Back to News & Insights
AI & Data September 2, 2026 5 min read

The AI Claim in Your Marketing Needs a File Behind It

Two years into Operation AI Comply, the FTC applies the same substantiation standard to business-to-business AI claims as it does to consumer advertising.

Two words showed up on a lot of small business websites over the past eighteen months. AI-powered.

They went onto service pages, capability decks, proposal boilerplate and pricing sheets, usually because a vendor put them there first and the language traveled downstream. For most of that period nothing happened, which is the normal way a marketing habit becomes standard.

On August 18, Holland & Knight published an assessment marking two years of the FTC's Operation AI Comply, the enforcement initiative the Commission launched in September 2024. The short version is that the initiative did not fade with a change of administration, and the analysis notes the FTC brought more than a dozen AI washing cases in the past year.

Two elements of how the agency is applying it are worth a small business owner's afternoon.

The business-to-business part

A widespread assumption is that advertising law is a consumer protection matter. You sell to other businesses, your buyers are sophisticated, they do their own diligence, and the rules that govern late-night infomercials are not really pointed at your capability deck.

That assumption does not hold. The Commission is applying the same substantiation requirements and the same deception standards to business-to-business marketing that it applies to consumer marketing. The audience does not change the standard.

This lands hardest on firms that sell services rather than software. A claim that your process is AI-driven, that AI reviews every file, that AI cuts turnaround time by some percentage, is an objective performance claim about your service. It sits in the same category as a claim about response time or error rate, and it needs the same kind of backing.

The FTC's own framing, as the analysis describes it, is that this is pro-innovation rather than anti-technology. Chairman Ferguson's stated concern is that when AI marketing claims cannot be trusted, the whole marketplace absorbs the cost. That framing matters for predicting where enforcement goes next. An agency policing a market it wants to function is not going to stop once the obvious frauds are cleared out.

The part that reaches vendors, and through them, you

The second element is a doctrine with an unglamorous name and real consequences. Means and instrumentalities.

The principle is that the Commission can reach not only the company making a deceptive claim to a customer, but the company that supplied the deceptive marketing materials used downstream. A vendor that hands its resellers, partners and channel a set of claims becomes exposed to the claims those partners then make.

For a small business this cuts in two directions at once, and they are worth separating.

If you resell, implement or white-label somebody else's AI product, the vendor's exposure does not become your protection. You are the party who said it to your customer. The doctrine gives the FTC a path to the vendor as well. It does not give you an exit.

If you are the one producing materials others repeat, whether you are a small software company, an agency or a firm with referral partners, the claims in your partner deck now carry further than your own sales conversations.

The practical upshot is the same in both cases. The numbers in a vendor's slide deck are a claim you are adopting the moment you repeat them, and most vendor decks we review do not come with a source.

What holds up and what does not

The distinction the analysis draws is between aspirational language and present capability. That is the cleanest test we have seen for this, and it converts an abstract compliance question into an editing question.

Present tense claims need present tense evidence. If your site says the system does something today, someone should be able to sit down and watch it do that today.

Roadmap language needs to read as roadmap language. There is nothing wrong with describing where a product is going, as long as a reader cannot mistake it for where the product is.

The other requirement is timing, and it is the one most often missed. The standard is contemporaneous substantiation. The evidence needs to exist at the time the claim is made, not assembled afterward when someone asks. A file put together in response to an inquiry is a different thing from a file that was already there, and the difference is visible.

What we would actually do

Inventory the claims, not the tools. Pull every place your business asserts something about AI. Website, proposal templates, one-pagers, email signatures, the deck your salesperson has been editing privately for a year. The tool inventory is a separate exercise. This one is about sentences.

Sort each into present, aspirational, or unsupported. Most of what we find in this exercise falls into a fourth bucket that nobody expects: claims nobody in the company remembers writing, inherited from a vendor or a contractor and never reviewed. Those get decided rather than defended.

Open a file for the ones you keep. It does not need to be elaborate. What the claim is, what the tool actually does, who verified it, and the date. If a claim describes a measurable improvement, the measurement goes in the file. If you cannot produce a measurement, revise the claim to what you can support.

Ask your vendors for their substantiation before you repeat their numbers. This is a reasonable request and the answer is informative either way. A vendor with documented claims will send them. A vendor who cannot, or who treats the question as unusual, has told you what the number is worth. That answer also belongs in the file.

Set a review date. Claims drift. A statement that was accurate when a tool was configured one way survives the reconfiguration and quietly stops being true. Six months is a reasonable interval for most small businesses.

None of this requires you to stop saying you use AI, and we would not advise that. Being able to describe what these tools do in your business is increasingly the thing buyers are asking about. The work is making sure the description matches the practice, and being able to show that it did when you said it.

If you are working through what your firm can honestly claim about the tools it uses, that is a conversation we are glad to have.

Join the conversation

Straight talk welcome. Comments are moderated, no pitches, no spam.

Comments load once the Hyvor Talk Website ID is set in article.html.