The FTC Wants to Make 'Our AI Is Accurate and Objective' a Legal Claim. Read This Before You Write Your Next Landing Page.

On July 3, the Federal Trade Commission opened public comment (through July 31) on a proposed policy statement about AI accuracy. The one-line version making the rounds is that it's about "woke AI," and the culture-war framing is doing most of the talking. Strip that off and there's a plainer thing underneath that actually touches anyone shipping an AI product: the FTC is signaling that if you tell customers your system produces "the most accurate output possible" and then quietly steer it toward some undisclosed objective, that could be a deceptive act under Section 5 of the FTC Act. The Commission authorized the notice on a 2-0 vote.

I'm not a lawyer, and this is a proposed statement, not a rule: nobody's getting sued next Tuesday. But the direction is worth reading now, because the cheapest time to fix your marketing copy is before it's a compliance surface.

What the statement actually says

The core logic runs like this. AI companies have made representations (explicit ones on their marketing pages, implicit ones baked into the whole pitch) that their systems are designed to produce the best, most accurate, most faithful output they can within their constraints. Consumers, the statement argues, rely on that: it cites that people accept AI outputs without independent fact-checking more than 90% of the time. So there's a reasonable expectation that the system isn't secretly engineered to pursue goals that distort what it tells you. An AI product that does distort its outputs toward undisclosed objectives, while marketed as accurate and objective, could be deceiving consumers.

Then there's the safe harbor, and this is the part builders should actually care about. You can avoid Section 5 exposure by making a clear, conspicuous, and adequate disclosure that your system is designed to prioritize certain objectives over what the user asked for or would otherwise expect. In other words: you're allowed to have a system with a point of view or a set of guardrails. You're just not allowed to sell it as neutral while hiding that it isn't.

There's also a genuinely messy wrinkle. The statement singles out Colorado's AI Act as an example of a state law that could pressure companies to suppress accuracy to avoid disparate-impact liability, and takes the position that such a law is impliedly preempted where it conflicts with Section 5. So the "which rules do I even follow" question, which was already annoying for anyone operating across states, just got a fresh federal-versus-state fault line running through it.

Why this lands on solo operators specifically

The instinct is to assume this is aimed at OpenAI, Google, and Anthropic. It's written with them in mind. But Section 5 is about deceptive claims to consumers, and it doesn't care whether the company making the claim has ten thousand employees or one. If anything, the big labs have general counsel offices that already sweat this language. The solo operator shipping an "unbiased AI research assistant" or an "always-accurate contract analyzer" is the one writing bold marketing claims with nobody reviewing them.

And most AI products a solo operator ships are, under the hood, a model with a system prompt that absolutely does prioritize certain things. You've told it to be concise, or to favor your product, or to refuse certain topics, or to push users toward the paid tier. That's a point of view. It's usually a reasonable one. The exposure isn't that you have guardrails: it's the gap between "objective, accurate, unbiased" on the landing page and the steering you built into the prompt and never mentioned.

What I'd actually do

Go read your own marketing copy today, before you touch anything else. Search your site for "unbiased," "objective," "always accurate," "neutral," "tells you the truth." Every one of those is now a claim you might have to stand behind, and most of them you put there because they sounded good, not because you measured them.

Then close the gap in the cheap direction. You don't have to strip out your guardrails: you have to disclose them. A short, honest line does most of the work: what your system prioritizes, where it will decline, that it can be wrong and users should verify anything that matters. That's not lawyer-proofing; it's just describing your product accurately, which is what the safe harbor rewards. If your app makes decisions that carry real stakes for the user (money, legal, medical, hiring), take it more seriously than a blog-post generator would, because that's where "consumers relied on it" bites hardest.

And keep the receipts. If you're going to claim accuracy, have the eval that supports the claim, even a scrappy one. "We market it as accurate and here's the test suite that checks it" is a very different position from "we called it accurate because it felt accurate."

Here's the honest counter-take, because this could cut the other way. This is a proposed statement from one FTC under one administration; the emphasis on "undisclosed ideological objectives" and the preemption swipe at Colorado are politically loaded, and the whole thing could soften, get reworked, or get tied up before it means anything concrete. If you strip your product down to nothing or lawyer every sentence into mush based on a comment-period draft, you've overreacted. So don't. But the underlying move (say what your product actually does, disclose what it prioritizes, don't oversell "objective") is the kind of thing that's correct whether or not this specific statement survives. Do that part regardless. It's just good product hygiene that happens to also be the safe harbor.

Author

Lukas

@lukcombinator

Sources

• FTC Seeks Public Comment on Policy Statement Addressing AI Accuracy — Federal Trade Commission
• Federal Trade Commission's Proposed Policy Statement Concerning the Suppression of Accuracy in Artificial Intelligence Systems — FTC
• FTC Proposes Policy Statement on AI Accuracy and Ideological Manipulation of AI Outputs — Consumer Financial Services Law Monitor