Provenance
Chain of custody
Tier 1–3 source classification, full metadata, and an audit trail with evidence preservation — acceptable vs non-acceptable sources defined.
As AI assistants become the interface between customers and brands, boards, investors, regulators and auditors are beginning to ask a hard question of AI-visibility figures: can you prove this number?
Most current dashboards can’t. They draw on scraped search pages, cached outputs and non-reproducible telemetry — methods that produce a chart but not evidence. If a metric can’t be independently reproduced, it can’t support corporate governance, regulatory compliance or investor reporting. It’s a marketing indicator, not a defensible one.
“DIVM is not a tool — it is the standard against which tools should be measured.”
The Data Integrity & Verification Methodology (DIVM) is the data-trust backbone of the AIVO Standard: the protocol that makes every visibility metric proven, reproduced and defended — turning it from a marketing indicator into a board-grade metric.
The core
Auditor-grade reproducibility
The heart of DIVM is a reproducibility standard borrowed from measurement science, not marketing analytics. A DIVM-conformant figure has to satisfy five conditions.
- Live, attributable interrogation. Every measurement comes from a live query to the model — with the answer text, citations, and model version captured — never from scraped or cached data.
- Statistical reliability. Results carry confidence intervals, coefficient of variation and intra-class correlation — so the number reports its own certainty. CI · CV · ICC
- A replay harness. Any measurement can be independently re-run and must reproduce within tolerance — the test of a defensible result.
- A stated tolerance. Reproduction is held to a ±5% standard; a second analyst re-running a sample must land inside it. ±5%
- Temporal & version control. Every measurement records the model version and timestamp, so genuine change is distinguished from model drift.
Trust, provenance and compliance
Built to survive scrutiny
Reproducibility is one half; the other is provenance and governance — the chain of custody that lets an external auditor trust and verify the result.
Assurance
Third-party replay
Disclosure levels, a standard disclosure pack, replay verification and trust seals — so an outside party can independently confirm a claim.
Security
ISO-aligned
Information-security alignment (ISO 27001 / 27701), privacy and DPIA handling, incident response and retention policy.
And it’s built for a tightening regulatory environment — designed to future-proof AI-visibility reporting against emerging law:
EU AI ActColorado AI ActEU Cyber Resilience ActEU Data ActGDPR / CCPAISO 27001 / 27701
Held to its own standard
The methodology verifies itself
DIVM applies its integrity discipline to its own publication. Every version is formally versioned, approved by the AIVO Governance Council, and carries a SHA-256 document hash logged in a public registry — so any auditor can confirm they’re reading the exact, untampered methodology a measurement was made under.
Document integrity, demonstrated. When a third-party assurance report references DIVM, it references the exact version and hash used at the time of measurement — the same chain-of-custody discipline the methodology requires of the data it governs.SHA-256 · c6f8766b2ffd7a76851af4787b6110c5b1e8436ac72fbaf77e470001c2f8a602
Related research
Where this sits
The full methodology
Read the complete DIVM standard
This article is an overview. The full v1.0 methodology sets out the data-integrity framework, the measurement & reproducibility standards, the governance and compliance model, the transparency and assurance protocol, and the implementation lifecycle — published open-access on Zenodo with a permanent DOI.
Citation: Sheals, P. (2025). AIVO Standard: Data Integrity & Verification Methodology (DIVM) v1.0. AIVO Standard. Zenodo. https://doi.org/10.5281/zenodo.17428849 · Licensed CC-BY-4.0.