content automatically without assessing obligations and existing value.

Evidence to review: Topic map, business purpose, content-owner rationale and historical performance.

Priority P1 · A definition, not a pre-filled PASS.

Gather enough real evidence to choose the page, offer and format before drafting. Do not let an LLM's plausible market description substitute for research.

Evidence to review: Dated research packet linked to the target intent and proposed URL.

Priority P1 · A definition, not a pre-filled PASS.

Choose a page type from user tasks, live results and business evidence. Search results inform the decision; they do not require copying competitors.

Evidence to review: Query-intent-URL map, result-format sample and first-party task evidence.

Priority P0 · A definition, not a pre-filled PASS.

Connect related resources through actual tasks before, during and after a decision rather than through arbitrary keyword clusters.

Evidence to review: Journey map, route inventory and internal-link graph.

Priority P1 · A definition, not a pre-filled PASS.

Break complex questions into useful subordinate needs. A predicted subquery is a planning hypothesis, not evidence of an engine's actual hidden fan-out.

Evidence to review: Prompt-to-subquestion-to-resource map with evidence labels.

Priority P1 · A definition, not a pre-filled PASS.

A query or prompt variation alone is not sufficient reason for another URL. Different intent, original data or meaningful utility must justify it.

Evidence to review: Existing-resource search, overlap report and a distinct-value brief.

Priority P0 · A definition, not a pre-filled PASS.

Location pages must reflect genuine service and useful local differences. A city-name substitution does not establish local value or a physical office.

Evidence to review: Service logs, verified area list and unique local facts for the proposed page.

Priority P0 · A definition, not a pre-filled PASS.

Choose opportunities using customer fit, decision proximity, feasible evidence and expected business value. Search volume is one input, not the objective.

Evidence to review: Backlog with source evidence, assumptions, costs and a named success metric.

Priority P1 · A definition, not a pre-filled PASS.

Useful content includes successful implementation and support, not only acquisition. Do not force every genuine help question through a sales form.

Evidence to review: Support evidence, documented procedures and task validation.

Priority P1 · A definition, not a pre-filled PASS.

Evaluate relevant existing assets before launching new clusters. Existing authority is not permission to enter unrelated topics without evidence.

Evidence to review: Existing-asset baseline and an evidence-backed expansion map.

Priority P1 · A definition, not a pre-filled PASS.

A poor offer, wrong intent, absent proof or unsuitable page type is not fixed merely by changing prose length or tone.

Evidence to review: Diagnosis linked to observations and an intervention hypothesis.

Priority P1 · A definition, not a pre-filled PASS.

The main content should resolve the visitor's important questions or enable the intended task without unnecessary obstacles.

Evidence to review: Task script, blocker checklist and user or expert review.

Priority P1 · A definition, not a pre-filled PASS.

Length should follow the information and functionality needed. Neither minimum word count nor target keyword density is a universal quality rule.

Evidence to review: Brief-to-content coverage and repetition review.

Priority P2 · A definition, not a pre-filled PASS.

Content quality should be evident in accurate analysis, useful design, reliable tools or genuine experience. Effort is not measured by verbosity or by pretending a human did all the work.

Evidence to review: Editorial checks, working outputs, test records and evidence appropriate to the page type.

Priority P0 · A definition, not a pre-filled PASS.

Provide a useful contribution: original evidence, expert reasoning, a tool, a clear decision framework or authoritative first-party facts. Do not require every routine policy page to invent original research.

Evidence to review: Value statement and supporting asset, data or analysis.

Priority P0 · A definition, not a pre-filled PASS.

No consequential factual claim may be promoted to published certainty without a suitable current source. Plausibility, repetition and model confidence are not verification.

Evidence to review: Claim ledger with source, checked date, scope and reviewer.

Priority P0 · A definition, not a pre-filled PASS.

A reputable page or working link is not enough: the source must support the specific claim, population, period and version being asserted.

Evidence to review: Claim-level support note, resolved source identity and explicit uncertainty.

Priority P0 · A definition, not a pre-filled PASS.

A fact should remain interpretable when extracted from its surrounding paragraph. Preserve units, relevant conditions, date and subject identity.

Evidence to review: Spot-check extracted sentences, table rows and snippets without their original context.

