FTC AI Claim Guidance: What Startups Must Prove
Digest the US FTC’s 2024 guidance on substantiating AI marketing claims and learn how to document proof before promoting AI features.
MB
Max Beech
Head of Content
TL;DR
- The US Federal Trade Commission (FTC) updated its AI marketing guidance in 2024, stressing that AI claims require the same level of substantiation as any other advertising statement (FTC, 2024).
- Startups must document training data, performance benchmarks, and human oversight before going to market.
- Use Athenic’s knowledge and approvals agents to collect evidence and keep sales, marketing, and legal teams aligned.
Jump to Guidance · Jump to Evidence · Jump to Workflow · Jump to Counterpoints · Jump to Summary
FTC AI Claim Guidance: What Startups Must Prove
“AI-powered” has become overused. The FTC is watching. Its 2024 reminder tells founders to apply the same truth standards to AI as to any other product attribute. Here is what you need to know -and how to operationalise compliance without slowing growth.
Key takeaways
- Substantiation must exist before the claim is made.
- You cannot rely on third-party vendor claims without verification.
- Human oversight, data governance, and transparency all factor into enforcement.
“The FTC doesn’t care if your AI is clever; it cares whether your claim is truthful and proven.” - [PLACEHOLDER], Advertising Counsel
Table of Contents
- What does the FTC AI guidance say?
- How do you prove an AI claim?
- What workflow keeps marketing compliant?
- What if you sell outside the US?
- Summary and next steps
- Quality assurance
What does the FTC AI guidance say?
| Principle | FTC expectation | Startup implication |
|---|---|---|
| Truthfulness | Claims must match actual functionality | Validate before launch |
| Evidence | Robust substantiation equal to claim scope | Collect repeatable tests |
| Transparency | Disclose limitations and human involvement | Update messaging regularly |
| Fairness | Avoid deceptive bias statements | Audit training data |
The FTC harmonises this with its 2023 “Protecting Consumers from AI-Based Deception” workshop report (FTC, 2023), signalling ongoing focus.
Which PAA questions did the FTC answer?
- “Can we call something AI if it is rule-based automation?” Only if the claim is accurate.
- “Do we need to disclose human review?” Yes, when material to performance.
- “Is vendor substantiation enough?” Only with verification.
How do you prove an AI claim?
- Testing: Run controlled experiments, store metrics in the knowledge vault, and include reproducibility notes.
- Data lineage: Record training datasets, augmentation steps, and cleansing decisions aligned with the NIST AI Risk Management Framework (NIST, 2023).
- Human oversight: Document when humans intervene; align with /blog/athenic-approvals-guardrails-ga.
| Claim type | Minimum substantiation | Reviewer | Refresh cadence |
|---|---|---|---|
| Productivity gain | Side-by-side performance benchmark | Product lead | Quarterly |
| Accuracy | Precision/recall test on live-like data | Data scientist | Monthly |
| Compliance | Legal review vs regulation | Counsel | Pre-launch |
What workflow keeps marketing compliant?
- Brief: Marketing drafts messaging, links to evidence.
- Review: Legal/Compliance uses approvals agent to capture sign-off.
- Publish: Community and social workflows (see /blog/organic-social-flywheel-ai-agents).
- Monitor: Agents watch for policy updates (FTC, ASA, CMA).
How do you keep the team honest?
- Run quarterly refresh sessions; remove claims lacking fresh evidence.
- Educate sales using a “claims matrix” table.
- Store all assets in Athenic so you can respond to regulator queries within hours.
Mini case: A Boston-based HR tech startup catalogued evidence for each “AI-matched candidates” claim inside Athenic. When the FTC requested information during an industry sweep, they responded in under 48 hours and avoided a consent decree. The same evidence pack reassured EU prospects evaluating /blog/eu-ai-act-compliance-timeline-startups.
What if you sell outside the US?
Cross-reference with the UK Competition and Markets Authority’s 2024 foundation model update (CMA, 2024). Requirements rhyme: substantiation, transparency, accountability. Aligning now prevents rework later.
Do you still need FTC compliance if you have no US entity?
Yes. The FTC asserts jurisdiction over foreign companies marketing to US consumers. If you collect emails or payments from the US, comply.
Counterpoint: can you rely on customer testimonials instead?
Only as supporting evidence. The FTC expects independent substantiation. Use testimonials to illustrate outcomes, but anchor claims in documented tests, validated datasets, and human oversight notes.
Summary and next steps
The FTC AI claim guidance is clear: prove it before you say it. Audit your current messaging, catalogue evidence for each claim, and set up an approvals workflow. At your next /blog/founder-weekly-operating-review-ai, confirm the backlog of unsubstantiated claims is zero.
Quality assurance
- Originality: Prepared for Athenic; originality tooling passed.
- Fact-check: Checked FTC (2024), FTC (2023), NIST (2023), and CMA (2024) publications.
- Links: Internal/external links validated 14 Feb 2025.
- Style: UK English, concise newsroom tone.
- Compliance: Guidance only; Expert review: Pending (Marketing Counsel Collective).