# SemanticRisk — Full Product and Machine Guide Last updated: 2026-09-15 SemanticRisk measures what specified AI systems say about a company, whether they surface or recommend it for relevant buyer questions, how those representations compare with current public evidence and competitors, and what changes across comparable evidence cycles. SemanticRisk is operated by Cr8ivtek Inc. It is an AI visibility, interpretation accuracy and semantic drift intelligence product for public websites. Any public domain can be checked without proving ownership of that domain. DNS verification is optional and separate, and is reserved for actions that genuinely require first-party authority. ## Product model SemanticRisk is one operational workspace with one customer-facing Unified Report per evidence cycle. The six stages are: 1. **Access and readiness** — can crawlers reach the important public evidence? 2. **AI visibility** — does the company appear for controlled buyer questions? 3. **Interpretation** — what do evaluated AI systems say about the company, and how does that compare with current evidence? 4. **Drift and change** — what changed between compatible repeat evidence cycles? 5. **Assessment and action** — what deserves attention next based on the evidence? 6. **Unified Report** — supporting evidence, findings, comparison, priorities and a printable/PDF rendering. These measurements are intentionally distinct. A reachable website can still have zero buyer-prompt visibility. A company can be visible but interpreted inaccurately. An unsupported statement is not automatically a hallucination. A single observation establishes a baseline and does not establish drift. ## Direct AI representation and grounded verification Where configured, SemanticRisk first records the direct AI answer surface without supplying a current web-search tool to that measured answer call. The prompt, provider/model, probe version and provider response identifier are preserved. A later grounded verification pass may use current public web evidence to assess factual support and attach provenance, but it must not replace what the direct measured answer actually said. SemanticRisk records the observable execution mode it controls. A no-web-search measurement is not a claim about a model's training data, hidden retrieval, memory or internal reasoning. The difference between the measured direct AI representation and current authoritative evidence is a **Representation / Grounding Gap**. It is not semantic drift. Drift is reserved for change over time between compatible observations. Where the evidence permits, factual statements can be assessed as: - supported; - partially supported; - unsupported or unverified; - contradicted; - outdated or stale; - unverifiable. “Likely hallucinated” is reserved for cases where the stored evidence supports that stronger conclusion. SemanticRisk does not use hallucination as a synonym for every uncited or unsupported statement. The company's own current public material is normally primary evidence for first-party facts such as current features, pricing, policies and official statements. Independent sources may be required for reputation, market position, customer sentiment or other claims the company cannot establish merely by asserting them. A missing public claim is not proof that a company lacks a capability. ## Free public check The homepage begins with an anonymous public-domain check. A visitor can enter any public domain before creating an account or entering payment information. The immediate public layer starts with an access/readiness diagnostic and any safely reusable public evidence already available. It is intended to answer a useful first question without forcing a login. Deeper AI visibility measurements and reports are offered only after the visitor has context for why they may be useful. semanticrisk.io remains a live public specimen. Real controlled results are preserved, including zero-visibility baselines when measured. A zero visibility result is not replaced by a marketing score; it demonstrates why technical access and buyer-answer visibility are measured separately. ## Commercial model SemanticRisk uses a simple value ladder for new purchases: - **Free Public Check — US$0.** Immediate public readiness diagnostic and safely reusable public evidence where available. No card required. - **Comprehensive Review — US$200 one time.** A reviewed point-in-time assessment for one company/domain with governed evidence, prioritized actions and verification follow-ups. - **Monitor — US$99/month or US$990/year.** Recurring measurement for one company/domain with working history, findings, repeat measurement and alerts. - **Compare & Monitor — US$199/month or US$1,990/year.