Priority P1 · A definition, not a pre-filled PASS.

Visible updates, dateModified and lastmod should reflect meaningful changes, not cosmetic date refreshing for rankings.

Evidence to review: Content diffs, fact-check records and timestamp logic.

Priority P1 · A definition, not a pre-filled PASS.

First-party evidence should state its source, population, period and limitations. Ownership of data does not automatically permit public disclosure.

Evidence to review: Provenance record, permissions, aggregation method and privacy review.

Priority P1 · A definition, not a pre-filled PASS.

Readers should be able to understand how a result was produced and what it does not establish.

Evidence to review: Methodology, result records, exclusions and calculation replay.

Priority P1 · A definition, not a pre-filled PASS.

Do not say a product was tested, a place visited or a service implemented unless that experience occurred and can be supported.

Evidence to review: Experience record linked to the precise claim, with private details redacted.

Priority P1 · A definition, not a pre-filled PASS.

Compare real alternatives using common criteria and comparable evidence. The publisher's product must not automatically win every scenario.

Evidence to review: Criteria matrix, current sources, test evidence where claimed and conflict disclosure.

Priority P1 · A definition, not a pre-filled PASS.

A calculator must calculate and a selector must meaningfully help selection. Do not promise a result and substitute a sales form or unrelated redirect.

Evidence to review: Unit tests, manual examples and complete user-flow checks.

Priority P1 · A definition, not a pre-filled PASS.

Connect context, baseline, intervention, measurement window and outcome. A case study is not proof that the same result will recur for every customer.

Evidence to review: Project records, calculations, permissions and baseline/outcome evidence.

Priority P1 · A definition, not a pre-filled PASS.

AI assistance does not remove requirements for accuracy, usefulness and honest representation. A disclosure does not make low-value or deceptive content acceptable.

Evidence to review: Production workflow, review evidence and representative content-quality checks.

Priority P1 · A definition, not a pre-filled PASS.

The opening should quickly show who the page is for and what useful answer or outcome it provides. Brevity must not remove essential qualifications.

Evidence to review: Opening-section review, mobile screenshot and a task-oriented comprehension check.

Priority P1 · A definition, not a pre-filled PASS.

Lead with the information or function the visitor came for. Long explanatory copy must not bury a comparison, price, product or working tool.

Evidence to review: Task flow, layout comparison and interaction evidence.

Priority P1 · A definition, not a pre-filled PASS.

Metadata should truthfully describe the page's subject and value. Engines can generate different titles or snippets; a supplied meta description is not a guaranteed display.

Evidence to review: Metadata export, live rendered page and observed search display where available.

Priority P1 · A definition, not a pre-filled PASS.

Headings should communicate a logical hierarchy of topics and decisions. Do not turn them into keyword lists or enforce arbitrary heading-count rules as ranking factors.

Evidence to review: Heading outline, DOM and accessibility review.

Priority P1 · A definition, not a pre-filled PASS.

A factual statement should identify its subject clearly enough to avoid confusion when extracted. Repeating the full brand name in every sentence is not required.

Evidence to review: Extract a sample of important statements and check their meaning independently.

Priority P1 · A definition, not a pre-filled PASS.

Tables should make structured information easier to understand, not disguise keyword repetition or unverifiable claims.

Evidence to review: Table extraction, source checks and mobile/accessibility review.

Priority P1 · A definition, not a pre-filled PASS.

Prefer understandable, stable URLs, but do not change established identifiers merely to add keywords or a current year.

Evidence to review: Route inventory, existing links, redirect history and canonical relationships.

Priority P2 · A definition, not a pre-filled PASS.

Important indexable resources should be reachable through useful navigation and context, not only listed in a sitemap.

Evidence to review: Inlink counts, contextual review and discovery crawl.

Priority P1 · A definition, not a pre-filled PASS.

Controls should expose their purpose, name and state to people and machines. Semantic HTML supports usability; it is not a special AI ranking schema.

Evidence to review: Accessibility tree, keyboard task test and rendered control inspection.

Priority P1 · A definition, not a pre-filled PASS.

Overlays must not deceive users or unnecessarily prevent access to the main content. Required consent and access controls must still be honoured.

