---
title: What can I claim about my AI product in marketing?
description: How the FTC and other regulators treat claims about AI capability, accuracy, results, data use, and reviews, what changed in 2025 and 2026, and how to check your launch copy before it ships.
category: AI features
service: ai-feature-review
updated: 2026-11-17
---

## Short answer

You can say what your product does, as long as you can prove it when you say it. US law doesn't restrict the word "AI." The risk is in claims a buyer would rely on: what the AI can do, how accurate it is, what results it produces, what it does with data, and what other customers say about it. The FTC and state attorneys general treat an unsupported or misleading claim as deceptive whether or not AI is involved, and they have brought a steady run of cases where AI was the selling point.

## Where the rules come from

- **The FTC Act.** Section 5 prohibits deceptive acts and practices. A claim is deceptive if it's likely to mislead a reasonable consumer about something that matters to their decision. Business buyers are protected too.
- **Substantiation.** Under the FTC's long-standing policy, you need a reasonable basis for an objective claim before you make it, not after someone asks. Claims about accuracy, speed, savings, or results need testing that matches what the claim says.
- **State consumer protection laws.** Every state has its own version, enforced by the attorney general and, in many states, by consumers who can sue.
- **Competitors.** A competitor harmed by a false comparison or a false claim about your product can sue under the federal Lanham Act.
- **Investors.** Statements to investors about your AI are covered by securities law. The SEC has brought "AI washing" cases against firms that overstated how they used AI.

## Claims that have drawn enforcement

- **Accuracy numbers.** The FTC's 2025 order against Workado, which sold an AI content detector, bars it from advertising the product's accuracy without competent and reliable evidence. A percentage on a landing page is a factual claim that needs testing behind it, on data like your customers' real inputs.
- **"Replaces a professional."** In its 2025 order, DoNotPay agreed to pay $193,000 and to notify past subscribers after the FTC challenged its "robot lawyer" claims. If you say your product does what a lawyer, doctor, or accountant does, you need evidence that it performs at that level.
- **Results and earnings.** Many of the FTC's AI cases involve claims that AI tools will produce income or business growth: Air AI (sued in 2025 and settled in March 2026, with a ban on the company and its owners marketing business opportunities), Ascend Ecom, FBA Machine, and the "AI-powered ecommerce empire" case against Empire Holdings.
- **What the AI does with data.** In 2026 the FTC finalized orders requiring Cox Media Group and two other firms to pay a combined $930,000 over allegedly deceptive claims about an "Active Listening" AI marketing service. Claims about what your AI collects, hears, or infers are capability claims too.
- **Reviews and testimonials.** The FTC's Consumer Reviews and Testimonials rule, in effect since October 21, 2024, bans fake reviews and testimonials, including ones from people who don't exist, such as AI-generated reviews. Violations can bring civil penalties, which most Section 5 cases can't on a first offense.

## What changed in 2025 and 2026

In December 2025, the FTC reopened and set aside its 2024 order against Rytr, an AI writing tool that the original complaint said could be used to write fake reviews. The Commission said the complaint didn't meet the FTC Act's standards and that the order burdened AI development. The practical message is that the FTC is less likely to pursue a company only because its tool could be misused, and remains willing to pursue false claims about what a product does. Every case above, except Rytr, is about claims.

In July 2026, the FTC proposed a policy statement on AI systems whose accuracy is suppressed or altered in ways users wouldn't reasonably expect. It's a proposal, but it signals interest in whether AI products behave the way their makers say they do.

## Claims that hold up

- **Say what you've tested, in the terms you tested it.** "Answers 94% of billing questions without escalation, in our tests on 2,000 real support tickets" is safer than "answers 94% of questions."
- **Put limits next to the claim.** If accuracy drops for some languages, file types, or edge cases, say so where the claim appears, not only in the terms.
- **Don't imply human review that doesn't happen.** "Reviewed by experts" means a person reviews each output.
- **Match privacy claims to the system.** "We never train on your data" has to be true of your model provider's settings, your logs, and your analytics tools, not only of your own code. See [Can I send user data to an LLM provider?](/guides/sending-user-data-to-llm-providers) and [Can I train or fine-tune a model on my users' data?](/guides/training-models-on-customer-data).
- **Keep the evidence.** Save the test set, method, date, and model version behind each claim. Rerun the tests when the model or provider changes, since an upgrade can make an old number false.
- **Use real reviews from real customers.** Don't write, generate, or buy reviews, and disclose any incentive you gave for one.
- **Check comparisons.** "More accurate than ChatGPT" needs a fair head-to-head test, repeated when either product changes.
- **Hold investor materials to the same standard.** A pitch deck that overstates your AI is a securities-law problem, not just a marketing one.

## Checklist an agent can run against the codebase

- [ ] Find every place marketing copy lives in the repository: the marketing site, app store listings, onboarding screens, pricing pages, email templates, and in-product upsells.
- [ ] List each objective claim: accuracy or success rates, speed, savings, results, "replaces" or "as good as" comparisons, and claims about what the AI does or doesn't do with data.
- [ ] For each claim, find the evaluation, test set, or metric behind it, or note that none exists.
- [ ] Check whether a model or provider change since the claim was written could have changed the number.
- [ ] Check for claims of human review, expert oversight, or "no AI" that the code doesn't support.
- [ ] Check privacy claims against the model provider's configuration (training and retention settings), logging, and analytics events.
- [ ] Check review and testimonial widgets, seed data, and fixtures for invented, generated, or placeholder reviews that could reach production.
- [ ] Check whether incentivized reviews are labeled as such.

## Sources

- [FTC Policy Statement Regarding Advertising Substantiation](https://www.ftc.gov/legal-library/browse/ftc-policy-statement-regarding-advertising-substantiation)
- [FTC Rule on the Use of Consumer Reviews and Testimonials, 16 CFR Part 465](https://www.ecfr.gov/current/title-16/chapter-I/subchapter-D/part-465)
- [FTC: Artificial intelligence cases and statements](https://www.ftc.gov/industry/technology/artificial-intelligence)

## If this applies to you

An **AI feature review** screens your AI claims against FTC guidance alongside the rest of the feature: provider terms, disclosures, and consent. For one change, such as a new launch page or a new accuracy number, a **pre-build feature check** reviews that claim before it ships.
