AI claim review
What an AI washing checker actually checks
Most searches for "AI washing checker" are really looking for the same thing: a way to tell whether an AI product claim is specific enough to rely on, and what evidence it would need to hold up. This page explains what that check involves, shows three common claim patterns, and links to the tool.
What the check looks at
An AI washing claim review does not judge whether a company is honest or fraudulent. It looks at the evidence burden a claim carries — and whether the public-facing wording provides enough scope, limitation, and supporting conditions to be relied on by a buyer.
The review asks: does the wording create a higher evidence expectation than the vendor has disclosed? If so, that is an evidence signal, not a verdict.
Every review returns an evidence-burden note, evidence needed, buyer questions, and a wording boundary — the scope and conditions a claim would need to carry lower evidence risk.
Three common claim patterns and what they need
"Our AI is 98% accurate for any document type."
Accuracy / Performance
Numeric accuracy without visible test scope, content type, or error-rate breakdown.
Test set design, content types covered, model version tested, false-positive and false-negative rates, retest cadence.
Does the 98% figure hold for the document types in your workflow?
"Fully automated — no human review needed."
Automation / Replacement
Full-automation claim with no stated human review boundary, escalation path, or failure handling.
Task scope, human oversight steps, escalation triggers, failure mode documentation.
Which tasks still route to a human reviewer, and what triggers that escalation?
"AI-powered compliance for HIPAA and GDPR."
Compliance / Safety
Compliance promise tied to a named regulatory framework without stated scope, audit, or customer responsibility.
Framework scope, covered controls, audit or assessment method, customer responsibilities, and known exclusions.
Does this cover your deployment configuration, data flows, and sub-processor chain?
Why evidence burden matters for buyers
A vendor claim carries a higher evidence burden the more it promises — in scope, accuracy, compliance, ROI, or exclusivity. When evidence is not disclosed, the buyer carries that gap.
- Accuracy or performance claims need benchmark design, sample size, error rates, and scope limits. A single number without those is not enough to evaluate reliability in your workflow.
- Full-automation claims need a human review boundary, escalation path, and failure mode. Without those, the buyer cannot assess operational risk.
- Compliance or safety claims need a framework scope, audit method, and statement of customer responsibility. A claim that something is "fully HIPAA compliant" tells a buyer nothing about what controls are in place.
- Vague AI-powered descriptors like "AI-native" or "agentic AI" need a description of what AI is actually doing and what its boundaries are.
None of this is a legal determination. Risk level describes evidence burden and wording risk only — not legality, compliance status, or purchase safety.
How to use the checker
- Paste the vendor's public product page URL, or paste the specific AI claim text directly.
- The tool extracts AI-related claims and returns an evidence-burden note, evidence needed, buyer questions, and a wording boundary for each claim.
- Review the Claim Receipt — copy it, share it, or bring it into a vendor evaluation or due-diligence document.
The checker works on public product pages, landing pages, marketing copy, and case study excerpts. It does not support PDFs, private pages, or batch scanning.
Sources
- ASA/CAP AI as a marketing term report Guidance on vague AI descriptor wording and evidence expectations for marketers.
- FTC AI claims enforcement — Content at Scale AI detector accuracy claim and evidence-retention case.
- FTC Business Guidance on AI FTC guidance on keeping AI claims substantiated.
What this tool does not do
- It does not determine whether a company is honest, deceptive, or committing fraud.
- It does not provide a legal opinion, compliance certification, or investment guidance.
- It does not rank companies or publish a blacklist.
- It does not make procurement decisions or approve vendors for purchase.