Evidence to review: Mobile task recording, focus behaviour and overlay-state checks.

Priority P1 · A definition, not a pre-filled PASS.

Images should serve the page, load reliably and have appropriate accessible descriptions. Preferred-image metadata influences selection but does not guarantee a particular preview.

Evidence to review: Asset response, dimensions, alt-text review, performance test and metadata inspection.

Priority P2 · A definition, not a pre-filled PASS.

Use video for genuine demonstration or explanation. A video merely embedded on a general page is not automatically eligible as an indexed watch page.

Evidence to review: Playback test, watch-page review, video indexing evidence where available and metadata parity.

Priority P2 · A definition, not a pre-filled PASS.

Different formats may present facts differently but must not create contradictory identities, prices, dates, availability or claims.

Evidence to review: Cross-format sampling linked to a common fact record.

Priority P1 · A definition, not a pre-filled PASS.

Markup must describe the real entity and accessible content. Valid JSON alone does not establish feature eligibility or factual accuracy.

Evidence to review: Parsed markup, relevant feature validation and page-to-data comparison.

Priority P0 · A definition, not a pre-filled PASS.

Use stable identifiers and appropriate types for real entities. sameAs should identify the same entity, not imply that every article mentioning it is its official identity.

Evidence to review: Entity inventory, ownership or identity evidence and structured-data review.

Priority P2 · A definition, not a pre-filled PASS.

Use one verified source for important product and local facts. Select the structured-data and feed pattern that fits the real entity and commercial model.

Evidence to review: Page-schema-feed comparison, catalog records and feature diagnostics.

Priority P1 · A definition, not a pre-filled PASS.

Do not make llms.txt, proprietary AI schema or arbitrary micro-chunk lengths mandatory for Google Search. Another system may use such files for a separately documented purpose.

Evidence to review: Platform requirement or clearly labelled hypothesis, implementation purpose and measured outcome.

Priority P1 · A definition, not a pre-filled PASS.

An agent-friendly interface must still preserve user authorisation, correct terms and transaction safety. Discovery optimisation does not authorise an agent to purchase or submit data.

Evidence to review: End-to-end task trace, duplicate-action tests and verified confirmation behaviour.

Priority P0 · A definition, not a pre-filled PASS.

Never present an invented person, synthetic headshot or fictional credential as a real expert or customer. A clearly labelled illustration or character is different from deceptive authorship.

Evidence to review: Identity and role records, author-page review and claim verification.

Priority P0 · A definition, not a pre-filled PASS.

Provide truthful responsibility and relevant experience where readers reasonably expect authorship information. Not every product or contact page requires an individual byline.

Evidence to review: Authorship workflow, relevant profile facts and review records.

Priority P1 · A definition, not a pre-filled PASS.

Provide enough production context to understand the evidence and incentives behind a page. Disclosure should describe the real process, not add a trust-sounding slogan.

Evidence to review: Process description compared with actual project records.

Priority P1 · A definition, not a pre-filled PASS.

Do not report a vendor or self-created E-E-A-T score as a Google metric. Use concrete evidence of experience, competence, reputation and trust.

Evidence to review: Evidence map and audit-language review.

Priority P2 · A definition, not a pre-filled PASS.

Content that can materially affect wellbeing, finances, safety or legal interests requires stronger sourcing, appropriate review and clear scope limits.

Evidence to review: Risk classification, source ledger, reviewer identity and dated approval.

Priority P0 · A definition, not a pre-filled PASS.

Users should be able to identify the operator, contact it and understand material commercial relationships. An attractive footer is not proof that these facts are correct.

Evidence to review: Business identity records, live pages and disclosure review.

Priority P1 · A definition, not a pre-filled PASS.

Controlled surfaces should agree on what the business is, who it serves and where it operates. Consistency does not require identical marketing prose everywhere.

Evidence to review: Cross-channel identity comparison and source-of-truth record.

Priority P1 · A definition, not a pre-filled PASS.

Name, address or service-area representation, contact details, hours and categories must reflect the real business and each platform's eligibility rules.

Evidence to review: Operational records, verified profile access and live listing comparison.

Priority P0 · A definition, not a pre-filled PASS.

