“Our AI delivers 100% accurate answers.”
The absolute accuracy claim is not substantiated by the reviewed documentation and evidence.
Replace absolute language with a qualified performance statement that reflects tested conditions and limitations.
AI Claims Validation
AI capabilities evolve. The claims your business makes about them should remain accurate, evidence-backed, and reviewable.
Assessed against documented configurations, supporting evidence, and relevant vendor documentation.
01 /The Problem
AI capabilities, configurations, and vendor conditions change. Business claims can quickly fall out of sync with the systems behind them.
A statement that was once accurate may no longer reflect the current configuration, provider, deployment route, or contractual conditions. Without clear traceability, teams struggle to establish what still supports a claim—and where its limitations begin.
As systems change, claims can become
“Customer data is never used to train AI models.”
“AI requests are not retained.”
“All AI processing stays within a specified region.”
ContextThese claims may depend on provider, configuration, service route, plan, contractual terms, and exceptions.
02 /The Tolvay Approach
Tolvay reviews AI-related claims against documented configurations, supporting evidence, and relevant vendor information to assess what is supported, what needs qualification, and what remains unverified.
Scoped, evidence-based claim review—not blanket certification or compliance approval.
Documented configuration and current vendor terms substantiate the stated claim.
Claim holds only under specific deployment routes, plan tiers, settings, or exceptions.
Required configuration records, policies, or vendor documentation are missing or stale.
| Claim basis | Evidence reviewed | Assessment |
|---|---|---|
| Direct match | Documented configuration and current vendor terms substantiate the stated claim. | Supported |
| Conditional scope | Claim holds only under specific deployment routes, plan tiers, settings, or exceptions. | Needs qualification |
| Evidence gap | Required configuration records, policies, or vendor documentation are missing or stale. | Unverified |
Review principles
03 standards
We review the actual statement being made—not broad assumptions about an AI provider or model.
Findings are grounded in available documentation, configurations, records, and relevant sources.
Conditions, limitations, exceptions, and missing information remain visible.
03 /The Process
Every review follows a structured process designed to understand the claim, examine its supporting information, identify gaps, and deliver actionable findings.
Define what is being asserted, the relevant AI systems or services, and the boundaries of the assessment.
Examine relevant AI configurations, provider information, available documentation, and customer-provided materials.
Identify the supporting information, conditions, limitations, exceptions, and evidence gaps behind each claim.
Assess the supportability of each claim and identify where wording, evidence, or conditions require attention.
Provide practical next steps, suggested claim corrections, qualifications, and areas requiring additional evidence.
Delivery model
Human-reviewed initial assessment grounded in submitted claims, available documentation, and vendor information—not automated monitoring or certification.
04 /Review Scope
AI claims rarely exist in isolation. Their accuracy depends on the models, providers, configurations, data practices, documentation, and conditions behind them.
Review what the AI system is capable of doing, how capabilities are described, and whether claims accurately reflect the underlying model and its limitations.
Identify relevant AI providers, underlying services, third-party dependencies, and distinctions between the products or services being represented.
Examine claims concerning data usage, retention, training, processing, and privacy-related conditions against available documentation and evidence.
Consider relevant settings, deployment choices, model selection, routing, and configuration dependencies that may affect whether a claim holds true.
Evaluate the available documentation, technical references, customer materials, and evidence supporting individual claims.
Identify limitations, exclusions, conditional statements, exceptions, and circumstances in which a claim may not accurately apply.
Review boundary
Evaluated through human review of provided documentation, configurations, and public vendor materials—not automated scanning, live system integrations, or certification.
05 /Illustrative Findings
A claim can sound definitive while its supporting evidence tells a different story. Here’s how an AI claims review can surface the distinction.
The absolute accuracy claim is not substantiated by the reviewed documentation and evidence.
Replace absolute language with a qualified performance statement that reflects tested conditions and limitations.
06 /Who it's for
When AI becomes part of your product, your claims become part of your customer relationship. Tolvay helps teams assess whether those claims are supported by the evidence behind them.
Companies embedding AI into their products and communicating AI capabilities, limitations, data practices, or model behavior to customers.
Product claims evolve as models, configurations, providers, and AI features change.
Founders, CTOs, and engineering leaders responsible for the accuracy of AI-related product and company claims.
Need confidence in what their company communicates without spending weeks manually reviewing every technical statement.
Teams preparing for enterprise procurement, security reviews, vendor questionnaires, and AI-related due diligence.
Customer questions increasingly require clear, supportable answers about how AI is used and configured.
Teams using AI claims in websites, sales decks, proposals, demos, and customer conversations.
Overstated or outdated claims can create friction and undermine trust during sales conversations.
07 /Start with Five
Choose five AI-related claims from your website, product documentation, or sales materials. We’ll review them at a high level and identify where they appear supported, overstated, conditional, or in need of further evidence.
Requests are sent directly by email—sending your email is the submission step. Include up to five AI-related claims and their source URLs (fewer is fine, and additional context is optional). Tolvay manually reviews each request and responds where appropriate.