AI vendor claim evidence checklist: walk one page or claim set
Last reviewed May 30, 2026
Use this checklist when one vendor page or pasted claim set is open in front of you. Walk each AI-related sentence in order, assign the claim type, match the evidence a buyer would expect, and note where scope limits or absolute wording exceed what is shown.
Evidence buyers verify
- Exact claim text and the vendor's page URL.
- Evidence matched to the claim type, not a general product document.
- Scope limits visible in the vendor's user-facing language.
Opens the checker for this claim type. Paste your vendor's exact wording there. Evidence questions only — not a blacklist or fraud detector. Not sure what a result looks like? See a sample receipt.
Sources this guide draws from
- · Case timeline through February 27, 2024
Source for AI-linked ecommerce income, profitability, passive-investor, and storefront outcome claims.
- · April 28, 2025
Source for AI detector accuracy and average-user performance wording.
- · April 22, 2025
Source for AI-powered accessibility compliance wording.
- · December 3, 2024
Source for facial recognition accuracy, bias, and anti-spoofing marketing claims.
Public claims with documented evidence gaps
"use artificial intelligence to boost earnings for consumers' e-commerce storefronts"
ROI / Outcome- Source and date
- FTC Automators case page · Case timeline through February 27, 2024
- Evidence signal
- AI contribution is used to support earnings wording without separating the AI workflow from coaching, inventory, storefront, and marketplace factors.
- Evidence gap
- A buyer should ask for store-level outcome distribution, total costs, unsuccessful-store rate, time to revenue, and a method for isolating the AI feature from surrounding services.
- Buyer question
- For an AI-boosted ecommerce earnings claim, what result remains after coaching, inventory, marketplace, and customer-labor effects are separated?
"developed using a wide range of material, including blog posts and Wikipedia entries"
Accuracy / Performance- Source and date
- FTC Workado proposed order release · April 28, 2025
- Evidence signal
- Training-material breadth used to support average-user accuracy without showing the content mix or benchmark context.
- Evidence gap
- A buyer should ask for the training and test content categories, general-purpose benchmark, false positive rate, false negative rate, and comparison baseline behind this wording.
- Buyer question
- For a wide-range training-material claim, what content categories and benchmark show the detector works outside academic content?
"AI product could make websites compliant with accessibility guidelines"
Compliance / Safety- Source and date
- FTC accessiBe · April 22, 2025
- Evidence signal
- Broad compliance promise with no visible exclusions.
- Evidence gap
- A buyer should ask for the standard version, issue coverage, audit method, manual remediation boundary, and maintenance limitation.
- Buyer question
- For an accessibility compliance claim, what scope limit should appear directly in the marketing copy?
"trained on millions of faces"
Accuracy / Performance- Source and date
- FTC IntelliVision · December 3, 2024
- Evidence signal
- Training-data scale wording used to imply performance or bias coverage without showing dataset composition or subgroup results.
- Evidence gap
- A buyer should ask for dataset composition, representativeness, duplicate handling, subgroup coverage, and the measured link between training data and deployment performance.
- Buyer question
- For a trained on millions of faces claim, what dataset documentation shows the claimed scale supports the intended demographic and deployment setting?
Claim review checklist
- Copy the exact AI claim before editing Claim text and source date
What exact words will appear on the page: accuracy number, compliance promise, replacement claim, outcome claim, or first-of-kind wording?
- Assign the claim type Accuracy, automation, compliance, ROI, first/only/best, or vague AI-powered
Which claim type creates the highest evidence burden in this sentence?
- Match the claim to evidence Benchmark, audit, subgroup test, customer sample, comparison set, or workflow documentation
Does the evidence test the same use case, content type, audience, and date implied by the copy?
- Check whether a scope limit is visible in the claim Content type, model version, standard version, customer segment, or human review boundary
Can a buyer see where the claim applies and where it does not apply without opening a separate document?
- Identify absolute wording that exceeds the evidence scope Wording boundary reference and limitation note
Does the vendor's copy use words like any, zero, fully, first, only, best, or replace that exceed the tested scope?
- Keep a source record Source URL, checked date, method note, and owner
If a buyer asks for support later, can the team produce the source, date, method, limitation, and owner behind the claim?
Match each claim pattern to the evidence buyers need
| Claim pattern | Evidence needed | Buyer question |
|---|---|---|
| Accuracy number in headline copy | Benchmark, sample size, model list, input categories, error rates, and test date. | What details should be visible next to the performance number or linked from it? |
| AI replaces a professional task | Task boundary, human review point, escalation process, unsupported cases, and quality testing. | Which word in the claim could make a reader think no expert review is needed? |
| Compliance, safety, or bias-free claim | Standard version, audit method, subgroup metrics, exclusions, and retest cadence. | What narrower result can be stated without implying full coverage? |
| First, only, best, or most advanced claim | Comparison universe, date checked, source of comparison, and update responsibility. | Is the comparison still supportable if a competitor changes positioning tomorrow? |
Evidence to request
- Exact claim text and the vendor's page URL.
- Evidence matched to the claim type, not a general product document.
- Scope limits visible in the vendor's user-facing language.
- A record of source URL, checked date, and claim owner.
- An assessment of whether the vendor's wording matches the available evidence.
Questions to put in front of the vendor
- For this claim on the page, which phrase creates the highest evidence burden?
- What source, benchmark, audit, customer sample, or workflow note would a buyer reasonably ask for after reading this sentence?
- Does the copy distinguish a current capability from a roadmap capability?
- Does the vendor's copy use absolute or broad wording that exceeds the evidence scope?
Wording boundaries to compare against
- Reported X% on a named benchmark covering specified inputs and model versions.
- AI drafts first-pass output for human review in defined use cases.
- Supports selected accessibility remediation tasks; final review and maintenance remain separate.
- One of the early products built for a named workflow, based on a stated comparison date.
Have your vendor's exact claim wording ready?
Check this vendor page or claim set How the evidence method works