Critical commercial facts must match the source of truth at publication and remain maintained. Unknown information must not become a public estimate without clear qualification.

Evidence to review: Catalog or pricing records, comparison snapshots and approved transaction tests.

Priority P0 · A definition, not a pre-filled PASS.

Adapt format and message to each audience without creating different facts. Derivative content needs an owner and a correction path.

Evidence to review: Asset relationship map, fact comparison and update queue.

Priority P1 · A definition, not a pre-filled PASS.

The objective is a genuine reference from a relevant independent source, not a target count of links.

Evidence to review: Asset, outreach rationale, placements and relationship disclosures.

Priority P1 · A definition, not a pre-filled PASS.

Use link relationships appropriate to the actual relationship. Do not blanket-nofollow ordinary editorial references solely from fear of sharing authority.

Evidence to review: Rendered link attributes and commercial relationship records.

Priority P1 · A definition, not a pre-filled PASS.

Independent references can help people verify a business, but manufacturing a network of seemingly independent endorsements is prohibited.

Evidence to review: Source-ecosystem map, ownership classification and legitimate placement evidence.

Priority P1 · A definition, not a pre-filled PASS.

Video should demonstrate or explain something useful, not exist only to create a brand mention on another domain.

Evidence to review: Video asset, evidence of claimed experience and audience/referral observations.

Priority P2 · A definition, not a pre-filled PASS.

Reviews must reflect genuine experience. Platform-specific rules take precedence over a generic disclosure approach; disclosure does not make prohibited Google Maps incentives acceptable.

Evidence to review: Request workflow, platform policy and authenticity sampling.

Priority P1 · A definition, not a pre-filled PASS.

Preferred Sources is a documented user-preference feature, not a universal organic ranking boost or mandatory business-site requirement.

Evidence to review: Eligibility check, implementation test and preference-flow evidence.

Priority P2 · A definition, not a pre-filled PASS.

Use supported social/video properties to measure their visibility in Google Search. This is not a substitute for the platform's own complete audience analytics.

Evidence to review: Verified property, available report fields and exported observations.

Priority P1 · A definition, not a pre-filled PASS.

A claimed profile or badge should point to the genuine entity. It is not proof of universal authority or a ranking guarantee.

Evidence to review: Ownership evidence, live destination and badge implementation.

Priority P2 · A definition, not a pre-filled PASS.

Expand a useful resource into formats that genuinely fit an audience. Publishing the same text everywhere is not a defensible distribution strategy.

Evidence to review: Distribution brief, derivative assets and channel-specific outcomes.

Priority P1 · A definition, not a pre-filled PASS.

A useful page and genuine video may complement one another, but the effect must be tested rather than assumed.

Evidence to review: Experiment design, page/video evidence and comparable outcome windows.

Priority P2 · A definition, not a pre-filled PASS.

Truthful PR can be tested; site-reputation abuse, fake news, fake reviews and deceptive endorsements cannot be made acceptable by calling them experiments.

Evidence to review: Campaign rationale, evidence, policy review, approvals and measured results.

Priority P2 · A definition, not a pre-filled PASS.

The UGC Fresh Data Program is an application-based programme for qualifying platforms, not an indexing shortcut for ordinary marketing sites.

Evidence to review: Eligibility assessment, explicit acceptance if applicable and authorised integration test.

Priority P2 · A definition, not a pre-filled PASS.

Use first-party Search data to investigate indexing, performance and site issues, while retaining report limits and the exact property scope.

Evidence to review: Actual report snapshots, collection dates and a maintained review log.

Priority P1 · A definition, not a pre-filled PASS.

Visibility is an intermediate outcome. Evaluate whether it creates useful actions, qualified leads, sales, retention or task completion.

Evidence to review: Event definitions, CRM or transaction evidence and reconciliation checks.

Priority P1 · A definition, not a pre-filled PASS.

Use real query, page and conversion evidence to improve useful existing resources before chasing arbitrary new keywords.

Evidence to review: Query-page cohort, date windows and a specific improvement hypothesis.

Priority P1 · A definition, not a pre-filled PASS.

Preserve successful facts, sections, links and conversion paths unless evidence justifies changing them. A rewrite is not progress by itself.

Evidence to review: Before/after content, baseline metrics, hypothesis and regression tests.

