Sample Claim Receipt
See how one public AI claim becomes evidence questions for a vendor. Demo data only. Hypothetical example — not from a real company.
Questions you could send to a document-review vendor
Example claim: “Our AI achieves 98% accuracy and fully automates document review for any document type.”
Before we rely on the stated accuracy and automation, could you share:
- Which document types, languages, and sample size were included in the 98% result?
- What false-positive and false-negative rates were measured for documents like ours?
- Which tasks and uncertain outputs still need human review?
These questions request evidence. They do not establish whether the claim is true.
Claim Receipt
- AI-related claim
-
"Our AI achieves 98% accuracy and fully automates document review for any document type."
- Claim type
- Accuracy / Performance + Automation / Replacement
- Evidence rationale
- Numeric accuracy wording is paired with broad automation wording, without named test scope or human-review limits.
- Evidence needed for wording
- High
Before relying on this wording, ask for the benchmark scope, test conditions, excluded tasks, and where human review remains required.
- Claim wording signals
-
- Numeric accuracy without benchmark scope
- Full automation wording without human-review boundary
- Evidence to request
-
- Benchmark name, sample size, evaluation date, and document types tested.
- False-positive and false-negative rates for the buyer's likely workflow.
- Escalation criteria and task types that still require human review.
- Buyer questions
-
- Which document types were included in the 98% result?
- Does the benchmark include the same layouts, languages, and edge cases we use?
- Which outputs still route to a human reviewer before use?
- Safer wording examples
-
- "Reported 98% accuracy on a named benchmark for selected document types."
- "Designed to automate defined review steps, with escalation for uncertain cases."
- Review basis
-
- Input reviewed
- https://example.com/ai-document-review
- Scope
- Hypothetical claim written to demonstrate the report format. No vendor page was fetched or assessed.
- Checked
- Not applicable — this is a static example.
- Review mode
- Static example, not a completed check
- Method note
- Signal Mapping demo. Real receipts link to the current method note.
- Source and provenance
-
- Public page fetched?
- No. This example does not come from a real vendor page.
- Source URL recorded only?
- The example URL is a placeholder.
- Source context
- Listed sources provide context for the evidence question. These are not findings about the demo vendor.
- Source page title
- Example AI Document Review Platform
- Publisher
- Example Vendor
- Snapshot hash
- None — no page was fetched.
- Source read note
- No source page was read. A real receipt explains whether its source was fetched, supplied as text, or recorded only.
- Boundary
- Demo receipt only. This is not legal advice, a compliance certification, a procurement approval, or a truth determination. Real receipts are unlisted and noindexed, but anyone with a shared receipt link can view that receipt. Do not use confidential vendor material.
How real receipts differ by input
- URL receipt Can show source URL, page title, publisher, checked date, and snapshot hash when the page is readable.
- Pasted text receipt Reviews only the text provided. URL provenance fields are marked not applicable.
- Pasted text with source URL Reviews the pasted claim text and records the URL as context only; the URL is not fetched or snapshotted.