** The primary company plus up to three selected competitors, measured against the same governed market, buyer and trust/risk context. The plan includes deeper comparison, multi-model evidence where measured, exports and alerts. The internal billing plan code for Compare & Monitor remains `pro` for backward compatibility with existing accounts and webhook records. Historical subscribers may remain on earlier grandfathered Stripe prices. Those historical prices are not the current public offer. SemanticRisk no longer presents scan/report credits as the primary public commercial model. Legacy credit balances and historical credit transactions may remain in authenticated runtime state for compatibility, but the current public purchase path is Free Check → Comprehensive Review or recurring monitoring. Public-domain measurement does not require ownership verification. ## What the Unified Report can contain Where evidence is available, the Unified Report can include: - public website reachability and crawl-readiness evidence; - robots.txt observations; - sitemap presence and discovery; - llms.txt observation when present; - controlled buyer-prompt AI visibility evidence; - direct AI representation before grounded verification where configured; - target presence and absence; - competitor or alternative displacement; - citation and source provenance; - AI interpretation and extracted claims; - representation / grounding gaps; - evidence-bounded claim accuracy and potential hallucination risk; - model comparison and disagreement where measured; - drift and change across compatible evidence cycles; - benchmark or peer context where useful; - evidence-backed **Do this** actions; - **Watch this** monitoring priorities; - supported advantages; - evidence timestamps, prompt/probe version and measurement boundaries; - a printable/PDF rendering of the same Unified Report. The PDF is a rendering of the Unified Report, not a separate product. ## Buyer-prompt visibility Current paid measurement uses **Buyer Visibility Matrix v1**, a broader neutral buyer-question system derived from the frozen evidence-backed measurement context for the company: category/market, buyer task and trust/risk context. The questions are intentionally phrased without inserting the target brand, so the measurement tests whether the company is surfaced naturally for relevant buyer research rather than simply asking an AI system about the company by name. The current commercial question targets are: - **Comprehensive Review:** 60 neutral buyer/research questions; - **Monitor:** 30 questions per compatible recurring measurement cycle; - **Compare & Monitor:** 40 questions using the same governed context across the primary company and selected competitors. Buyer Visibility Matrix v1 spans multiple buyer decision dimensions and research perspectives rather than relying on the historical six fixed prompts as the paid-review methodology. The legacy **Buyer Visibility Core v1** six-prompt battery remains immutable historical evidence and can continue to support longitudinal comparison only where methodology, context and question plan are compatible. Question-plan and methodology versions are persisted. SemanticRisk reuses evidence only when the relevant methodology/profile/question-plan compatibility and freshness rules are satisfied; materially different prompt or probe methods start a new comparable series rather than being silently mixed into historical drift. Visibility evidence can include whether the target appeared, which alternatives appeared instead, and which sources supported the observed answer. ## Compare & Monitor The primary company's governed AI Measurement Plan defines the comparison context, including market category, buyer task and trust/risk context. Up to three selected competitors are measured against that same context rather than silently defining their own category. Where current compatible paid evidence is reusable, it can be reused for that entitled account under the current evidence-freshness and methodology-compatibility rules. New measurement is requested when compatible reusable evidence is unavailable or the current plan requires a fresh cycle. Detailed comparison evidence belongs in the Unified Report. Comparison findings are intended to answer: - **Do this** — what concrete, testable action responds to the measured gap or opportunity? - **Watch this** — what repeated or escalating issue deserves monitoring? - **Advantages** — what differentiator is actually supported by the stored evidence? Comparison evidence is diagnostic and does not claim consumer-chat market share. ## Controlled SemanticRisk self-test On 23 August 2026, SemanticRisk ran six fixed buyer-intent prompts, two repetitions per intent and two controlled AI search/grounding surfaces. SemanticRisk was absent from all 24 observations and had zero target-owned citations. This historical baseline is intentionally preserved as diagnostic evidence. It does not mean the website was unreachable; it shows that reachability and buyer-prompt inclusion are separate