AI-first company

Governance and risk · Adopt

Claim only what your AI has been tested to do

If you tell customers that your product or service uses AI, or how well it works, hold the evidence before you say it. Regulators treat “AI-powered” and “98% accurate” as factual claims like any other.

By
Last reviewed
Corrections
Report an error
When Do next. Once your first uses of AI are running.
Effort Low (days, one person)
Evidence Emerging

Six documented regulator cases, all in the United States, end the same way, with the outcome stated. No study compares companies that test their claims with those that do not.

Fits
Industries All industries
In the guide What fails

What it is

Check your own marketing. Before your company says “we use AI”, “AI-powered”, “98% accurate” or “does the work of a professional”, have someone confirm two things: the statement is true today, and the company holds the evidence.

This applies to any firm that sells a product or service with AI in it, and to any firm that mentions AI in its sales material.

What the evidence shows

Six cases brought by two regulators in the United States are the basis for this advice. No study has tested it.

  • Saying you use AI when you do not. Two investment advisers described AI they did not have. One said its AI drew on client data. The other called itself the “first regulated AI financial advisor”. They were fined $225,000 and $175,000.

  • An accuracy figure with no fitting test. A vendor advertised its AI-text detector as 98% accurate. The regulator said independent testing found 53% on general content.

  • “As good as a professional.” A service sold as a “robot lawyer” had never been compared with a lawyer’s work. It paid $193,000 and had to notify past subscribers.

  • A promise of compliance. A vendor said its AI plug-in could make any website meet accessibility standards. It paid $1 million. The regulator also acted on paid reviews presented as independent.

  • “AI” that mostly needed a person. Presto Automation told investors its voice AI removed the need for people to take drive-through orders. The securities regulator found that the “vast majority” of orders through Presto’s own system needed a person to step in. Presto agreed to a cease-and-desist order and paid no penalty, because it cooperated.

  • “AI that replaces staff” sold to small firms. The consumer regulator says Air AI marketed “conversational AI” it said could replace customer-service staff, and that some buyers lost up to $250,000. Under a proposed settlement announced in March 2026, the owners are banned from marketing business opportunities, and an $18 million judgment is mostly suspended because they cannot pay. These are allegations; the settlement admits nothing.

In all six, the regulator used rules that already existed: for advertising, for consumer protection, or for what companies tell investors. The securities regulator’s chair called overstating AI in this way “AI washing”.

How to do it

  1. Collect every public statement about AI. Website, brochures, sales decks, social media, press releases, tender documents, regulatory filings.
  2. For each, write down the evidence. What test, on what material, when, and by whom? If the answer is “none”, cut the claim or run the test.
  3. Match the claim to the test. “98% accurate” is true only for the kind of material you tested. Say what that was, or do not give the number.
  4. Avoid absolute words. “Any”, “always”, “fully compliant” and “replaces a lawyer” are promises about every case. One counter-example makes them false.
  5. Say what a person still does. If staff review the AI’s output, say so. The statement is true, and customers value it.
  6. Label paid reviews and endorsements. Disclose any payment or other link to the author.
  7. Keep the evidence on file, with the date. One of the regulators’ orders above requires exactly this.
  8. Re-check when the product changes. A claim that was true of last year’s version may not be true of this one.

What to measure

  • The share of AI claims in current material that have dated evidence on file.
  • The date of the last full check of marketing material.

Where it fails

  • Marketing writes ahead of the product. The claim describes what the product will do, not what it does today. Have someone who knows the product read every AI claim before it is published.
  • Old pages stay up. One adviser’s statements were made over four years, in filings, a press release and its website.
  • Borrowed figures. A supplier’s accuracy number is repeated as your own. Test it on your material first. See ask AI vendors for proof.

By company size

  • Micro and small: read your own website with this question: could I show a regulator the test behind each sentence about AI?
  • Medium: one named person signs off every AI claim before it is published.
  • Mid-market and enterprise: AI claims go through the same review as financial and product claims, with the evidence kept.

Company cases

What happened without it

Sources

No published study tests this practice directly, so it is rated Emerging. The rating rests on the company cases above and on the sources below.

  1. US Federal Trade Commission (25 September 2024). FTC announces crackdown on deceptive AI claims and schemes. Press release. ftc.gov
  2. US Securities and Exchange Commission (18 March 2024). SEC charges two investment advisers with making false and misleading statements about their use of artificial intelligence. sec.gov
  3. US Securities and Exchange Commission. In the Matter of Presto Automation Inc., File No. 3-22413, Release 33-11352 (January 2025). sec.gov
  4. US Federal Trade Commission (24 March 2026). Press release on the proposed settlement with Air AI and its owners. ftc.gov