When a vendor says "AI-powered", ask for the measurement.

The words “AI-powered” on a sales page should make a business owner ask for more proof, not less. We think an owner should refuse to buy any AI product until the vendor shows a measurement on work like the owner’s own. A demo and a confident pitch do not count. Our reason is a case that the Federal Trade Commission closed on August 27, 2026. The buyers in that case were small businesses.

The service that did not listen

Three marketing firms sold a service called “Active Listening”. According to the FTC’s complaints, they claimed it listened to people’s conversations through smart devices, in real time, and used what it heard to target ads. They told small businesses it would put their ads in front of consumers in the places those businesses wanted to reach.

The FTC said the service did not listen to conversations and did not use voice data at all. It also said the service did not place ads accurately in the locations the customers asked for. The final orders require Cox Media Group and two smaller firms to pay $930,000 in total. Cox Media Group pays $880,000 of it. These were settlements of allegations, not a court verdict.

Two things in this case should bother an owner.

The first is who was fooled. The buyers were businesses purchasing advertising, the same way a roofing company or a dental practice does. One of the three sellers is an established media company. A known name on the contract did not protect the buyers.

The second is stranger. The FTC pointed out that if the service had worked as advertised, collecting people’s voice data without proper consent would itself have broken the law. The buyers were sold something that was either false or illegal. Either way, the buyer loses.

This has a name

Calling a product AI when it is not, or claiming it does more than it does, is often called “AI washing”. The FTC has gone after it before. In September 2024 it announced five cases at once under the name Operation AI Comply. One company agreed to pay $193,000. In another case, the complaint said a scheme that promised “AI-powered” tools for online stores had taken at least $25 million from consumers. To be fair to the record, the FTC later set aside one of those five orders, in December 2025.

A regulator acts after the money is spent. The FTC first announced the Active Listening settlement in May 2026, and the orders became final in August. An owner cannot wait for the FTC to check a vendor.

Why small companies are the easy sale

A company with a technology team can test a claim. Someone on staff asks how the product works, and can tell a real answer from a vague one. A company with no technology team has to take the vendor’s word.

In a Goldman Sachs survey from March 2026, 76% of small businesses said they use AI, and only 14% said it is fully part of their core operations. The owners named what holds them back: a lack of technical expertise, difficulty choosing tools, and worries about data privacy. Those are owners who want to buy and say they cannot judge what they are buying.

What proof looks like

We will use our own work as the example, because we know exactly where it is weak.

We redesigned the tools of an AI agent and claimed it now reads 61 to 73% fewer tokens for each task. Tokens are the unit that AI providers bill. Behind that claim is a test. We ran the old design and the new design on the same 24 tasks. The wording was the same each time, and the database was restored before each run. There were 595 runs across three AI models, and 589 of them were scored.

The claim also has limits. It is our own benchmark and not an independent audit. We measured tokens and time, not dollars. Each model is compared only with itself.

We print those limits next to the number. That is the standard we would hold any vendor to, and it does not take a technology team to apply it. Ask the vendor what they measured, on how many cases, and what the result does not show. If they give you one good number and no limits, treat it as advertising.

Then ask to see the product run on your own work before you sign. Bring a week of real invoices, real emails or real calls. A product that works will survive that. Ask what happens when it gets something wrong, and who at your company finds out.

And ask one more question, the one this case raises: if this does what you say it does, is it legal?

If you are about to buy an AI product and cannot tell if the claim is real, book a 30 minute call and we will look at it with you.