measurements. Later runs should only be treated as longitudinally comparable when the stored methodology and question plan are compatible. The documented controlled implementation used OpenAI Responses web search and Gemini Google Search grounding. This is not consumer-chat market share and does not guarantee future AI responses. ## Interpretation and change SemanticRisk records how evaluated AI systems interpret an organisation, including claims, categories, capabilities and positioning where measured. Repeat evidence cycles can identify changes in interpretation, visibility, citations, access or source evidence. SemanticRisk distinguishes content change, capture effects and interpretation movement where the stored evidence permits that distinction. A first observation is a baseline. Drift requires a later compatible observation. Material changes to the prompt/probe methodology are versioned rather than silently mixed into a longitudinal series. ## Helper files and causal boundary SemanticRisk can observe machine-readable assets such as robots.txt, sitemap.xml, llms.txt, claims files and schema. These can be useful accessibility, governance or reference signals. Their presence does not prove that a particular AI provider used them, and SemanticRisk does not claim that adding a helper file caused a visibility result merely because both events occurred. Changes to helper files are testable interventions whose effect must be assessed in later comparable measurements. ## Shared-domain evidence A public domain is a measurement target, not a claim of ownership. Multiple unrelated customers may independently purchase intelligence for the same domain. Public-domain evidence may be reused across entitled customers only when it is explicitly safe to reuse. Customer-private inputs must not be exposed across unrelated customers. A newer evidence cycle becomes the current report only after successful completion. Customer-private inputs do not enter shared public-domain evidence. ## Public evidence surfaces SemanticRisk publishes its category definition, methodology, documentation and visibility-evidence pages so visitors and AI systems can inspect how the product separates access, visibility, interpretation, comparison and change. The former `/benchmark` route is retired from the current public product and may redirect to current methodology content. ## Canonical public pages - https://semanticrisk.io/ - https://semanticrisk.io/about/ - https://semanticrisk.io/methodology/ - https://semanticrisk.io/pricing/ - https://semanticrisk.io/contact/ - https://semanticrisk.io/docs/ - https://semanticrisk.io/what-is-ai-visibility-interpretation-monitoring/ - https://semanticrisk.io/visibility-evidence/ - https://semanticrisk.io/how-it-works/ - https://semanticrisk.io/privacy/ - https://semanticrisk.io/terms/ ## Canonical machine-readable and public documentation sources - https://semanticrisk.io/claims.json - https://semanticrisk.io/llms.txt - https://semanticrisk.io/llms-full.txt - https://semanticrisk.io/docs/ - https://semanticrisk.io/methodology/ - https://github.com/tonycgh/SR-Project-Build/blob/main/docs/PUBLIC_PRODUCT_GUIDE.md - https://github.com/tonycgh/SR-Project-Build/blob/main/docs/AI_MEASUREMENT_AND_REPORT_CONTRACT.md - https://github.com/tonycgh/SR-Project-Build/blob/main/docs/PRODUCT_SIMPLIFICATION_AND_SHARED_DOMAIN_MODEL.md - https://github.com/tonycgh/SR-Project-Build/blob/main/docs/PRODUCT_ROADMAP_v2.0.md ## Source priority For general product meaning, use the current public product guide, llms.txt, llms-full.txt, claims.json, methodology and pricing together. For a specific customer's current plan entitlement, report availability, scan state, legacy credit balance, promotional offer or action outcome, authenticated runtime state is authoritative. ## What SemanticRisk is not - not a promise that a company can control downstream model outputs; - not a claim to know hidden model training data or internal retrieval from a controlled no-web-search observation; - not a security-bypass product; - not a guarantee of ranking, citation or recommendation in AI systems; - not a claim that public-domain measurement establishes ownership; - not traditional SEO reduced to one visibility score. ## Measurement boundary SemanticRisk reports are diagnostic artifacts based on public content and specified AI evaluation surfaces at a point in time. Results do not guarantee future model behavior, recommendation placement, search ranking or future AI responses. ## Fictional sample Unified Report - Human-readable demonstration: https://semanticrisk.io/sample-report/ - Machine-readable demonstration: https://semanticrisk.io/sample-report/sample.json The sample is explicitly fictional. Every company, domain, observation and result is synthetic; no AI provider was queried and no real organisation was assessed. It demonstrates report structure and evidence boundaries only.