Reviewed findings & product research

Interpretation movement, visibility evidence, and the sources behind both.

SemanticRisk reviews repeated scans and controlled buyer-prompt observations to distinguish website change from interpretation change, recommendation change and citation-environment change.

22 August 2026 · New research layer

Visibility Evidence enters controlled testing

See the live pilot page
Repeated self-test

SemanticRisk was absent in both repetitions of the first neutral category-discovery prompt

The exact same OpenAI Responses API web-search prompt was executed twice. SemanticRisk was not surfaced in either observation. The first answer contained 15 cited URLs and the second 10; neither contained a SemanticRisk-owned citation.

Why this matters: the target outcome reproduced while the citation environment changed. An aggregate visibility score would collapse those two facts into one number; the evidence layer retains both.
Method direction

Repeatability and evidence movement are now measured separately

The admin review now reports repeated target inclusion/absence, citation-count range and citation-host overlap for each fixed prompt, alongside the exact prompt and source URLs.

Next validation

Broader buyer intents come next

The first prompt is only one category-discovery test. The remaining fixed buyer prompts are being run before SemanticRisk publishes category-level conclusions or a market benchmark.

January–July 2026

Six-month AI interpretation review

Read the full review
11,461scan results
49,974claim observations
798extraction-drift events
72.92%average four-model agreement
Evidence review

Content change is not semantic change

Only 3.84% of 7,976 ordinary content-change comparisons were classified as materially interpretive.

Interpretation drift

Stable content can produce changed output

All 798 extraction-drift events occurred while the normalized content fingerprint remained identical; 87.72% were classified as material.

Method refinement

Paraphrase must be separated from meaning change

Near-balanced claim additions and removals show why semantic matching, split/merge detection and claim-family separation are the next measurement priorities.

July 2026

Recent company-level review

Open live benchmark
Positive movement

monday.com

Moved from 45.55 to 57.75, with stronger structure and semantic clarity signals tied to clear AI/work-platform positioning.

Capture volatility

walmart.com

A negative movement appeared alongside AI-readable text dropping from roughly 30.7k to 3.6k characters.

Capture volatility

paypal.com / staples.com

Both returned successful status while producing substantially smaller captured-text samples than prior scans.

See the evidence, then test your own domain.

Explore the public benchmark, read reviewed findings, or ask about a controlled Visibility Evidence pilot.

Visibility EvidenceOpen benchmark