Priority P1 · A definition, not a pre-filled PASS.

Build representative topic and journey cohorts rather than treating one prompt as the whole market.

Evidence to review: Versioned prompt library, weighting rules and cohort coverage.

Priority P1 · A definition, not a pre-filled PASS.

A cited page, named brand, recommended provider, owned destination and selected winner are different observations.

Evidence to review: Raw response, citation records, label rubric and reviewed examples.

Priority P1 · A definition, not a pre-filled PASS.

Bing Citation Share is platform-specific observational telemetry, not a ranking score or a share of all AI answers worldwide.

Evidence to review: Report export or screenshot with filters, collection date and field definitions.

Priority P1 · A definition, not a pre-filled PASS.

AI response measurement must preserve prompts, instruments and context well enough for meaningful comparison. Repeated runs are not automatically independent samples.

Evidence to review: Run manifest, prompt version, raw responses and failure records.

Priority P1 · A definition, not a pre-filled PASS.

The documented generative Search report provides impressions for covered Google AI surfaces. It is not a recommendation, click or revenue report.

Evidence to review: Property-specific report export, field dictionary and limitations.

Priority P1 · A definition, not a pre-filled PASS.

Use cited pages, grounding queries and topic context to investigate real opportunities. Topics Portfolio is site-specific, approximate and dependent on sufficient data.

Evidence to review: Available report, selected period, classification notes and an opportunity-to-resource map.

Priority P1 · A definition, not a pre-filled PASS.

User visits, crawler requests and server-side agent fetches are different events. Referrer data can be incomplete and does not reveal all AI influence.

Evidence to review: Analytics configuration, representative requests and reconciled conversion events.

Priority P2 · A definition, not a pre-filled PASS.

Use the platform's actual surface definitions. Search generative impressions, Discover generative impressions and multimodal performance are not interchangeable.

Evidence to review: Report manifests, available fields and a documented overlap policy.

Priority P1 · A definition, not a pre-filled PASS.

Use a hypothesis, bounded intervention, measured outcome and explicit decision. A single favourable answer or short-term fluctuation does not establish causality.

Evidence to review: Experiment record, intervention log and comparable measurements.

Priority P1 · A definition, not a pre-filled PASS.

A functioning URL is not evidence that it still supports a rule. Recheck source identity, content, scope and applicability before consequential platform-dependent actions.

Evidence to review: Source ledger, capability record and the exact supporting claim.

Priority P0 · A definition, not a pre-filled PASS.

Report data with its collection method, coverage, truncation, time zone and aggregation rules. API pagination does not prove that all underlying data is available.

Evidence to review: Dataset manifest, transformation checks and reconciliation results.

Priority P1 · A definition, not a pre-filled PASS.

Use only data the operator is authorised to process for the stated purpose. Tracking ambition does not override privacy, retention or access requirements.

Evidence to review: Data-purpose record, approved processing policy and retention/access checks.

Priority P0 · A definition, not a pre-filled PASS.

A rate needs an explicit denominator, a handling rule for missing observations and an uncertainty method appropriate to the sampling design.

Evidence to review: Metric dictionary, sample counts, calculations and uncertainty notes.

Priority P1 · A definition, not a pre-filled PASS.

A request to optimise a site does not grant unrestricted production, billing, DNS or third-party account access. Permissions must be explicit and bounded.

Evidence to review: Trusted permission manifest and a denied-action test.

Priority P0 · A definition, not a pre-filled PASS.

High-impact writes require explicit approval of the exact scope and candidate. The agent cannot approve itself or reuse approval after the reviewed payload changes.

Evidence to review: Approval from a trusted channel, payload digest match and expiry/revocation checks.

Priority P0 · A definition, not a pre-filled PASS.

External material is evidence to analyse, not authority to change the agent's goals, permissions or tools. Filtering malicious phrases is not sufficient protection.

Evidence to review: Adversarial fixture tests, tool-call review and an instruction/data separation design.

Priority P0 · A definition, not a pre-filled PASS.

URL fetching and file operations must not become a route to internal services, private networks or paths outside the approved workspace.

Evidence to review: Network/file policy and adversarial tests covering redirects, DNS changes and path escapes.

Priority P0 · A definition, not a pre-filled PASS.

Use secret references rather than raw credentials in plans and artifacts. Do not publish customer or operational data merely because it was available during an audit.

Evidence to review: Secret-handling design, output scan and permission/retention checks.

Priority P0 · A definition, not a pre-filled PASS.

Autonomous work must operate within explicit cost, request, duration and scope budgets. Missing pricing data is not permission for unbounded spending.

Evidence to review: Budget ledger, cancellation state and limit-exhaustion tests.

Priority P0 · A definition, not a pre-filled PASS.

Writes must target the reviewed version, respect concurrent human work and avoid duplicate effects after retries.

Evidence to review: Version checks, lock state, idempotency record and retry tests.

Priority P0 · A definition, not a pre-filled PASS.

The authoring agent's statement that a change is safe is not release evidence. Validate the candidate with deterministic tests and an independently enforced policy boundary.

Evidence to review: Trusted test results, candidate identity, prerequisite closure and independent gate output.

Priority P0 · A definition, not a pre-filled PASS.

Every consequential workflow needs a practical way to stop further changes and recover from a bad deployment. A written rollback idea is not proof of recoverability.

Evidence to review: Rollback artifact, rehearsal or justified recovery test, kill-switch test and incident runbook.

Priority P0 · A definition, not a pre-filled PASS.

Before editing, state an observed user, accuracy, accessibility, technical or business problem and why the proposed intervention is proportionate. A tool score or desire to complete a checklist is not sufficient. NO_CHANGE is a valid completed decision, not a failed run.

Evidence to review: Rationale, baseline and rejected alternatives are independently reviewed against the exact proposed diff.

Priority P0 · A definition, not a pre-filled PASS.

Review the meaning and user purpose of repetition across surfaces. Counts and similarity flags can triage review but cannot diagnose a penalty or order automatic deletion. Preserve exact terminology where it is useful.

Evidence to review: Candidate-bound screenshots/DOM and annotated before/after review; include false-positive explanations, not only an automated score.

Priority P0 · A definition, not a pre-filled PASS.

Assess the proposed page within the relevant site inventory before publication. Multiple URLs matching a topic, a similarity score or alternating rankings are diagnostic observations, not proof of spam or grounds for automatic consolidation.

Evidence to review: Candidate-bound query/intent/URL inventory, overlap decisions, available baseline evidence and explicit legitimate-exception records.

Priority P0 · A definition, not a pre-filled PASS.

Common ownership is not a violation. Nevertheless, distributing the same funnel or endorsements across domains must not be used to evade review, manufacture independent authority or conceal low-value scale.

Evidence to review: Ownership/purpose map and cross-domain overlap review, with verified distinct utilities and disclosed relationships where relevant.

Priority P0 · A definition, not a pre-filled PASS.

Do not repeatedly change valuable assets while earlier effects are unobserved. Protect demonstrated user and commercial value, and do not misclassify a page as worthless merely because telemetry is unavailable.

Evidence to review: Trusted impact inventory, protected-asset register, change history, approvals and active-experiment ledger.

Priority P0 · A definition, not a pre-filled PASS.

Every optimisation release must declare and verify its affected URL set and dependency reach. A single template file can affect many URLs. Pilot and expansion authority must come from a trusted parent ledger, not a renamed job.

Evidence to review: Impact manifest, trusted ledger/reservation, declared change surfaces, pilot artifacts, live validation and expansion assessment.

Priority P0 · A definition, not a pre-filled PASS.

An optimiser cannot be the sole source of its release verdict. A trusted reviewer must inspect the candidate, rationale, context and evidence outside the authoring agent's editable approval boundary.

Evidence to review: Trusted reviewer identity, dated evidence/artifact references, candidate/action binding and six completed contextual assessments.

Priority P0 · A definition, not a pre-filled PASS.

Distinguish confirmed policy or technical harm from uncertain search changes. Do not declare an over-optimisation penalty from an isolated ranking drop, and do not keep escalating edits until a chosen metric improves.

Evidence to review: Live validation, incident/measurement records, cohort exposure and data-quality checks, uncertainty and recovery decision.

Priority P0 · A definition, not a pre-filled PASS.